doi
stringlengths 28
28
| title
stringlengths 19
311
| abstract
stringlengths 217
5.08k
| plain language summary
stringlengths 115
4.83k
| article
stringlengths 3.87k
161k
|
|---|---|---|---|---|
10.1371/journal.pgen.1000535 | Telomere Recombination Accelerates Cellular Aging in Saccharomyces cerevisiae | Telomeres are nucleoprotein structures located at the linear ends of eukaryotic chromosomes. Telomere integrity is required for cell proliferation and survival. Although the vast majority of eukaryotic species use telomerase as a primary means for telomere maintenance, a few species can use recombination or retrotransposon-mediated maintenance pathways. Since Saccharomyces cerevisiae can use both telomerase and recombination to replicate telomeres, budding yeast provides a useful system with which to examine the evolutionary advantages of telomerase and recombination in preserving an organism or cell under natural selection. In this study, we examined the life span in telomerase-null, post-senescent type II survivors that have employed homologous recombination to replicate their telomeres. Type II recombination survivors stably maintained chromosomal integrity but exhibited a significantly reduced replicative life span. Normal patterns of cell morphology at the end of a replicative life span and aging-dependent sterility were observed in telomerase-null type II survivors, suggesting the type II survivors aged prematurely in a manner that is phenotypically consistent with that of wild-type senescent cells. The shortened life span of type II survivors was extended by calorie restriction or TOR1 deletion, but not by Fob1p inactivation or Sir2p over-expression. Intriguingly, rDNA recombination was decreased in type II survivors, indicating that the premature aging of type II survivors was not caused by an increase in extra-chromosomal rDNA circle accumulation. Reintroduction of telomerase activity immediately restored the replicative life span of type II survivors despite their heterogeneous telomeres. These results suggest that telomere recombination accelerates cellular aging in telomerase-null type II survivors and that telomerase is likely a superior telomere maintenance pathway in sustaining yeast replicative life span.
| Telomeres are the specialized structures at the ends of eukaryotic linear chromosomes. The simple guanine-rich DNA repeats at telomeres and their associated proteins are important for chromosome stability. Most eukaryotic species have evolved an enzyme named telomerase to replicate their telomeric DNA. Telomerase usually contains a protein catalytic subunit and a RNA template subunit. A few eukaryotic species can use either telomere recombination or retrotransposon-mediated transposition to accomplish telomere elongation. Interestingly, the baker's yeast Saccharomyces cerevisiae can use both telomerase and recombination to replicate telomeres. In this study, we utilize this unique eukaryotic model system to compare the efficiency of these two mechanisms in the maintenance of cellular function and life span. Telomerase-null cells that used recombination to elongate telomeres were able to maintain relatively stable chromosomes; however, they exhibited a shortened replicative life span which may represent a novel aging pathway. Reintroduction of telomerase inhibited telomere recombination and restored the replicative life span of these cells, implying that telomerase is superior to telomere recombination in the regulation of yeast replicative life span.
| Telomeres are the physical ends of linear eukaryotic chromosomes and are composed of specific repetitive DNA sequences and binding proteins [1],[2]. The functional integrity of telomeres is required for cell proliferation and survival because they protect chromosome ends from nucleolytic degradation and help to distinguish normal chromosome ends from DNA double-strand breaks [3]–[5]. Additionally, telomeres can compensate for the incomplete replication of chromosomal DNA by conventional DNA polymerases [6],[7].
In eukaryotes, telomeres can be maintained by three different mechanisms, namely a telomerase-dependent pathway, a recombination pathway and a retrotransposon-mediated transposition pathway [8],[9]. Telomerase-dependent telomere replication has been documented in the vast majority of eukaryotic species including in budding yeast and humans. In these species, repetitive telomeric DNA is added to the chromosome ends by telomerase, a specialized reverse transcriptase that catalyzes the addition of telomeric DNA using its intrinsic RNA template [10],[11]. Recombination-dependent telomere maintenance has been reported in a few organisms that naturally lack telomerase, including the lower dipterans Chironomus and Anopheles and the plant Allium spp [12]–[14]. Retrotransposon-mediated telomere maintenance has been well adapted by the fruit fly Drosophila melanogaster [15]–[18]. The wider use of telomerase in eukaryotes suggests that it has been evolutionarily selected for as an advantageous mechanism for maintaining telomere integrity and stability, however, the reasons why telomerase has been adopted by so many eukaryotic species are not clear.
Interestingly, some organisms are likely capable of using both telomerase and recombination to replicate their telomeres. For example, previous studies have reported that 85% of the human cancer cells are telomerase positive, however the other 15% cancer cells are telomerase negative [19] and maintain their telomeres by recombination pathway, also termed alternative lengthening of telomeres (ALT) [20]. In the budding yeast Saccharomyces cerevisiae, a RAD52-dependent homologous recombination pathway can be employed by a minority of telomerase-negative cells as an alternative method for telomere maintenance. These cells are called post-senescent survivors. Two types of post-senescent survivors exist and are distinguishable by their characteristic telomere patterns. Type I survivors exhibit amplification of Y' elements and have very short TG1–3 repetitive tracts on the chromosome ends [21]. Type II survivors show a variable pattern of long tracts of TG1–3 repeats and only modest Y' amplification [22]. Because type II survivors have long and heterogeneous telomeric repeats, and their telomere maintenance requires both RAD50 and RAD52, they resemble human ALT cells [20], [23]–[25]. In budding yeast, telomerase seems to be the preferential telomere-elongation pathway. Introduction of the telomerase component EST1 back into an est1Δ type I survivor that exhibits extensive Y' amplification results in the elongation of the terminal telomeric tract back to wild-type length, as well as a substantial reduction in Y' copy number [21],[22]. Similarly, following reintroduction of telomerase into a type II survivor, telomeres gradually return to a wild-type telomeric structure that can be confirmed by examining telomeric restriction patterns via Southern blotting [22].
To investigate whether recombination is inferior to telomerase in preserving an organism or cell under natural selection, we compared cellular traits in telomerase-null post-senescent type II survivor cells to cellular traits in telomerase positive cells. Type I recombination survivors were not included in this report because they have a severe growth defect and highly abnormal karyotypes [21],[26]. In this report we demonstrate that recombination was as efficient as telomerase in maintaining cell survival and overall genome stability, but telomerase-null cells using recombination-only maintenance of telomeres had a shortened replicative life span (RLS) when compared to telomerase-positive cells. In yeast, RLS is defined as the total number of daughter cells generated by a mother cell before cell death [27]. RLS was significantly reduced in type II survivors. The decline in RLS was not due to a defect in the canonical aging regulation pathways and the reintroduction of telomerase activity immediately restored the RLS of type II survivors to that of a wild type cell. Our results provide experimental evidence supporting the notion that telomerase is superior to telomere recombination in the regulation of yeast replicative life span.
To determine whether the recombination pathway is as efficient as the telomerase pathway in maintaining cell survival, we mated either type II survivors (ALT II pathway in humans) or telomerase-proficient cells (TERT pathway in humans) with yeast cells whose telomerase was recently inactivated (pre-survivors, SEN) (Figure 1A). The viability of the resulting diploids was examined. Diploids generated by crossing two SEN populations senesced and underwent crisis (Figure 1A, sectors 3 and 4). In contrast, diploids generated by crossing a type II survivor with a SEN (Figure 1A, sectors 5 and 6) grew as vigorously as the diploids created by mating a TERT with a SEN (Figure 1A, sectors 1 and 2). These data are consistent with previous reports [28] and suggest that telomere maintenance by recombination is as efficient as telomere maintenance by telomerase in maintaining cell survival.
Genome integrity is maintained in telomerase-proficient cells because telomeres are not recognized as DNA double-strand breaks [3]. To ascertain if genome stability was altered in telomerase-negative type II survivors, we carried out several phenotypic analyses and compared our results to those obtained from telomerase-positive (wild-type) cells. First, we determined that type II survivor cells grew as robustly as the wild-type cells (Figure 1B) [22]. Second, we found that type II survivors exhibited similar levels of sensitivity to four separate DNA damage-inducing agents when compared to wild-type cells (Figure 1B). Third, telomere position effect (TPE), a silencing mechanism combining telomere architecture and classical heterochromatin features, was slightly enhanced in type II survivor cells, indicating that the heterochromatic state of telomeres has not been damaged (Figure 1C). This observation was consistent with previous reports where increasing the length of telomeres was found to enhance TPE [29]–[31]. Fourth, normal telomere positioning at the nuclear periphery was maintained in type II survivors, as shown by immunostaining of the telomeric repeat binding protein Rap1p (Figure 1D). Fifth, pulsed-field gel electrophoresis revealed that haploid type II survivors contained linear chromosomes that were indistinguishable from wild-type cells (Figure 1E). Finally, the gross-chromosomal rearrangement (GCR) analyses showed that the type II survivors had a very low GCR rate, in contrast with the rad50Δ mutants that displayed a significant increase of GCR events (Figure 1F). This behavior of the type II survivors was similar to that seen in wild-type cells as previously reported (Figure 1F) [32]. These results suggested that the elevated recombination at type II survivor telomeres has not caused any noticeable defects in DNA replication or repair, and the chromosomes of the type II survivors are stably maintained.
A link between an increase of repetitive rDNA recombination and cellular aging has been well established in S.cerevisiae [33]. With the activation of homologous recombination at telomeric repeats, the type II survivor cells showed more heterogeneous lengths of telomeric TG1–3 tracts and modest Y' telomere amplification (Figure 2A) [22]. Conversely, DNA instability was not observed in another part of the genome when gross chromosomal rearrangements were examined in type II survivors (Figure 1F). This discontinuity of results led us to wonder if the replicative capacity of the type II survivors was comparable to that of telomerase-positive cells. A population of budding yeast cells, can be grown indefinitely in culture under optimal conditions using either telomerase or recombination for telomere maintenance (Figure 1A) [21],[22]. However, the replicative capacity of a single yeast cell is finite. This is because of the activity of other functional aging pathways, and this finite life span holds true whether the telomeres are maintained by telomerase or by recombination throughout the life span [33],[34]. Since yeast cells reproduce by asymmetric cell division with a larger mother cell giving rise to a smaller daughter cell, the two cells can be separated right after each cell division by micromanipulation. The number of daughter cells that a mother cell can produce before senescence defines the replicative life span of that cell. The average number of cell divisions undergone by a group of mother cells defines the mean replicative life span of a yeast strain [27],[35]. A young cell will become mother cell after production of its first daughter.
Replicative life span (RLS) analysis showed that the mean life span of the type II survivors was 15.3 generations, which was much shorter than 27.8 generations of wild-type cells (Figure 2A). Apparently, the reduction in the replicative capacity of type II survivors was neither mating-type nor strain-specific (Figure S1A, S1B). Consistently, the type II survivor cells derived from deletion of the telomerase subunit EST3 also showed significantly decreased life span (Figure S1C). Previous studies showed that, type II survivors are stable over time, but their telomeres experience a cycle of continuous shortening and abrupt elongation during the outgrowth (Figure 2B) [22]. To determine whether the accelerated cellular senescence persists during the outgrowth, we analyzed the life span of the 1st-, 10th-, and 25th-restreaked type II survivors. All these survivors exhibited significantly reduced life span, despite the changes in telomere length (Figure 2B). Thus, for the first time, a severely reduced life span was observed in the telomerase-null type II survivors.
As previously reported, critically short telomere(s) cause telomerase-deficient cells to abruptly cease cell division, and senesce at the G2/M checkpoint (Figure S2) [36]. To determine whether the accelerated senescence of type II survivors was caused by critically short telomere(s), the morphology of cells at the end of their life span was indexed according to a previously established method where the fraction of unbudded, small-budded, and large-budded cells is determined [37]. In our experiments, the large-budded proportion of old type II cells was comparable to that of old wild-type cells (Figure 2C), suggesting that type II survivors may age by a process that is independent of critically-short-telomeres.
Long telomeres have also been proposed to affect replicative life span. For example, rif1Δ cells have much longer telomeres than wild-type cells and their replicative life span is reduced [38]. These data with rif1Δ cells led to the generation of a hypothesis that long telomeres may reduce life span by competing for Sir silencing factors with the non-telomeric loci [38],[39]. Because the telomerase-null type II survivors possess long and heterogeneous telomeres, it is possible that long telomeres per se in these cells result in the life span decline. To test this possibility, yeast cells that harbored long telomeres by temporarily over-expressing telomerase were subject to a life span assay. Results show that these cells had regular life span (Figure 2E), indicating that long telomeres per se may not be the direct cause of the shorter life span we saw in the type II survivors.
To confirm that the telomerase-null type II survivors died of replicative aging but not general sickness, we examined age-dependent sterility in type II survivors. Sterility due to loss of silencing at HMLα and HMRa loci has been reported as an aging-specific phenotype in budding yeast [40],[41]. We determined the percentage of sterile cells as cells aged by documenting the inability of cells to respond to a mating pheromone, α factor. Similar to the wild-type cells, type II survivors became sterile at a higher frequency the older they got (Figure 2D). Together, the data from all the phenotypic assays in the section led us to conclude that telomerase-null type II survivors aged prematurely in a manner that was phenotypically indistinguishable from that of telomerase-positive wild-type cells.
Calorie restriction (CR) is an intervention which slows the aging process and increases life span in many organisms [42]. In yeast, CR can be executed by reducing the glucose concentration of growth media from 2% to 0.5% (or 0.05%), resulting in a significant increase of life span (Figure 3A) [43],[44]. Life span extension by CR in yeast involves at least three nutrient-responsive kinases: TOR (target of rapamycin), Sch9, and protein kinase A (PKA) [43], [45]–[47]. To better understand life span regulation in type II survivor cells, we examined whether these canonical aging pathways function in type II survivors. Although type II survivors exhibited accelerated replicative aging, CR (in 0.05% glucose) significantly extended the survivors' life span from 12.6 to 16.7 generations (Figure 3A). The deletion of TOR1 also significantly extended the mean and maximum life span of type II survivors (Figure 3B). Subsequently, we found that the inactivation of Tor1 affected neither the senescence nor survivor-arising rate of est2Δ cells (Figure S3). In addition, tor1Δ type II survivors exhibited shorter life span than that of tor1Δ telomerase-positive cells (Figure 3B). Based on these data, we propose that TOR1 regulates the replicative life span of type II survivors independently of telomere recombination. The extension of life span by either calorie restriction or Tor1 deletion further indicated that type II survivors are not simply sick cells.
In every way we examined them phenotypically, type II survivors seemed to resemble wild-type cells with the exception that their telomeres were elongated through telomere-telomere recombination and not telomerase (Figure 2A). Because telomerase has been suggested to play a role in capping of telomeres and facilitating cell proliferation [48], there is a possibility that the lack of telomerase capping provides an essential senescence signal. Alternatively, the increased telomere recombination per se in telomerase-null survivors induced chromosomal instability at telomere and this may be the cause of the decreased life span. To distinguish between these two possibilities, we examined the RLS of telomerase-deficient pre-survivors. Heterozygous yeast diploid cells with a single EST2 deletion were dissected and subjected to serial restreaking on YPD plates every 48 hours. Both the spores taken immediately from the tetrad dissection and the cells that were grown on YPD plates for 48 hours had wild-type life span (Figure 4A and 4B). Decreased RLS was only observed in telomerase-null pre-survivors during the later serial restreaks when there was severe loss of telomeric DNA (Figure 4C). Reintroduction of telomerase restored the low RLS of pre-survivors by elongation of short telomeres, suggesting that the short RLS of late passages was caused by critically short telomere(s) (Figure 4C). Together, these data supported a model where the absence of Est2p did not directly cause a reduction in replicative capacity and the shorter life span of type II survivors was not caused by the lack of telomerase capping.
The observation of loss-of-productivity in late passages of telomerase-null pre-survivors mirrors what is known to happen to est2Δ cells on serial streak-outs (Figure 1A), or in liquid-growing culture (Figure S2A) [36],[49]. Mother cells exhibited the same rate of cell viability reduction as the logarithmically growing cells, the vast majority of which are young cells, suggesting that the older mother cells and younger cells do not have appreciable difference in their ability to maintain telomeres in the absence of telomerase.
In budding yeast, telomerase appears to be the preferred pathway for telomere maintenance [21],[22]. To examine whether reintroducing telomerase activity in survivors may inhibit telomere recombination and rescue the short RLS, we performed a mating assay in which est2Δ type II survivor cells (MATα) were mated with wild-type cells (MATa). The telomerase-positive diploids (named “survivor diploids”) had a heterogeneous telomere-length (Figure 5A). However, they possessed similar replicative capacity to the wild-type diploids (Figure 5A), indicating that telomerase is dominant over recombination in regulating cellular life span, and further suggesting that heterogeneous long-telomeres are not the cause of premature aging. The diploid cells in this background (BY4743) lived significantly longer than haploid cells, and this phenomenon was also observed by Kaeberlein et al. [50]. Next, both the telomerase-negative and positive haploid cells were obtained by tetrad dissection from the “survivor diploids”. Interestingly, all the spores had long heterogeneous telomeres but possessed the same replicative capacity as the wild-type haploids (Figure 5B). These results are supportive of the following conclusions: (i) reintroduction of telomerase activity is able to extend the short RLS of type II survivors; (ii) the reduced replicative capacity of telomerase-null type II survivors resulted from telomere alteration; (iii) lack of telomerase capping does not cause a decline of replicative capacity, consistent with the results shown in Figure 4; (iv) long telomeres per se do not affect cellular life span, consistent with our previous results shown in Figure 2E; (v) telomerase may restore RLS by inhibiting telomere recombination rather than regulating telomere length.
To further verify the idea that reintroducing telomerase activity into survivors may inhibit telomere recombination and rescue the cellular life span, we reintroduced EST2 into the type II survivors which were originally derived from cells with an EST2 deletion. Reintroduction of telomerase activity caused a gradual shortening of very long telomeres in type II survivors, and eventually recovered the wild-type telomere pattern (after ∼25 restreaks, see middle panels of Figure 6A and 6B). These data suggested that when compared to recombination, telomerase was the preferred mechanism for telomere maintenance and the presence of telomerase suppressed telomere-telomere recombination [22]. Consistent with the results of our mating assay (Figure 5), the RLS of type II survivors was restored immediately after the reintroduction of EST2 in spite of the long telomeres (Figure 6C). Accordingly, a plasmid-borne EST3 restored the life span of est3Δ type II survivors immediately after being transformed (Figure 6D). Reintroduction of a catalytically inactive est2 (DD670-1AA) into est2Δ type II survivors, on the other hand, failed to restore either the telomere length or the typical life span (Figure 6). These observations suggested that functional telomerase is required for the life span restoration in type II survivors.
In budding yeast, a change of rDNA recombination rate causes reciprocal changes in the cellular life span. For example, elimination of the replication block protein FOB1 or over-expression of SIR2 significantly extended cellular life spans by reducing rDNA recombination, whereas the deletion of SIR2 had the opposite effect [51]–[53]. In the telomerase-null type II survivors, recombination was presumably activated at the telomere loci. We wondered whether increased telomere recombination affected the stability of rDNA loci. A marker loss assay was performed as described previously [54] to analyze the rDNA recombination rate in type II survivors. Interestingly, the rDNA recombination rate in type II survivors was 3-fold lower than that observed in wild-type cells (Figure 7A), thereby suggesting that extra-chromosomal rDNA circles (ERCs) do not contribute to the acceleration of aging process in type II survivors.
Like in wild-type cells, deletion of SIR2 in type II survivors led to a decline of life span from 14.6 to 8.6 generations (Figure 7B). Unexpectedly, neither the deletion of FOB1 nor introduction of an extra-copy of SIR2 into type II survivors could extend their life span (Figure 7C and 7D). Double deletion of FOB1 and SIR2 did not affect the life span of wild-type cells [53], whereas combined deletion of FOB1 and SIR2 reduced the life span of type II survivors (Figure 7E). We were unable to determine why FOB1 deletion or Sir2 over-expression did not positively influence the RLS of type II survivors. It is possible that the increased telomere recombination sequestered recombination factors and caused a reduction in rDNA recombination (Figure 7A). Thus, either FOB1 deletion or Sir2 over-expression could no longer reduce rDNA recombination levels that were already lower.
In this study, we reported that telomerase-null type II survivors, which employ homologous recombination to efficiently maintain telomeres, exhibited normal chromosomal stability in an assay that measures gross chromosomal rearrangement rates, but accelerated cellular senescence. The reduced replicative life span of type II cells could be extended by either calorie restriction or inactivation of the TOR pathway, but not by FOB1 deletion or SIR2 over-expression. Reintroduction of telomerase restored the life span of type II survivors to wild-type level, indicating the superiority of telomerase over homologous recombination in guaranteeing full replicative potential.
In most eukaryotic species studied so far, telomere replication involves either telomerase or a recombination pathway [9]. Stable maintenance of telomeres is required for cell proliferation, survival and preservation of a species [55]. Reactivation of telomerase or telomere-recombination is associated with immortalization of mammalian cells grown in tissue culture, including human cells [19],[56]. Similarly, the budding yeast S. cerevisiae, can be grown indefinitely in culture under optimal conditions with either telomerase or telomere-recombination activated for telomere maintenance (Figure 1A) [21],[22]. However, for a single yeast cell, its replicative capacity is finite due to the activity of other aging pathways regardless of how telomeres are maintained throughout the life span [33],[34]. Surprisingly, we found that the budding yeast type II survivor cells, which have adopted homologous recombination to replicate their telomeres, possessed shorter replicative life span (Figure 2 and Figure S1). Type II survivor cells differ from wild-type cells by the nature of their repetitive telomeric DNA sequences, the physiological challenges they may face, the length of their heterogeneous telomeres, the absence of telomerase capping, the heterochromatin structure at these telomeres, and the telomere recombination status. Since each of these differences alone or a combination of these differences may be responsible for the shortened RLS in type II survivors, we examined further the contributions of these differences in the premature senescence phenotype.
Compared to wild-type cells, the type II survivors did not exhibit altered sensitivity to various DNA damage-inducing reagents (Figure 1B). Additionally, they did not show an increase in the rate of gross chromosomal rearrangement events (Figure 1F). Consistent with the genetic assay, type II survivor cells and wild-type cells displayed identical chromosomal banding patterns when compared using pulsed-filed gel electrophoresis (Figure 1E). Moreover, telomere silencing was slightly enhanced in type II survivors (Figure 1C), and the telomere clustering at the nuclear periphery remained similar to the wild-type cells (Figure 1D). These results indicate that the type II survivors are phenotypically healthy, instead of generally “sick” cells.
When examining cell morphology at the end of the life span, type II survivor cells showed similar fractions of cells that were large-budded and small-budded when compared to wild-type cells at the same stage of life span (Figure 2C). In addition, the type II survivors exhibited the aging-associated sterility in a manner that was nearly identical to that of wild type cells (Figure 2D). These observations on one hand raise the argument that the type II cells are premature aging instead of premature death, and on the other hand challenge the idea that the life span reduction of type II survivors is due to their critically short telomere(s) which could cause more cells to senesce at G2/M phase (Figure S2). Additionally, the life span of type II survivors was extended by calorie restriction or inactivation of the TOR1 pathway (Figure 3), and reduced by SIR2 deletion (Figure 7B), further supporting the argument that the type II survivors age prematurely.
Since telomerase has been shown to play a capping function in maintaining telomere integrity [48], it remains unclear how telomere capping is maintained in the type II survivors. In the telomerase deficient pre-survivors with a moderate loss of telomeric DNA, the defect of telomerase capping due to lack of Est2p did not detectably affect the replicative capacity (Figure 4A and 4B). Consistent with those data, the telomerase-null cells newly derived from EST2/est2 hybrids (crosses between type II survivors and wild-type haploids) have a comparable life span to the telomerase proficient cells regardless of the length of their heterogeneous telomeres (Figure 5B). Thus, the life span reduction in telomerase-null type II survivors did not appear to be a consequence of a loss of telomerase capping by Est2p.
Recent studies on the role of telomere length in aging have expanded from the cellular level to the anatomical/organismal level. Telomerase-deficient mice with critically short telomeres exhibit decreased viability associated with diminished proliferative capacity of B and T cells [57]–[59]. However, when telomere length is kept above the critically short length, the relationship between telomere length and life span seems to be controversial. In nematode Caenorhabditis elegans, Joeng et al. showed that worms with longer telomeres live longer [60]; whereas Raices et al. demonstrated that telomere length contributed little to the normal aging process [61]. In Drosophila melanogaster, longer telomeres were found to have no effect on the life span of the adult flies [62]. In the yeast rif1Δ cells, long telomeres were proposed to contribute to accelerated cellular senescence by titrating away limiting pools of Sir silencing factors from non-telomeric silenced loci [38],[39]. Several lines of evidence presented in our current study do not support the hypothesis that longer telomeres alone contribute to a shortened life span in yeast. For example, the yeast cells that harbored long telomeres by temporarily over-expressing telomerase exhibited wild-type life span (Figure 2E). Additionally, the hybrid diploid cells obtained from mating wild-type and type II haploids had full replicative capacity in spite of heterogeneous long telomeres (Figure 5A). Moreover, reintroduction of telomerase slowly restored the telomere-length homogeneity, but immediately restored the life span (Figure 6). Finally, over-expression of the essential silencing factor Sir2p had no effect on the replicative life span of type II survivors (Figure 7D). These results indicate that long-telomere length per se in the type II survivors is not associated with the accelerated cellular aging we observed. Most likely, type II survivors aged prematurely in a telomere-length independent manner. The results presented in our current work are different from the ones reported previously [38], where rif1Δ or tlc1 mutants were exploited to characterize the relationship between telomere length and life span. In our experiments (Figure 2E, Figure 5, and Figure 6), no parameters other than telomere length have been changed, and this might help to explain the discrepancy of our results and those reported previously [38].
In contrast to the controversial role of telomere length in longevity determination, loss of genome integrity is generally believed to contribute to the finite life span of organisms from yeast to humans [63]. A causal link between repetitive DNA instability and aging has been previously established in S. cerevisiae. The rate of aging in mother cells is dictated by the stability of the rDNA, which is present in 100–150 tandem arrays of 9.1-kb repeats [33],[64],[65]. During the aging of mother cells, extra-chromosomal rDNA circles (ERCs) are formed by homologous recombination between rDNA repeats. Importantly, ERCs are self-replicating via an origin in the rDNA repeat-unit during subsequent cell cycles and they display biased segregation to mother cells due to a lack of CEN element [66]. Thus, ERCs accumulate with the aging of mother cell in a Septin- and Bud6-dependent manner, and likely contribute to cellular senescence once a threshold level is reached [67],[68]. In type II survivors, rDNA recombination was decreased compared to the wild-type cells (Figure 7A). So it is unlikely that the ERCs contributed to the acceleration of the aging process in type II survivors, and it is likely that other aging pathway(s) dominated the aging process. Accordingly, SIR2-overexpression or FOB1-deletion did not extend RLS in type II survivors because the ERC pathway was recessive in the aging process of these cells. However, SIR2-deletion should still further shorten RLS because it makes the ERC pathway dominant again.
As telomeres are arranged in TG-rich repeats, we could not rule out the possibility that telomere circles might affect cellular life span in the same way as rDNA circles. However, qualitative and quantitative determination of telomeric DNA-containing rings shows that telomere circles exclusively exist during the time when survivors are being generated, but not after survivors are established [69]. In addition, telomere repeats do not contain self-replicating origin elements. It's unlikely that telomere circles would accumulate during the aging process of survivors. Given that the recombination has been increased at telomeric loci in type II cells in a manner similar to that of rDNA recombination in aging cells [70], telomere recombination may titrate away vital transcription and/or replication factors that play a role in preventing cellular senescence. Accordingly, we did observe significantly reduced rDNA recombination in telomerase-null type II survivors (Figure 7A). However, we could not detect any DNA replication or repair defect on the general chromosome loci as shown by several lines of evidence. The telomerase-null type II survivors did not exhibit altered sensitivity to various DNA damage-inducing reagents when compared to wild-type cells (Figure 1B), thereby indicating there was no obvious DNA replication or repair defect on the general chromosome loci. In addition, the gross-chromosomal rearrangement (GCR) rate was not increased in type II survivors as previously reported (Figure 1F) [32]. Moreover, the chromosomal banding pattern of type II survivors was comparable to the wild-type cells as displayed by pulsed-field gel electrophoresis (Figure 1E). At this point, we could not explain why telomeres preferentially competed with the rDNA loci for recombination. One possibility is that our assays (Figure 1) are not sensitive enough to detect any replication or repair defect on the general chromosome loci. Alternatively, both telomeric and ribosomal DNAs are favorable substrates for certain factor(s), such as Sir2p binding, and an increase of recombination in either one would affect the rate of recombination at the other. However, SIR2 over-expression, which might compensate for the decrease of Sir2p at rDNA loci, did not extend the replicative life span of type II survivors (Figure 7D), thereby leading us to propose that at least Sir2p is not the factor that might be involved in regulating the relative recombination rates at both telomeres and the rDNA.
The shorter life span of the pre-survivors, which was potentially caused by severe telomere loss, could be rescued by reintroduction of telomerase, presumably due to recovery of telomere length by telomerase (Figure 4C). Interestingly, reintroduction of telomerase immediately restored the short life span of telomerase-null type II survivors despite insignificant changes in telomere length (Figure 5 and Figure 6), implying that a distinct mechanism is engaged in the life span regulation upon reactivation of telomerase. Reintroduction of telomerase activity caused a gradual shortening of very long telomeres in type II survivors and eventually re-established the wild-type telomere Southern blotting banding pattern (Figure 6A and 6B, middle panels) as previously reported [22]. This observation suggested that the presence of telomerase somehow suppressed telomere-telomere recombination. Catalytically inactive telomerase failed to inhibit telomere recombination as reflected by the continuous presence of heterogeneous telomere pattern (Figure 6A and 6B, right panels), thus, it could not recovery the replicative life span of type II survivors (Figure 6C). We therefore propose a model where telomere recombination leads to accelerated cellular aging in telomerase-null survivors and functional telomerase rescues the replicative life span of type II survivors by inhibiting telomere recombination. In conclusion, telomerase has evolved to be as a superior mechanism to telomere recombination in regulating cellular life span. Telomerase likely plays a duel role in regulating life span. It helps maintain the telomeres above the critically short length necessary to reach full replicative potential, while also inhibiting the telomeric recombination that otherwise leads to a decline of cellular replicative capacity.
Unless otherwise noted, all yeast strains used in this study were BY4742 (MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0), BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0), BY4743 (MATα/MATa his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 lys2Δ0/+ met15Δ0/+ ura3Δ0/ura3Δ0), and their derivatives. All strains were grown at 30°C and on YPD (10 g/L yeast extract, 20 g/L peptone, 2% dextrose) unless otherwise stated.
tor1Δ, sir2Δ and fob1Δ strains were from EUROSCARF consortium. est2Δ was a deletion of the EST2 open reading frame using a pRS303 plasmid which contained 800 bp homologous sequences to up- and down-stream of EST2 ORF. The same method was used for the deletion of the open reading frame of EST3. sir2Δ fob1Δ mutants were obtained by deletion of the SIR2 ORF in the fob1Δ strain. All gene disruptions were verified by PCR. Strains over-expressing Sir2p were constructed by genomic integration of an extra-copy of SIR2. Integration of SIR2 at LEU2 locus was accomplished by transforming cells with Hpa I digested plasmid pRS305-SIR2. In addition to the entire coding region of SIR2, 800 nucleotides of up-stream and down-stream sequence were included. Plasmid pRS305-SIR2 was constructed by ligation of the PCR-amplified products into the BamH I and Sal I sites of pRS305.
The pRS316-EST3 centromere plasmid was constructed as described [71]. The pRS316-EST2 centromere plasmid was a gift from Dr. Yasumasa Tsukamoto. The pRS316-est2 (DD670-1AA) was constructed using site-directed mutagenesis method.
Replicative life span assay of yeast cells was performed as described previously [72],[73]. Prior to analysis, strains were patched onto fresh solid medium and grown for 2 days at 30°C. Single colonies were then arrayed onto standard YPD plates using a micro-manipulator and allowed to grow for about 3 hours. Virgin daughter cells were isolated as buds from mother cells and subject to life span analysis. During life span experiments, plates were incubated at 30°C during the daytime and stored overnight (∼8 hr) at 4°C. Each experiment consisted of more than 50 mother cells and was independently repeated at least twice. Data shown in the results represent one single experiment. Statistical significance was determined by a Wilcoxon rank sum test using Stata 8 software. Differences are stated to be significant when the confidence is higher than 95%.
Genomic DNA prepared from each strain was digested by XhoI or 4 bp cutter (MspI, HaeIII, HinfI, AluI), separated on a 1.0% gel, transferred to Hybond-N+ membrane (GE Healthcare), cross-linked by UV and then probed with C1–3A/TG1–3 telomere-specific probe as described previously [22].
Ten-fold dilutions of each strain were patched on YPD containing the indicated doses of phleomycin (Sigma), methyl methanesulfonate (MMS; Sigma), hydroxyurea (HU; Sigma), or exposed to UV with indicated doses, or grown at 23°C, 30°C, and 37°C. Photos were taken after two days.
Each strain that contains URA3-marked telomere VII-L was grown to log phase at 30°C. Ten-fold serial dilutions were plated on YC complete medium and YC containing 5-FOA (5-Fluoroorotic Acid) at 1 g/L. Plates were incubated at 30°C for two days and then photos were taken.
Immunostaining of Rap1 was performed as described previously [74].
Agarose plugs for pulsed-field gel electrophoresis (PFGE) were prepared as described previously [75]. PFGE was performed on a Bio-Rad CHEF-DR-III system in 0.5×TBE at 14°C using the following program: step 1, voltage 3.6 V/cm, switch 120 s, time 20 hr; step 2, voltage 3.6 V/cm, switch 300 s, time 24 hr. After electrophoresis, DNA was visualized by ethidium bromide staining.
GCR rate in indicated strains was determined as previously described [76]. The nonessential gene, HXT13, located distal to the CAN1, was replaced with a second selectable marker, the URA3 gene. Each strain was inoculated into YPD medium and grown at 30°C until the culture reached saturation. Cells of suitable dilutions were spread on YC plates in the presence or absence of 60 mg/L L-canavanine and 1 g/L 5-FOA. A fluctuation test and the method of the median were used to assess GCR rate [77].
Sensitivity to α factor was performed as previously described [41]. Cells of various ages were scored for their ability to undergo cell cycle arrest and schmooing in response to the yeast mating pheromone, α factor. After 4 hours of α factor challenge, cells were transferred to fresh medium to complete their life span. All cells documented underwent at least one cell division after being removed from the presence of α factor.
The rate of marker loss in rDNA was measured as described [54]. Strains carrying a URA3 marker integrated into the rDNA array were grown in YC medium lacking uracil until the culture reached saturation. Cells of suitable dilutions were spread on YC plates with and without 5-FOA. Plates were incubated at 30°C for two days and colonies were counted. The number of colonies on 5-FOA plates divided by the number of colonies on YC plates was reported as the rate of marker loss. A Student's t test was used to determine the statistical significance of the data.
|
10.1371/journal.pntd.0002217 | Is Pregnancy Associated with Severe Dengue? A Review of Data from the Rio de Janeiro Surveillance Information System | Dengue is a reportable disease in Brazil; however, pregnancy has been included in the application form of the Brazilian notification information system only after 2006. To estimate the severity of maternal dengue infection, the available data that were compiled from January 2007 to December 2008 by the official surveillance information system of the city of Rio de Janeiro were reviewed.
During the study period, 151,604 cases of suspected dengue infection were reported. Five hundred sixty-one women in their reproductive age (15–49 years) presented with dengue infection; 99 (18.1%) pregnant and 447 (81.9%) non-pregnant women were analyzed. Dengue cases were categorized using the 1997 WHO classification system, and DHF/DSS were considered severe disease. The Mann-Whitney test was used to compare maternal age, according to gestational period, and severity of disease. A chi-square test was utilized to evaluate the differences in the proportion of dengue severity between pregnant and non-pregnant women. Univariate analysis was performed to compare outcome variables (severe dengue and non-severe dengue) and explanatory variables (pregnancy, gestational age and trimester) using the Wald test. A multivariate analysis was performed to assess the independence of statistically significant variables in the univariate analysis. A p-value<0.05 was considered statistically significant.
A higher percentage of severe dengue infection among pregnant women was found, p = 0.0001. Final analysis demonstrated that pregnant women are 3.4 times more prone to developing severe dengue (OR: 3.38; CI: 2.10–5.42). Mortality among pregnant women was superior to non-pregnant women.
Pregnant women have an increased risk of developing severe dengue infection and dying of dengue.
| Dengue represents a major worldwide public health problem. According to the WHO, up to 50 million dengue infections occur each year. The occurrence of dengue fever and dengue hemorrhagic fever has increased in Brazil, in part due to the simultaneous circulation of DENV-1, DENV-2 and DENV-3. Although a primary infection with one serotype confers a partial or transient immunity against other serotypes, any subsequent infections harbor the risk of increased morbidity/mortality. Several case reports have been published regarding maternal and fetal outcomes from dengue infection, but it is still inconclusive if pregnancy is associated with severity. To estimate the severity of maternal dengue infection, available data that were compiled from 2007 to 2008 by the official surveillance information system of the city of Rio de Janeiro were reviewed. The cases of dengue were analyzed using the 1997 WHO classification. Pregnant women were 3.4 times more prone to developing severe dengue than non-pregnant women. Mortality among pregnant women was superior to non-pregnant women. The increased risk of severe outcomes in pregnant women merits further attention to effective public health and medical interventions that will aid in avoiding morbidity/fatalities within this population.
| Since the reintroduction of DENV-1 in 1986 in RJ, dengue has become a major public health problem in Brazil [1]. The occurrence of dengue fever (DF) and dengue hemorrhagic fever (DHF) has increased over the past several years in Brazil, in part due to the rapid spread and simultaneous circulation of the DENV-1, DENV-2, DENV-3 [1]. In 2008, over 600,000 cases of DF and 4,455 cases of DHF were reported in Brazil, with 40% and 42%, respectively, occurring in the state of RJ [2], [3].
A surveillance information system of reportable diseases, SINAN, was implemented in Brazil in the early 1980s [4], and since then, dengue has been a compulsory reportable disease. However, pregnancy was a reportable item on the form only after 2006.
Globally, there are increasing reports of dengue during adulthood, increasing the risk for dengue during pregnancy. In the literature only approximately 400 cases of dengue during pregnancy have been reported, primarily describing the maternal and fetal outcomes [5], [6]. If diseases such as malaria and cholera are more severe during pregnancy, would dengue also be more severe? During the 2007/2008 epidemic in the city of RJ, the highest rate of laboratory-positive dengue samples was among those in the age group under 15 years, followed by those 15–29 years; 99% of all births during this period occurred in mothers aged 15–49 years [7].
To estimate the severity of maternal dengue, the available data provided by SINAN related to the epidemic period of January 1, 2007, through December 31, 2008, in the city of RJ, were reviewed. Laboratory-confirmed dengue cases in reproductive-age women (15–49 years) were included. Mortality and severity of the disease were compared between pregnant and non-pregnant women.
A suspected dengue case is routinely reported to SINAN within 24 hours of attendance in a healthcare unit, using a standardized form [8]. When the laboratory results are available, the form is completed by a health staff member who reviews the chart information and adds the final dengue classification, usually after a period of no more than 3 months. Suspected cases are reported from all healthcare facilities in RJ.
The SINAN form includes information on basic demography, laboratory data, hospitalization and outcomes (death or cure). Dengue cases are classified according to the WHO 1997 [9], adapted by the Brazilian Ministry of Health to include the category of dengue with complications [10] for the cases that do not fulfill all three criteria for DHF. Laboratory confirmed cases were considered when either virus isolation, PCR testing, paired IgM or IgG testing or single IgM test was positive.
Pregnancy is categorized in the SINAN form according to trimester: 1st trimester (up to 14 weeks of gestation), 2nd trimester (14–28 weeks), 3rd trimester (after 28 weeks) or unidentified gestational age.
Patients were categorized according to the WHO 1997 classification system as DF, DHF or DSS [10]. Dengue classification of patients (n = 117) categorized in the SINAN form as ‘dengue with complications’ were reviewed. If patients had evidence of plasma leakage they were categorized as having DHF/DSS and thus considered as severe cases. Otherwise, patients were categorized as having DF.
The Mann-Whitney U test was used to test the difference between the mean age of pregnant and non-pregnant women and the difference between the mean age of pregnant women by dengue classification (DF and DHF/DSS).
A chi-square test was used to evaluate the differences in the proportion of dengue severity between pregnant and non-pregnant women. A p-value of <0.05 was considered significant in all statistical tests.
A univariate analysis was performed using DHF/DSS (dependent variable) and pregnancy, maternal age (as a continuous variable) and trimester (independent variables) using the Wald test. Multiple logistic regression analysis was used to determine whether statistically significant variables were independently associated with dengue severity. Variables with a p-value<0.05 in the univariate analysis were included in the multivariate analysis. Finally, the residuals of the fitted model were analyzed. With this modeling, the odds ratio and their respective confidence intervals (95%) were obtained. All statistical analyses of data were performed using R software, version 2.11.1.
Our study was reviewed and approved by the Ethical Committee of the Municipal Secretary of the City of Rio de Janeiro: Comitê de Ética em Pesquisa da Secretaria Municipal de Saúde e Defesa Civil. Protocolo de pesquisa: 51/08. CAAE: 0122.1.314.000-08 e 0130.1.314.000-08. Inform consent was not obtained because the data were analyzed anonymously.
The incidence of laboratory confirmed dengue among women in reproductive age was 234/100,000 inhabitants/2y, with similar rates between pregnant (238/100,000) and non-pregnant women (233/100,000). Mortality of dengue was 3,6/100,000 inhabitants/2y among pregnant women and 1,7/100,000 inhabitants/2y among non-pregnant women. Case fatality rate was 7,4 and 1,5% respectively.
Data on 546 eligible reproductive-age women who had confirmed cases of dengue were analyzed: 99 (18.1%) were pregnant and 447 (81.9%) were not (table 1). The mean (± standard deviation) maternal age was significantly different: 26.3±8.5 years in pregnant women compared with 31.5±10.7 years in non-pregnant women (p<0.05). No significant difference was observed in the mean age between pregnant women with DHF/DSS (25.5±8.2) and DF (26.9±8.5).
Most cases were classified as DF (n = 417, 76.4%), 123 as DHF (22.5%) and 6 as DSS (1.1%). A higher proportion of pregnant women than non-pregnant women had DHF/DSS (table 1).
Hospitalization information available for 186 (34.1%) patients occurred in 61 (34.1%) pregnant women, and in 118 (65.9%) non- pregnant women. The proportion of severe dengue among hospitalized women was similar: 73.8% and 66.9% for pregnant and non-pregnant women, respectively.
Information on death was available for 395 (72.3%) of the eligible cases: three pregnant and five non-pregnant women died (table 1). Shock syndrome (n = 3) and cavity effusion (n = 2) were associated with deaths. The cause of death was unknown in three patients.
A higher prevalence of DHF/DSS that increased with gestation age was observed (table 2). Pregnant women were 3.4 times more likely to have DHF/DSS, primarily in the last trimester; OR 3.38; CI 2.1–5.42 (table 3).
This study suggests that dengue during pregnancy can increase maternal mortality, as previously reported [11]. It also suggests that pregnancy is associated with DHF/DSS and that the susceptibility to severe disease increases with pregnancy age.
Severe dengue has been associated with maternal deaths, with fatality rates ranging from 2.9%–22% [5]–[6], [11]–[13]. The maternal dengue fatality in this study was 7.4%. The differences in dengue fatality in pregnant women likely result from differences in the designs and in the heterogeneity of the studies sample sizes. Additionally, it may represent different regional management of dengue in pregnant women.
More than half of pregnant women were hospitalized and it was twice the rate of hospitalization for non-pregnant women, since it was a recommendation of Rio de Janeiro's healthcare authorities to prevent dengue complications in this group. Moreover, the proportion of DHF could still be underestimated as the identification of plasma leakage syndrome through the hemoconcentration or hypoproteinemia may be compromised from the seventh to the 32rd week of gestation, by the physiological increase of intravascular volume of this period [14].
The reasons for the association of DHF/DSS with pregnancy were not assessed in this study. The amount of vascular leakage during early versus late pregnancy may have different effects on the clinical presentation and on the perceived severity level. The higher risk for developing severe disease in the 2nd and 3rd trimesters should be confirmed by prospective studies as the selection bias related to admission because of risk of preterm delivery cannot be excluded.
The non-laboratory confirmed dengue cases were not analyzed to avoid a detection bias, and the confusion of dengue with pregnancy complications, such as HELLP syndrome.
The findings of the study are based on a retrospective review of routinely collected data, with laboratory confirmed dengue, which introduces some limitations such as bias resulted from incomplete data and possible misclassification. Although pregnant women were more likely to be hospitalized for fever and illness in general compared to their non-pregnant counterparts, it would be expected a lower frequency of severity among this group as pregnant women had a preventive hospitalization.
As all the uncompleted data about death were attributed to non-pregnant women, the mortality rate among pregnant women might still be underestimated.
SINAN has also been used in Brazil to conduct studies on dengue [15]. Although citywide surveillance system of information has no specific clinical plasma leakage signs data and may be incomplete, it is a population-base registry from which maternal dengue severity could be inferred by the access to dengue classification. Further longitudinal studies are needed to confirm these findings and to determine on how these two subgroups presents clinically and how their presentations differ.
|
10.1371/journal.pntd.0005567 | Prevalence of depression and associated clinical and socio-demographic factors in people living with lymphatic filariasis in Plateau State, Nigeria | Lymphatic filariasis is a chronic, disabling and often disfiguring condition that principally impacts the world’s poorest people. In addition to the well-recognised physical disability associated with lymphedema and hydrocele, affected people often experience rejection, stigma and discrimination. The resulting emotional consequences are known to impact on the quality of life and the functioning of the affected individuals. However, the management of this condition has focused on prevention and treatment through mass drug administration, with scant attention paid to the emotional impact of the condition on affected individuals. This study aimed to determine the prevalence and severity of depression among individuals with physical disfigurement from lymphatic filariasis in Plateau State, Nigeria.
A cross-sectional 2-stage convenience study was conducted at 5 designated treatment centers across Plateau State, Nigeria. All available and consenting clients with clearly visible physical disfigurement were recruited. A semi-structured socio-demographic questionnaire, Rosenberg Self-esteem and a 9-item Patient Health Questionnaire (PHQ-9) were administered at the first stage. Those who screened positive (with a PHQ-9 score of five and above) were further interviewed using the Depression module of the Composite International Diagnostic Interview (CIDI).
Ninety-eight individuals met the criteria and provided consent. Twenty percent of the respondents met criteria for depression, with the following proportions based on severity: Mild (42.1%), Moderate (31.6%) and Severe (26.3%). History of mental illness (OR 40.83, p = 0.008); Median duration of the illness was 17 years (IQR 7.0–30 years) and being unemployed (OR 12.71, p = 0.003) were predictive of depression. High self-esteem was negatively correlated (OR 0.09, p<0.004).
Prevalence of depression is high among individuals with lymphatic filariasis and depression in sufferers is associated with low self-esteem and low levels of life satisfaction.
| Lymphatic filariasis is a chronic illness that is disabling and often results in disfigurement. Affected people experience rejection, and stigma and discrimination, which can result in significant emotional consequences. Overall functioning and the quality of life of such individuals can be further affected by this exclusion and psychosocial impacts. Little or no attention is presently paid to the emotional impact of this disease in the overall management of people affected. The study, therefore, aimed to determine the prevalence and severity of depression, as well as associated socio-demographic factors, in individuals with physical disfigurement from lymphatic filariasis in Plateau State, Nigeria. Ninety-four consecutive consenting individuals with physically disfiguring lymphatic filariasis at 5 established treatment centers across Plateau State, Nigeria, were recruited and had semi-structured sociodemographic, Patient Health (PHQ-9), the depression module of Composite International Diagnostic Interview (CIDI) and Rosenberg Self-esteem questionnaires administered using a 2-stage design. Twenty percent of the sample were found to be depressed, while history of mental illness, duration of the illness, being unemployed, and religion were predictive of depression. High self-esteem was negatively correlated. The study underscores the need to go beyond just the physical needs of individuals with lymphatic filariasis. Management must be holistic and attention must be focused on the emotional sequelae of lymphatic filariasis.
| Neglected Tropical Diseases (NTDs) are a group of disabling conditions that are among the most common infections affecting the world’s poorest people [1]. Many NTDs lead to chronic and often disfiguring conditions that result in significant disability and affect more than 1 billion people across the world [2].
Lymphatic filariasis (LF) is a mosquito-borne disease caused by filarial parasitic worms like Wuchereria bancrofti, Brugia malayi and Brugia timori [3]. Global estimates suggest that 120 million are affected in 80 countries throughout the tropics and subtropics with people at risk exceeding 1.3 billion [4, 5]. LF often manifests as enlargement of the entire leg or arm, the genitals (scrotal hydrocele in men), vulva and breasts leading to significant physical disfigurement [3]. Significant social stigma is associated with this stage of the disease [6], and the psychosocial problems linked with the condition are believed to be more severe than the physical ones [7].
In the 2010 Global Burden of Disease study, the Disability Adjusted Life Years (DALYs) associated with depressive illness in lymphatic filariasis was found to be twice that of the physical consequences of the disease (5.09 million global DALYs vs 2.78 million respectively) [8]. Despite this report, very few studies have been carried out to explore the association between depressive illness and lymphatic filariasis.
This study, therefore, aimed to determine the prevalence and severity of depression and the social and clinical factors associated with depression in individuals with lymphatic filariasis in Plateau State, North Central Nigeria.
The study was carried out in five catchment areas in Plateau State where people living with LF periodically meet to receive physical care and support. These centers are run by the Carter Center, a US-based international Non-Governmental Organization (NGO). They include Tudunwada (Jos North Local Government Area (LGA)), Nyes (Mangu LGA), Ampier (Kanke LGA), Dadur (Langtang LGA) and Gwamlar (Kanam LGA). These sites include both rural and urban communities and were chosen as they have a high prevalence of LF.
The study population included all individuals accessing care at these five centers in Plateau State with obvious physical manifestation of the condition and who were available on site on the day of the visit of the research team. Advance notice was sent to all the Centers about a month before the visit, inviting potential participants to attend. There is no reason to believe that the attendees on these clinic days vary systematically from those attending on other days, though they are characterised by their access to clinical and rehabilitative care, which is not universal in Nigeria.
A cross-sectional 2-stage descriptive convenience study design was adopted. It involved the recruitment of all available and consenting individuals with physical manifestations of LF. Data was collected using a semi-structured socio-demographic questionnaire designed by the research team and a 9-item Patient Health Questionnaire (PHQ-9) for the first stage. Those who screened positive (with a score of 5 and above) were further interviewed using the Depression module of the Composite International Diagnostic Interview (CIDI) to make definitive diagnosis and rate the severity of the depression. Rosenberg Self-esteem scale was used to assess self-esteem in the study population.
On each day of the visit to a particular center, all participants were briefed on the purpose of the study. Each participant was then approached, given a more detailed explanation, consent obtained and the interview conducted in private. Six researchers (two consultant psychiatrists, a clinical psychologist, a social worker and two resident doctors in psychiatry) administered the socio-demographic, Rosenberg self-esteem and the PHQ-9 questionnaires to the participants and scored the PHQ-9. Those who scored 5 and above were sent to 2 consultant psychiatrists trained in the use of CIDI for the second stage. The participants were then formally diagnosed and rated for severity of the depression.
All those who screened positive for depression using the CIDI were given advice and information (psycho-education), and those who had moderate and severe depression were referred to a nearby comprehensive health facility for further treatment. They have subsequently been followed up in an on-going psychosocial service being developed at the Centers.
Ethical clearance for the study was obtained from the Institutional Research Ethical Committee of Jos University Teaching Hospital, Plateau State. Signed or thumb-printed written consent was individually obtained from each participant after due explanation of the purpose of the study and the voluntary nature of their participation. Respondents below the age of 18 years, required the consent of their parent or guardian in addition to their assent. The medical data as well as respondent scores on the instruments were anonymized in order to protect confidentiality.
Data analysis was carried out using the Statistical Package for Social Sciences (SPSS), version 21, software using descriptive statistics to yield frequencies, percentages and proportions. Level of significance was kept at 5%. Logistic regression was used to identify predictors of depression among the respondents in this study using a few of the co-variates.
A total of ninety-eight participants, who met the inclusion criteria and gave consent for the study, were interviewed. Ninety-four (95.9%) had full documentation and were analysed. The majority, 58 respondents (61.7%), were female. Other socio-demographic details are provided in Table 1.
The median duration of illness was 17 years (IQR 7.0–30.0 years). Twenty-three (24.5%) had the illness for more than 30 years. 21 respondents (22.3%) rated their level of functioning as poor. Other details of functioning, as well as perceived adequacy of support are presented in Table 2.
Nineteen respondents (20%) met criteria for depression, using CIDI, with the severity of the depression being Mild [8 (42.1%)], Moderate [6 (31.6%)] and Severe [5 (26.3%)]. See Fig 1.
Fourteen of the depressed respondents (73.7%) were female. Furthermore, 69 respondents (73.4%) reported low self-esteem. See Table 3.
Logistic regression analysis revealed that history of mental illness (OR 40.83, p = 0.008); duration of the illness between 11–20 years (OR 5.02, p = 0.079), being unemployed (OR 12.71, p = 0.003) and Self Esteem (OR 0.09, p = 0.004) were predictive of depression in the cohort. High self-esteem was negatively correlated to depression. See Table 4 below.
Lymphatic filariasis cuts across all age-groups affecting both the elderly and the young with a slight preponderance in the mid to older age group. The over-representation of women (61.7%) in the study population is in keeping with the population of people receiving care at the Centers. The percentage of the participants who were separated/ divorced was higher than the expected in the general population (35.1% as against 6.8% in the general population) [9] and is in keeping with typical findings among this study population in relevant literature. They are less likely to be married, often described as the “last choice” in qualitative studies, and when the disease manifest itself during their married life, it may lead to divorce (7, 8). Nearly half of the study population (46.8%) had no formal education but this is not a sharp departure from typical rates in the rural population of the country [9]. The participants were largely self-employed as farmers and traders, which reflects the usual occupation of the people in the study areas. The average monthly income of more than three-quarter of the study population was less than N10, 000 (<$50 monthly) and this was described by more than 80 percent of the participants as inadequate. This is in keeping with the fact that NTDs principally impact the world’s poorest populations, as well as being a driver to poverty by itself [1]. The median duration of illness was 17 years (IQR 7.0–30.0 years), with about a quarter being over 30 years, illustrating relatively early onset and the chronic nature of the condition.
The percentage of those found to be depressed among the people with lymphatic filariasis (20%) is higher than the reported prevalence of depression among adults in Nigeria (3.1–5.2%) [10, 11, 12] and in primary care patients [13]. It is, however, similar to reported prevalence of depression in chronic medical conditions [14], who have been found to have two- to three- fold higher rates of major depression when compared with age- and gender- matched primary care patients [15, 16].
More women were found to be depressed among the study population, with a female to male ratio of almost 3:1. This large difference cannot be explained by the higher representation of women (F:M ratio of 1.6:1) in the study sample, and is also higher than the reported two-fold greater prevalence of Major Depressive Disorder in women than men in the general population [17]. This may be associated with greater social exclusion or stigmatisation experienced by women than men.
History of mental illness is an important and recognised risk factor, as such individuals are at greater risk of a relapse or development of another mental illness. Individuals with previous history of mental illness are almost 41 times more likely to be depressed than those without a previous history in our study.
Participants with low self-esteem, as measured by the Rosenberg self-esteem questionnaire in this study, were more likely to have depression when compared to those with high self-esteem. This could be attributable to the fact that people with depression are more likely to have low self-esteem. Our study, and many others in the region [18] and elsewhere have found that low self-esteem is often seen in people with NTDs possibly as a consequence of self-stigma, discrimination and limitations in functioning. This would suggest that high self-esteem is protective against the development of depression among this group. This is in keeping with recognised mechanisms linking stigma and depression, and may point towards potential therapeutic or health promotion interventions [19]. The link between social stigma and mental illness, including depression, are reported to be cyclical and reinforcing [19]. Self-stigma, a self-imposed restriction of expectations is common and may negatively affect expectations and motivation to engage socially [20]. In fact, people affected by LF will experience both the stigma associated with their physical condition and of their comorbid mental health problems.
Duration of the illness may be associated with higher prevalence of depression for a number of reasons. This may include the chronicity of exposure to risk factors inherent in the condition (disfigurement, pain, disability) which would be further compounded by the severity of the illness (likely to be greater in longer-term illness). It is also possible that there will be a greater extent of social drift over time. In keeping with this, unemployment was also found to be associated with greater prevalence of depression. This is the pattern found with many psychiatric disorders, where there is a vicious cycle of mental illness provoking poverty (through mechanisms summarised as social drift) and poverty being a risk factor for mental illness (social causation, or poverty as a social determinant) [21].
Beyond the physical burden of living with lymphatic filariasis, people with the condition also have significant psychiatric complications (particularly depression). Appropriate treatment for depression has been found to improve outcomes in other chronic conditions [22]. Therefore, given the high prevalence of depression, providing access to mental health screening and interventions should be integral to NTD programmes. Thus, in addition to attending to physical needs, emotional needs should also be routinely assessed and catered for.
It is recommended that staff should be sensitised to the high risk of depression and be trained to recognize basic signs and symptoms of depression. Simple screening instruments for depression, such as the PHQ-9 can also be utilized in routine clinics, and those found to be depressed can be treated or referred appropriately. Health promotion strategies geared towards assessing and addressing factors associated with depression (for example self-esteem) should be incorporated in routine community engagement, including health talks in the clinic. These are simple steps that can easily be incorporated into specific NTD services, and general health care for endemic populations, providing the possibility of improving the quality of life of affected persons, and reducing the negative impact of depression on an already marginalised population.
|
10.1371/journal.pntd.0005082 | Rift Valley Fever Virus Circulating among Ruminants, Mosquitoes and Humans in the Central African Republic | Rift Valley fever virus (RVFV) causes a viral zoonosis, with discontinuous epizootics and sporadic epidemics, essentially in East Africa. Infection with this virus causes severe illness and abortion in sheep, goats, and cattle as well as other domestic animals. Humans can also be exposed through close contact with infectious tissues or by bites from infected mosquitoes, primarily of the Aedes and Culex genuses. Although the cycle of RVFV infection in savannah regions is well documented, its distribution in forest areas in central Africa has been poorly investigated.
To evaluate current circulation of RVFV among livestock and humans living in the Central African Republic (CAR), blood samples were collected from sheep, cattle, and goats and from people at risk, such as stock breeders and workers in slaughterhouses and livestock markets. The samples were tested for anti-RVFV immunoglobulin M (IgM) and immunoglobulin G (IgG) antibodies. We also sequenced the complete genomes of two local strains, one isolated in 1969 from mosquitoes and one isolated in 1985 from humans living in forested areas. The 1271 animals sampled comprised 727 cattle, 325 sheep, and 219 goats at three sites. The overall seroprevalence of anti-RVFV IgM antibodies was 1.9% and that of IgG antibodies was 8.6%. IgM antibodies were found only during the rainy season, but the frequency of IgG antibodies did not differ significantly by season. No evidence of recent RVFV infection was found in 335 people considered at risk; however, 16.7% had evidence of past infection. Comparison of the nucleotide sequences of the strains isolated in the CAR with those isolated in other African countries showed that they belonged to the East/Central African cluster.
This study confirms current circulation of RVFV in CAR. Further studies are needed to determine the potential vectors involved and the virus reservoirs.
| Rift valley fever virus (RVFV) is an arthropod-borne virus that causes serious illness in both animals and humans. RVFV is transmitted by direct contact with infectious tissues or by the bites of infected mosquito species of the Aedes and Culex genuses. Its distribution in tropical forests in central Africa is poorly documented. We assessed the current circulation of RVFV among livestock and humans in the Central African Republic (CAR) by detecting anti-RVFV immunoglobulin M (IgM) and immunoglobulin G (IgG) antibodies in sheep, cattle and goats and in people living in Bangui who were considered at risk. We also sequenced the complete genomes of two local strains, one isolated in 1969 from mosquitoes and one isolated in 1985 from humans living in forested areas. Sheep were the most frequently infected ruminants. IgM antibodies were found only during the rainy season; the frequency of IgG antibodies did not differ according to season. No evidence of recent RVFV infection was found in humans at risk; however, 16.7% had evidence of past infection. Phylogenetic analysis showed a perfect match of CAR strains with the East/Central African cluster. Our results confirm current circulation of RVFV in CAR. Further studies should be conducted to determine the vectors involved and the virus reservoirs.
| Rift Valley fever (RVF) is a viral zoonosis that affects mainly animals but is also found in humans. It is caused by an RNA virus of the Phlebovirus genus (Bunyaviridae family), the genome consisting of three RNA segments: large, medium, and small [1,2]. RVFV is transmitted mainly by infected mosquitoes of the Aedes and Culex genuses, but humans can be contaminated by direct contact with blood (e.g. aerosols, absorption) or tissues (e.g. placenta of stillborns from infected animals) [3,4].
The virus was first identified in 1930 during an epidemic that caused deaths and sudden abortions among sheep on the shores of Lake Naivasha in the Great Rift Valley in Kenya [5,6]. Since then, the virus has spread to most African countries. The disease occurs in endemic and epidemic forms along the east and south coasts of Africa, in West Africa, in Madagascar [7,8] and as far north as Egypt, with a recent outbreak in the Arabian Peninsula [9,10]. Severe episodes of RVF have been reported among humans and animals in southern Africa [11,12,13,14]. The animals most frequently infected are sheep and cattle, followed by goats, with heavy economic losses due to abortions and high mortality rates among juvenile animals [15,16]. RVFV antibodies have been detected in many wild animal species, including ungulates in Kenya [17,18], bats in Guinea [19], and small vertebrates in Senegal and South Africa [20,21]; however, their role in maintenance of the virus in the ecosystem during inter-epidemic periods and their contribution to amplifying outbreaks remain unknown.
RVFV was first isolated in the Central African Republic (CAR) in 1969 from a pool of Mansonia africana mosquitoes [22]. It was identified as the causative virus of RVF in 1983 [23,24]. RVFV-specific antibodies have since been detected in humans, and 15 strains of RVFV have been isolated from humans and sylvatic mosquitoes in CAR [25,26], although no RVF outbreak has been reported. Current circulation of RVFV in the CAR is unknown, after a gap of two decades without surveillance; however, as animal breeding plays a large part in the economy of the CAR, an epidemic involving humans and animals is possible.
We undertook a study to assess the current circulation of RVFV in livestock and humans in the CAR. We also sequenced the genomes of local strains isolated in 1969 from wild mosquitoes and in 1985 from humans in a forested area of the country in order to determine the genetic diversity of these strains which will serve as reference for the future.
We performed a prospective cross-sectional study in two phases between November 2010 and November 2012.
The national ethical and scientific committees in charge of validating study design in the CAR approved the study design (No. 9/UB/FACSS/CSCVPRE/13). The study was described orally before blood samples were collected from human participants, and participants were included only if they gave written consent; for participants aged ≤ 18 years, a parent or guardian provided written informed consent. The informed consent form included a clause permitting use of the participants’ biological specimens for future research. No endangered or sheltered animal species were used in the survey. Verbal consent for testing their animals was acquired from farmers after the objectives of the study had been explained. Once permission was obtained for blood sample collection, an experienced veterinarian bled the animals gently.
Samples were taken during the dry season in the first year and at the same sites during the rainy season in the second year. Blood was collected from sheep, cattle, and goats at three localities: the livestock market situated 13 km north of Bangui for cattle, the Ngawi market in Bangui commercial centre, and Ndangala village located 30 km south of Bangui for sheep and goats (Fig 1). The sex and age of each animal were noted. All animals under 3 years of age were considered juveniles and those over 3 years as adults. Blood samples were also collected from people at risk, such as stock breeders and people working in slaughterhouses and livestock markets. From both animals and humans, venous blood samples were collected in 5-mL Vacutainer tubes (Becton Dickinson, Franklin Lakes, New Jersey, USA), which were placed in a cooler, transported to the laboratory, and centrifuged at 2–8°C for 10 min at 2000 rpm. Each serum sample was separated on collection into two aliquots and stored at –20°C until analysis.
Each person who agreed to participate in this study completed an anonymous questionnaire that included demographics.
The mosquitoes were collected in sylvan environments in 1969, identified, and grouped into pools of up to 30 individuals per species per site; samples were stored at –20°C for a maximum of 4 days in the field, transported to the Institut Pasteur in Bangui, and stored at –80°C until virus isolation. The virus was isolated and amplified by four serial passages in suckling mice brain, as described by Saluzzo and colleagues [27]. The brain suspensions were then lyophilized and stored in sealed glass vials at room temperature until use.
Serum samples were analysed for the presence of immunoglobulin M (IgM) and immunoglobulin G (IgG) antibodies to RVF with a SPU-02 RVF IgM and IgG Biological Diagnostic Supplies Ltd (ELISA) kit according to the manufacturer’s instructions. Briefly, plates were coated with a recombinant nucleocapsid RVFV antigen diluted 1:1000 in sodium bicarbonate buffer (pH = 9.6), covered with plate seals and incubated at 4°C overnight. Unbound antigen was removed by washing three times for 15 s each with PBS-T. Plates were then blocked with 10% skimmed milk in PBS (PBS-SM) at 37°C for 1 h and then washed. Test sera were added in duplicate at a dilution of 1:400 in 2% PBS-SM and incubated for 1 h at 37°C. The plates were washed once more, and HRP-conjugated anti-human IgG antibody, diluted 1:25 000 in 2% PBS-SM, was added to each well and incubated for 1 h at 37°C. After a final wash, chromagenic detection of HRP and absorbance measurement were performed as described previously. Negative and positive control sera were included for each plate. Sera samples were considered positive if their calculated optical density (OD) was ≥ 0.29 (net OD serum/net mean OD positive control).
RVFV was isolated from three samples: two strains isolated in 1969 from mosquitoes (ArB1986 and HB74P59) and one from human serum in 1985 (HB1752). It was amplified by inoculation into the brains of newborn mice in a laboratory of biosafety level 3. A brain suspension was prepared, lyophilized, and stored in glass vials at room temperature, and viral RNA was extracted with a QIAmp viral RNA Mini kit (Qiagen, Valencia, California, USA) according to the manufacturer’s instructions [28]. The extracted RNA was treated with Turbo DNase (Life Technologies Inc., Carlbad, California, USA) and then retro-transcribed into cDNA with a SuperScript III First Strand Synthesis kit in the presence of random hexamers. The cDNA generated was amplified with Phi29 enzyme as described previously [29]. A fixed amount of amplified DNA was sequenced in an Illumina Hi-seq 2000 sequencer. An average of 30 × 106 single reads with 100 bases was obtained for each sample [28]. The quality of reads was assessed by FastQC, and the sequences were selected according to their quality. All reads corresponding to the mouse genome sequence were filtered by mapping with Bowtie 2.0 software on a Mus musculus Mn10 sequence. The viral reads corresponding to the RVFV genome were selected by a similarity approach with BLASTN search tools [30]. All selected viral reads were assembled with Ray software, with k = 25, to obtain the full-length viral genomes [28].
In order to validate our approach for obtaining the complete sequence of RVFV by high-throughput sequencing, the RNA was extracted from three viral strains (HB74P59, HB1752, and ArB1986), and fragments of the small, medium, and large segments were amplified and sequenced. The sequence obtained from HB74P59 was compared with those obtained previously by Bird et al. in 2007 [31]; no difference was found in three segments obtained by high-throughput sequencing and classical Sanger sequencing. Moreover, no difference was found in the three segments of strain ArB1986 isolated from an Aedes palpalis mosquito and that of a strain isolated in the same city, Loko-Zinga, in 1969 but from another arthropod, Mansonia africana.
The distribution of serological results for RVFV was analysed by species, season, and human age and sex and presented as proportions. The effect of each variables on the RVFV positivity was examined using the chi-squared test. P values < 0.05 were considered statistically significant. The variables were then put in a Logistic regression model in stepwise manner. The likelihood ratio test was used to compare the model with and without the variable. In case there was no evidence that the variable fitted, it was dropped (parsimonious model). Statistical analyses were performed with STATA software version 11.
A total of 1271 animals were sampled, comprising 727 cattle, 325 sheep, and 219 goats (Table 1). The overall seroprevalence in animals was 1.9% for anti-RVFV IgM antibodies and 8.6% for IgG antibodies. The seroprevalence varied significantly by ruminant. The IgM antibody titre, which indicates recent circulation of RVFV, was 4.3% in sheep, 1.4% in goats, and 1.1% in cattle (P < 0.0001), whereas the IgG seroprevalence was 12.9% in sheep, 7.8% in cattle, and 5.0% in goats (Table 2). However, the multivariate analysis showed that IgM and IgG seropositivity rates in sheep were higher than other species (OR = 1.8, 95% CI = 1.1–3.0) (Table 1). No significant difference in seropositivity was found between male and female animals (Table 3). The IgG seropositivity rate was 9.6% in adults and 5.1% in juveniles (P < 0.02), but no significant difference in positivity for anti-RVFV IgM antibodies was found between juveniles and adults (Table 3). The IgM and IgG seropositivity rates varied significantly according to the origin of the sample. No animals positive for IgG antibodies were found in Ndangala, whereas most of those positive for IgM antibodies were originated from this site, and cattle market are less likely to be IgM positive (OR = 0.1, 95% CI = 0.0–0.5) (Tables 1 and 3). All animals with positive IgM were found in the rainy season, and IgG seropositivity was more pronounced during dry season (Table 3). None of the livestock owners reported cases of abortion or death in the months before sampling that would indicate RVFV infection in their herds.
Blood samples were collected from 335 people who were regularly in contact with blood from the animals. The mean age (±SD) was 36.3 years (±18.1), and the sex ratio (M/F) was 6.0/1 (287/48). No evidence of recent RVFV infection (absence of IgM) was found in human samples; however, 16.7% had evidence of past infection (IgG alone). Of these, 7.7% were stock breeders, 6.6% were butchers, 1.5% were slaughterhouse workers, and 0.9% were veterinarians (Table 4). A higher positivity rate was observed among people over 25 years of age (P = 0.04), and 17.8% (51/287) of males and 10.4% (5/48) of females were positive for IgG (P = 0.29) (Table 4).
A third RVFV strain obtained by high-throughput sequencing was isolated in Bangui at the end of December 1984 (HB1752). Genomic analysis of three segments showed that it was identical to a strain isolated in Bangui 3 months earlier; however, RVFV strains isolated from vectors in 1969 and from human cases in the CAR several decades later had different nucleic sequences, even though they belonged to the same cluster (Fig 2). The complete genome sequence was made available to GenBank (HB1752 strain small, medium, and large accession numbers KJ782452, KJ782453, and KJ782454 respectively; ArB1986 strain small, medium, and large accession numbers KJ782455, KJ782456, and KJ782457, respectively).
This study, the first in the CAR since RVFV was isolated in 1985, shows that the virus continues to circulate in central Africa. The overall prevalence in animals in this study was lower than that reported in Comoros in 2009 (39% in sheep and 33.5% in goats) [32], in Madagascar in 2008 (24.7% in small ruminants) [33], and in Mozambique (35.8% in sheep and 21.2% in goats) [34]. The same ELISA kits were used in all these studies, suggesting that the differences are due to climatic factors, entomological parameters, agro-ecological conditions, or sampling strategies.
The higher prevalence in sheep is consistent with previous work, indicating that this species is preferentially infected with RVFV [15]. Most infected animals, especially sheep, were found in Ndangala, a rural forested area south of Bangui that has more rainfall than the rest of the country, with lower temperatures and constant humidity in the rainy season, during which time there is little husbandry. The high prevalence observed at this site, with the presence of IgM antibodies, suggests endemic virus circulation, which would be maintained by a sylvatic cycle involving wild animals and mosquitoes, as suggested by Olive et al. [35]. In a previous study in a forested area of the CAR, RVFV was isolated from wild mosquitoes, including Ae. palpalis [26]. In the rainy season, there are many potential breeding sites, which increases the density of vectors and subsequently increases transmission of arboviral diseases such as RVFV. IgM, which indicates recent infection, was present only during the rainy season, but IgG was also significantly associated with the rainy season. These finding are consistent with those in Mauritania and Senegal that indicate that the risk factors for RVF are linked to heavy rainfall and the presence of large temporary masses of surface water [36,37]. In a previous study, seropositivity for RVFV was associated with increased numbers of mosquito vectors [38]. Although we did not conduct entomological surveys, recent entomological surveillance for yellow fever identified several species of Aedes mosquito, including Ae. cuminsii, Ae. circumluteolus, and Ae. palpalis [39], which are known vectors of RVFV [40].
The seroprevalence of RVFV was higher in adult than in juvenile animals. Similar results were reported in Mauritania and Senegal [32,33], supporting the hypothesis of endemic circulation of the virus, as older animals would have longer exposure than younger ones [34]. The presence of IgG in young animals (< 3 years) in this study suggests recent circulation of the virus. This result is compatible with the IgM titres in each species, with high titres in samples taken from sheep in Ndangala (Table 2). The absence of IgG in animals from this region indicates that introduction of the virus south of Bangui is recent. Recent introduction of the virus associated with environmental modifications such as deforestation, population displacement due to the socio-political crisis, and introduction of a new vector competent for RVFV [41] could increase the risk for emergence of an epidemic.
As no epidemic of RVF or obvious clinical signs of the disease (such as abortions) was observed, the infections were minor or sub-clinical [34]. In a study in Madagascar, circulation of RVFV during the dry season did not result in clinical cases [33]. Viral activity may be maintained in mosquitoes near rivers that do not dry up during the dry season, resulting in a low level of transmission among domestic animals. These observations and the recent epidemics in East Africa illustrate the risk for introduction of pathogenic strains of RVFV to CAR from countries such as South Sudan, which shares a long border with CAR and has had many epidemics and epizootics of RVF.
The absence of anti-RVFV IgM antibodies among people regularly exposed to animals would appear to indicate that contact with the virus is uncommon and the public health risk is low. Nevertheless, the presence of IgG among breeders and butchers, who are in contact with the blood of these animals, should alert the authorities to strengthen surveillance of circulation of this virus. Studies to isolate the virus in the vector (mosquitoes) and longitudinal studies in sheep, goats, and cattle should be conducted to detect clinical cases, particularly among sheep and herders in Ndangala village, where evidence of current circulation of the virus was found. As most of the inhabitants of the rural areas in which the small ruminants were sampled are farmers who share the same environment as their animals and may have the same exposure to mosquitoes, a study should also be carried out to determine the prevalence of RVFV antibodies and to establish whether RVF occurs regularly in these zones and is thus a neglected cause of morbidity and mortality.
Our study was limited to Bangui and neighbouring areas because animals are brought to the capital from all regions of the country. As we were unable to isolate recent strains of the virus, we could not establish the precise geographical origin of the viral strains currently circulating in the CAR. Nevertheless, on the basis of genomic data for old strains, the same strain may be circulating relatively freely in several vectors in a defined geographical area over a long period. Furthermore, although IgG responses persist for several years in continually exposed people, we were unable to obtain a second sample at an interval of 2 weeks and therefore could not demonstrate seroconversion, which would support recent infection. The low IgG titres and high IgM titres found in the Ndangala region suggest recent introduction of the virus into this forested area, probably associated with illegal movement of sheep and goats from the Democratic Republic of Congo.
In view of the highly precarious situation in CAR, a large-scale molecular study will not be possible in the short term; furthermore, the socio-political upheaval in the country may change socioeconomic and environmental conditions and the time of infection before the study is conducted.
Comparison of the sequences of strains isolated from vectors and humans in the CAR with those isolated in other African counties show that they belong to the East/Central African cluster, confirming RVFV strain exchanges among geographical areas. Propagation of RVFV from East Africa to other regions was noted in Saudi Arabia and Yemen in 2000–2001 [42] and in Chad in 2001 [43] during previous RVF outbreaks.
The results of this first study conducted in both humans and animals in the CAR are important for public health. They shows a high prevalence of RVFV in an area where neither epidemics nor clinical cases of RVF have been reported previously. These results are also important because, in the forested area south of Bangui where there is little husbandry, there is nevertheless low virus noise, which might suggest the presence of a reservoir that has come nearer to human habitats. Unexpectedly, we found low IgM titres in regions of previous intensive animal husbandry, because animal density in these areas has fallen sharply due to the migration of herders in response to the continuing instability in the CAR.
Other studies are required to elucidate and measure the environmental risk factors for infection with RVFV in order to predict epidemics, and entomological studies should be performed to identify all the potential vector species to better understand the ecological and climate factors that favor the distribution of RVFV.
|
10.1371/journal.pbio.1002123 | Convergent Evolution of Mechanically Optimal Locomotion in Aquatic Invertebrates and Vertebrates | Examples of animals evolving similar traits despite the absence of that trait in the last common ancestor, such as the wing and camera-type lens eye in vertebrates and invertebrates, are called cases of convergent evolution. Instances of convergent evolution of locomotory patterns that quantitatively agree with the mechanically optimal solution are very rare. Here, we show that, with respect to a very diverse group of aquatic animals, a mechanically optimal method of swimming with elongated fins has evolved independently at least eight times in both vertebrate and invertebrate swimmers across three different phyla. Specifically, if we take the length of an undulation along an animal’s fin during swimming and divide it by the mean amplitude of undulations along the fin length, the result is consistently around twenty. We call this value the optimal specific wavelength (OSW). We show that the OSW maximizes the force generated by the body, which also maximizes swimming speed. We hypothesize a mechanical basis for this optimality and suggest reasons for its repeated emergence through evolution.
| How would animal life differ if it evolved again on Earth or any other habitable planet? If variation and selection can overwhelm all the other factors that might impede the approach to an optimum, then traits of animals that fulfill similar functional needs—such as camera-type eyes for seeing or wings for flying—are more likely to emerge independently and repeatedly. In aquatic animal swimming, one performance criterion is the Strouhal number (St), which specifies the frequency of fin movement for maximum propulsive efficiency in those animals that use the common “body/caudal fin” swimming mode, such as trout. Here, we use a combination of computational modeling, a robotic knifefish, and measurements of animal swimming behavior to study another widespread form of locomotion—“median/paired fin” swimming, used by animals as diverse as cuttlefish, triggerfish, and rays. Our studies provide quantitative evidence for a complementary performance criterion, called the optimal specific wavelength (OSW), which determines the wavelength of fin movement required for maximum propulsive force or thrust. Adherence to the OSW has independently emerged within eight clades of animals in three different phyla, including vertebrates and invertebrates, encompassing over a thousand species.
| How would life look if it evolved again on Earth, or for that matter, on any other habitable planet? The question of the role of chance versus necessity in evolution is a foundational issue in biology [1–3]. Gould gave us the metaphor of the “tape of life” for the evolution of life and argued that if it were somehow rewound and started again, life would have taken a very different course [4]. Conway Morris has argued that, on the contrary, the laws of physics limit the number of good solutions that are within reach of evolution, and that therefore we should expect life to take a similar course upon rewinding [5]. Examples of convergent evolution, such as wings on insects, birds, and mammals, are considered supporting evidence for this hypothesis. But our understanding of convergent evolution, as reflecting the dominance of natural selection plus variation over factors such as developmental constraints, pleiotropy [6], phylogenetic inertia, genetic drift, and other stochastic processes [3], is held back by a lack of quantitative arguments. Such arguments would expose the links from physical principles to the biological phenomena and help us understand where evolution is likely to converge to the same result or diverge to a wide variety of solutions [7].
Here, we present just such arguments for a phenomenon that unifies a vast diversity of swimming organisms, from invertebrates, like cuttlefish, to vertebrates, like cartilaginous and bony fish. Unlike the case of the convergently evolved wing, a morphological feature, here the evolved feature is a pattern of movement [8–12] that occurs across a morphologically diverse set of moving appendages on aquatic animals. These animals swim by undulating elongated fins while keeping their body relatively rigid (Fig 1). Their fins run lengthwise with the body rather than crosswise to the body like the tailfin of a trout. The fins occur along the midline of the body, either along the belly or back (Fig 1N–1S, knifefish), as a dorsoventral pair (Fig 1L, triggerfish), or as a left–right pair (Fig 1C–1K, cuttlefish, skates, and rays). Movement by means of these fins is therefore called median/paired fin swimming, in contrast to the more common swimming style of fish like trout, in which a portion of the body and the caudal tailfin is moved, termed body/caudal fin swimming [13].
We find that in all cases where the wavelength of the traveling wave of median/paired fins has been documented during steady swimming, or can be extracted from readily available videos, the wavelength is about twenty times the mean amplitude (ã) of movement (mean = 19.5, standard deviation (STD) = 3.0, n = 22, Fig 1 and S2 Table). This includes cases, such as swimming of the cownose ray, in which the wavelength is greater than the fin length. These fins appear to only oscillate up and down to a casual observer but in fact undulate in the same manner as described earlier.
The ratio of the wavelength of an undulation to its mean amplitude is called the specific wavelength (SW). This parameter has been used in the past to study undulatory swimming [14–17] under water and under sand [15,16]. However, the finding in this work that there is a particular value of the SW that is observed across a diverse set of animals is new. We call this the optimal specific wavelength, or OSW. The OSW identifies the optimality in kinematics associated with undulatory patterns, in contrast to prior work [18], in which the focus was on quantifying bends solely due to propulsor flexibility.
Many of the organisms that swim with the OSW, including invertebrates such as cuttlefish and flatworms, jawed vertebrates such as rays and skates, and ray-finned bony fishes such as gymnotiformes, gymnarchids, and notopterids have no known ancestor that swam by means of median/paired fins (See S1 Methods Section 2). Swimming through undulation of median/paired fins with the OSW has therefore independently arisen at least eight times in three phyla (Chordata, Mollusca, and Platyhelminthes) among the Bilateria (Fig 1).
The values of SW, shown in Fig 1 for a variety of median/paired fin undulatory swimmers, converge to a narrow range (mean = 19.5, STD = 3.0, n = 22, S2 Table). Is there a mechanical optimum due to the physics of undulatory swimming that has driven evolution, in multiple instances, toward this narrow range of values of SW? An initial clue was obtained by our group from simulations of undulating neotropical knifefish fins [19] and measurements of the swimming speed of an undulatory fin robot [20]. Prior experimental work and current simulations show that swimming speed is highest when the number of undulations present along the fin (the fin length divided by the wavelength of the traveling wave progressing down the fin) is around two, similar to what is found in measurements of live knifefish [21,22]. The simulations were carried out on two sizes of fins. The first fin was similar in size and shape as the ventral elongated ribbon fin of a specific type of knifefish, the black ghost knifefish (A. albifrons, Fig 1N), where the fin measures 10 cm long by 1 cm tall. The second fin was a smaller fin measuring 2 cm long by 0.4 cm tall. A smaller fin was used because simulating many cases of a knifefish-sized fin for a more detailed parametric study, which we report later in the paper, was not practical due to limited computational resources. The experiments were with a robotic fin that was approximately three times larger in length and height (32.6 cm long by 3.6 cm tall) compared to the knifefish-sized fin. A larger robotic fin was built because of limitations of motor packing density for the smallest motors that could provide sufficient torque while providing an adequate number of artificial fin rays (32 rays [20], compared to 148 found in A. albifrons [23]). In Fig 2, we show the free swimming speed of the robot [20] and a simulated knifefish-sized fin [24,25] as a function of number of undulations along the fin. We also plot the propulsion force generated by undulating robotic and simulated fins that are prevented from translating along their long axis. We find that swimming speed and propulsive force are well correlated. This correlation is expected, since maximizing propulsive force or thrust is likely to enable the fastest swimming speed. Hereafter, we will present our analysis only in terms of propulsive force.
As we varied the length of the simulated fin and the robotic fin for this study, while keeping other kinematic parameters the same, we were surprised to find that the number of undulations that resulted in maximal propulsive force (correlating to maximal free swimming speed) changed, as shown in Fig 3A. In this work, we show that if we examine force versus SW, instead of force versus number of undulations, the peaks of propulsive force all align at around a SW of twenty (Fig 3B). We find that the same is true when other kinematic and morphological parameters (amplitude of undulations, frequency, fin height, and fin shape) of the fin are varied (see below and S1 Methods Section 3). Because the propulsive force was maximal at an SW of around twenty across these examined parameters, we refer to SW ≈ 20 as the optimal SW or OSW.
The axial propulsive force F depends on several physical variables of the problem:
F=fn(ρ,μ,f,λ,L,h,a),
(1)
where fn denotes “function of," ρ is the density of water (0.9956×103 kg/m3 at 25°C, varying less than 5% across the temperatures and salinities of aquatic organisms), μ is the viscosity of water (0.89×10−3 N ∙ s/m2 at 25°C), f is the frequency of oscillation of the fin, λ is the wavelength of undulation of the fin, L and h are the length and the height of the fin, respectively, a = h sinθmax is the amplitude of undulation at the distal edge of the fin, and θmax is the peak angle of excursion of the fin from the midsagittal plane (see S1 Fig). We assume a condition of steady temporally periodic flow in Equation 1. In our experiments and simulations, the force was generated by a undulating but non-translating fin. Hence, the fin velocity is not a physical variable of the problem. If the fin is attached to a swimming body then the swimming velocity is another dimensional variable of the problem. The steady swimming velocity also depends on the kinematic parameters listed in Equation 1.
Other physical variables of interest that depend on those listed in Equation 1 are noted next. A is the surface area of the fin, which is approximately given by A = Lh(1 + (a / λ)2). For the parameter space of interest in this work, a / λ ≪ 1, implying A ≈ Lh. A measure of the lateral velocity of oscillation of the fin is Vl = f ã, where ã is the average amplitude of oscillation of the fin. In the definition of SW, we use the average amplitude ã rather than the amplitude a. The reason is as follows: the propulsive force generated by an elongated fin depends on the amplitude of oscillation of the fin. For a given angular excursion, the amplitude of oscillation depends on the height of the fin, which typically varies along the length of the fin (see Fig 1, and S2 Fig for a schematic). The contribution to propulsive force from the part of the fin with greater height would be larger than that part of the fin which has relatively smaller height. Therefore, we use the average amplitude ã to define SW = λ / ã (and other parameters), where a˜=hmeansin(θmaxavg)/2. hmean is the mean height of the fin and θmaxavg is the mean angle of excursion of fin rays (for more details and a fully worked example, see S1 Methods Section 1 and S1 Table).
Equation 1 shows that there are eight physical variables including the propulsive force. The viscosity and density of water are known. Thus, the propulsive force F depends on the five remaining physical variables of which the wavelength, or equivalent SW, is one variable. We find that although F depends on multiple variables listed in Equation 1, the maximum is always at SW ≈ 20, irrespective of the other parameters. This is demonstrated in Fig 4. It is seen that the OSW did not change over the range of frequencies, angles of excursion, fin heights, and fin lengths shown. Further details of this parametric study are discussed in S1 Methods Section 3. In addition to the effect of rectangular fin height, the effect of different nonrectangular fin shapes on the OSW was also investigated. Fin shapes ranging from rounded convex to triangular (such as those used in a robotic cownose ray [27]) had no effect on the value of the OSW. This is discussed further in S1 Methods Section 4.
For the parameter space we examined, the qualitative trend of F with respect to SW is independent of the size of the fin. Consequently, the OSW is independent of size over the range examined here. Although our experiments and simulations are limited to fins between 2 cm and 32 cm in length, data from 22 species of swimming animals in Fig 1 show that the OSW holds over a wide range of lengths that span two orders of magnitude, from a 2 cm long Pseudobiceros pardalis to a 255 cm long R. glesne (see S2 Table). Beyond our sample of 22 species, there are over 1,000 median/paired fin species (either within the same genera or in higher taxonomic categories) that we predict will exhibit the OSW.
We have chosen not to include the seahorse Hippocampus hudsonius within our results. This species is the sole exception we could find to the OSW, according to fin kinematics reported by Blake [28]. We decided not to include this case for the following reasons: the fin of the seahorse discussed in [28] was 2.1 cm long and around 1 cm deep. Blake reported that the fin moved with an amplitude of 0.63 cm at a frequency of 40 Hz while having 3.3 undulations. We cannot be confident that the reported kinematics are physically achievable by animals. To understand why, consider the kinematics of the black ghost knifefish [22]. The fin height is the same for the seahorse and the knifefish. The length of the seahorse fin is five times less than that of the knifefish. If the knifefish were to undulate its fin in an equivalent scenario, it would need to have around 15 undulations. The highest number of undulations along the fin of a knifefish is around four [22]. Thus, for conditions equivalent to the seahorse, the fin membrane would be subjected to extremely high strains and stresses, which seem physically unrealizable.
Convergent evolution, the repeated emergence of a trait from an ancestor that did not have the trait, is the most compelling evidence for what has been called the “robust repeatability thesis” [3], the view that the evolution of animals can be explained by the comparative selective advantage of evolutionary survivors over their inferior counterparts. This is in contrast to Gould’s view—that quite different animal forms would emerge if evolution were to rerun its course. This view of macroevolution, the “radical contingency thesis,” [3] is that early patterns of animal evolution were generated by a selection process that was unrelated to the comparative fitnesses of the selected lineages.
What is the mechanism for macroevolutionary repeatability? In the language of the calculus of variations, these examples of convergent evolution—if correctly identified as such—imply that there is a gradient in the fitness landscape toward some optimum with respect to trait in question, and this gradient is large enough to overcome competing factors such as developmental constraints, pleiotropy [6], phylogenetic inertia, genetic drift, cases where optimality in one trait results in suboptimality in another trait (e.g., [29]), and approximations of the trait which provide local but not global optimality. With a sufficiently steep gradient in fitness in place and evolutionary dynamics capable of achieving near-optimal solutions, it is only a matter of time before the mechanism of selection with variation can arrive at the optimum. As the derivative of the trait with respect to fitness is stabilizing, departures from the optimum would be self-correcting over evolutionary time.
What is the evidence of a gradient in the case of the OSW? To address this question, we examine how variation in the SW affects force production (and therefore swimming speed) across the observed range of SW (Fig 1). We then compare the results to the variation in force production for SW outside of the observed range.
By using simulation and experimental results presented in S1 Methods Section 3, we found that for a SW between 15 and 25, corresponding to the observed natural range (Fig 1), the respective propulsive force decreases by around 7.5% at most from the optimal value (see S1 Methods Section 5 and S3 Table for details). Outside the naturally observed range of 15–25, the reduction in force grows significantly. When we consider an additional 25% deviation in SW, for a range of 10–30, the decrease in propulsive force is almost 25% (see S1 Methods Section 5 and S3 Table for details). The effect of any decline in propulsive force—even less than one percent—from what it is at the OSW is amplified over the vast number of undulations an animal may make in its life. Thus, we can only speculate that the 7.5% decline in force occurring over the observed variation in SW is not large enough to overcome the many causes of suboptimality listed above, whereas the 25% decline we find beyond this range is large enough to cause selection pressure toward the OSW. Additional research is needed to establish whether this hypothesis is true.
Finally, we note another aspect of the stability of the OSW, which is its insensitivity to changes in key kinematic variables such as frequency and amplitude, as well as to changes in fin geometry such as height and shape (Fig 4, S1 Methods Sections 3 and 4).
While the previous discussion shows that once an elongated fin species has arisen it may be costly to depart from the OSW, elongated fin swimmers constitute a minority amongst all swimmers. For example, there are approximately 1,000 median/paired fin swimming species within the jawed fishes (based on ascending to the nearest non-median/paired fin swimmer at each node of the tree presented in Fig 1) compared to 33,000 total fish species currently listed within FishBase. The fastest fish, and therefore the fish that are able to produce the highest amount of force, are body/caudal fin swimmers [30,31]. The question is therefore why the slower forms of swimming exhibited by median/paired fin animals would emerge and thrive despite the prevalence of body/caudal fin swimming in ancestral species.
A similar question has arisen in simulation studies that show that a light-sensitive patch of skin can evolve through several intermediate forms into an advanced camera-type lens eye in only a few hundred thousand years [32]—why, then, are there so many existing animals with intermediate forms of eyes? Nilsson and Pelger’s answer is that camera-type lens eyes are only the best solution for certain animal–ecosystem combinations [32]. Our answer is similar: body/caudal fin swimming makes little sense in isolation. It is only within particular ecological contexts that some types of animals are able to survive better with this type of swimming than with alternative approaches.
In particular, median/paired fin swimming appears to be a low speed, low cost of transport specialization [21,30,33]. The lower amplitudes of fin movement that are possible in median/paired fin swimmers, compared to the very high amplitudes possible when the high power axial musculature is used in body/caudal fin swimmers, is therefore an advantage instead of a liability due to the lower energetic cost of transport of median/paired fin swimming [30]. The fact that median/paired fin swimming is used at lower speeds should not be confused, however, with the concept of maximizing speed by swimming at the OSW. Even when swimming at lower speeds (or whatever speed for that matter, which is determined by frequency and amplitude), for a given set of parameters (amplitude of undulations, frequency, fin height, and fin shape), if an animal swims with elongated median/paired fins, then its speed can be maximized for that set of parameters by swimming at the OSW.
The ecological circumstances in which median/paired fin swimming will be favored over the typically faster, higher cost of transport body/caudal fin mode are not clear. If we consider one group specifically, such as the Gymnotiformes, a number of ecological features stand out as possible factors. First, they live in habitats of murky water and are active at night. Therefore vision—and high speed predation based on visual guidance—is not a factor. Second, periodic reductions in oxygen levels of the rivers they inhabit [33] may favor the lower cost of transport of median/paired fin swimming. Third, species within this group sense using electric fields, a mode of sensing where there is fourth-power attenuation of signal with distance [34]. This quartic attenuation implies a penalty on gross movement in sensory acquisition tasks, since typically sensing range is extremely short [34–36]. Therefore, precise, small, and slow movements are most effective. Finally, because the field generator is in the trunk, movement of the body causes large distortions of the emitted signal, possibly favoring holding the trunk rigid [37,38] rather than using it for propulsion. Given these constraints, the elongated fins that are universally present within the more than 150 species comprising Gymnotiformes may be favored, but clearly a tremendous amount of work would need to be done to assess the relative importance of all of these factors in giving rise to this one group of median/paired fin swimmers.
While the existence of body/caudal fin swimmers and the existence of median/paired fin swimmers may or may not be subject to robust repeatability, what is clear is that if median/paired fin swimming with elongated fins and semirigid trunks emerges—as it has independently on multiple occasions according to Fig 1—it is very probable that the specific trait of swimming at the OSW will also emerge. This is because of the thrust-maximizing property of the OSW, the existence of the gradient in thrust efficacy with a stabilizing derivative mentioned above, and the insensitivity of the OSW to variations in key fin movement parameters such as frequency and amplitude and to variations in fin height and shape.
Maximizing thrust ensures that speed can be maximized for a given set of parameters, irrespective of whether the animal swims energetically efficiently during cruising, or possibly less efficiently during escape or attack maneuvers. Thus, it seems likely that aquatic animals could benefit from adhering to the OSW at all—or at least at most—times. An example of using the OSW for thrust maximization even when swimming speed is not a concern is provided by the phenomenon of counterpropagating waves along a single fin. Gymnotid fish such as the black ghost knifefish A. albifrons [22,39] and the glass knifefish E. virescens will generate undulations from head to tail (called the head wave) and undulations from tail to head (called the tail wave) along its median anal fin, producing antagonistic forces [40]. We and collaborators have shown that this enhances stability and maneuverability [40]. Published videos show indications of this pattern being more widespread than gymnotid fish, such as the oarfish R. glesne [41] and cuttlefish [42]. For E. virescens, where counterpropagating waves have been analyzed more fully, while the angular amplitude is relatively constant between the two waves, there is a distinct difference in the wavelength. As the fin is deeper rostrally, for a similar angular excursion of each fin ray, the resulting mean lateral amplitude ã of the fin will be greater in the head wave than in the tail wave (for fin profile, see S2 Fig from [40]). The fish compensates for these differences in mean amplitude with a longer wavelength for the head wave than the tail wave, keeping the SW closer to 20 than if a constant wavelength were present throughout the fin (see S4 Table). A. albifrons follows a similar trend, though more data is needed for a full analysis. Maintaining OSW for the head and tail counterpropagating waves maximizes the counteracting thrust forces which lead to increased stability and maneuverability [40].
Despite these uses of the OSW for force maximization, it should be noted that median/paired fin fish are able to switch to the higher force production mode enabled by body/caudal fin swimming when absolutely needed, such as C-starts in knifefish [43], gait changes in median/paired fin swimmers to body/caudal fin swimming as speed increases [30], and oarfish transitioning to anguilliform swimming for higher speed (for reference see the videos from [41]). A natural question is whether body/caudal fin swimmers, or those that undulate their entire body, such as eels and lamprey—or even terrestrial animals like snakes—abide by an OSW of ≈ 20. It is likely that some related criteria exist but the optimal value of the SW may be different, or it may not be a constant as in case of the animals considered here that swim by means of elongated fins while keeping their trunk semirigid. In this context, we note that in the characterization of sandfish swimming (effectively, a body/caudal fin swimmer in sand), Maladen et al. [14,15] find that the swimming speed of the sandfish is optimal at SW = 5. The issue of the applicability of an OSW-like optimum to other modes of swimming, although not within the scope of this work, merits future investigation.
One obvious way in which swimming animals could adhere to the OSW is through neural control. Whether at the level of spinal circuitry, or through descending control, this control would enforce the OSW through changing amplitude and wavelength so as to maintain a fixed ratio of around 20, as discussed below in remarks discussing the implications of the OSW for controlling robots. While outside the scope of this study, this hypothesis warrants further investigation, perhaps using the weakly electric fish model system, which is heavily used within certain domains of neuroscience [44].
Another approach is to build the OSW into the material properties of the structures that comprise the fin. The fin is made of a collagenous membrane that connects adjacent fin rays in teleosts [45] and a muscle and collagen matrix in cuttlefish [46,47]. This connective tissue limits the maximal strain, and thus the curvature of the fin cannot increase beyond a certain value. Suppose the amplitude of movement is increased for a given wavelength. This would increase the curvature and strain in some parts of the membrane. If the maximal allowable strain is being approached, then the material would relax to a larger wavelength for the new larger amplitude. A demonstration of this principle is seen in experiments in which the anterior end of a long, flexible, free-swimming foil is oscillated [48]. When the amplitude of a traveling wave along the foil increased by slightly over a factor of two, the wavelength increased by a factor of 1.6 (see Fig 4, [48]). Interestingly, the primary structural protein of aquatic animal fins, collagen, is an ancient protein predating the divergence of the three phyla where we have observed the OSW [49].
Whether by neural control or by material properties, there is an additional aspect of swimming by median/paired fins that could facilitate genetic aspects of the implementation of the OSW. For the more prevalent body/caudal fin swimming style, there are a host of other functions served by the body besides propulsion, such as holding internal organs, or providing a mirror-like surface to facilitate camouflage [50,51]. Having to move the mass of the trunk for propulsion, including such things as internal organs, is a disadvantage for the body/caudal fin style of swimming.
In contrast to most body/caudal fin swimmers, median/paired fin swimmers use their fins purely as thrust elements. In moving fluid to generate thrust, they are mostly putting energy into moving fluid instead of the mass of the membranous fin. Therefore, the collection of genes that give rise to axial elongation [52,53] and the associated elongated fins used for thrust production may have few other functions, perhaps facilitating their approach to an optimum in comparison to the genes controlling body shape in swimmers that move their trunk for swimming [6]. Further evidence in support of this idea is the finding of convergently evolved anatomical specializations for one type of body/caudal fin swimmer that minimizes trunk movement during tail fin movement, resulting in a similarly pure thrust element [12]. These are the thunniform swimmers such as tuna and lamnid sharks which largely restrict movement to their caudal fin propulsor, and in doing so are able to swim faster than most other fish. We hypothesize that regulation of the OSW, axial elongation, and fin elongation in the phyla in which we observe these traits may reflect the action of common genetic networks, as has been shown to underlie convergent evolution in the case of the electric organ [54].
Based on the OSW and the measured mean undulation amplitude ã, the number of undulations that should be present along the fin of a swimming animal, cruising at any velocity, can be predicted. This is given by
NUpredict=LOSWa˜.
(2)
We plot the measured number of undulations along with the predicted number of undulations in Fig 5 for all the same animals shown in Fig 1, using measurements of mean amplitude and assuming an OSW = 20. The number of undulations predicted is in excellent agreement with the observed number.
These predictions are for a single swimming velocity for each species. However, in two cases we have fin undulation data across several swimming velocities. Here too, the fin follows the pattern dictated by the OSW. The data for the African aba aba knifefish (G. niloticus) and South American black ghost knifefish (A. albifrons) below the dashed line in Fig 5A are at different swimming speeds. The superscripts on G. niloticus and A. albifrons relate to swimming speed, with the superscript number 1 corresponding to the slowest speed and higher numbers to higher speeds. The data show that as these fish swim faster, they reduce the number of undulations along their fins. This reduction in undulations along the fin is most likely because this is necessary to continue swimming at the OSW. To swim faster, the knifefish have to either increase the amplitude of fin undulations, or their frequency, or both. An important question for future investigation is whether the motor systems of undulatory animals provides for independent control of key kinematic parameters such as amplitude and frequency. For example, in a biologically inspired model of how coupled oscillators generate traveling waves along the anal fin of knifefish, the drive to the initial oscillator controlled frequency, while the gradient in drive across oscillators controlled the number of undulations [22], a situation in which there is no independent control of these key kinematic parameters. For the data plotted in Fig 5, to swim faster, the knifefish increase both frequency and amplitude [22,55]. Consequently, to compensate for the increase in amplitude, the wavelength is decreased to maintain the OSW.
Because the SW is independent of frequency, if an animal is swimming at the OSW and only frequency is varied to change speed, then the organism does not have to change the number of undulations along the fin to maintain the OSW. It appears that frequency control dominates at lower speeds for knifefish [21,22], which likely underlies the observation of a relatively constant number of undulations along the fin as swimming speed changes in a variety of electric and nonelectric knifefish [21,22]. Rosenberger [56,57] also mentions how some species of rays such as Gymnura micrura and T. lymma exhibit changes in fin motion with swim speed. As G. micrura and T. lymma increase their swimming speed, the number of undulations goes down. Rosenberger [57] also reported that T. lymma used smaller amplitudes at lower speeds. Consequently, to maintain the OSW, it would have to use more undulations when swimming slowly compared to when it is swimming rapidly using larger amplitudes. The same change in amplitude with speed has been mentioned qualitatively for G. micrura [56], and we would predict that quantitative measures would show that the OSW-prescribed pattern holds for this animal as well during changes in speed.
A final example of the correlation between amplitude and number of undulations is provided by the phenomenon of counterpropagating waves along a single fin discussed above. Despite traveling waves passing along the fin in opposite directions along fin membrane of differing depth (resulting in different amplitudes for similar angular excursions across fin rays), fish adjust the wavelength of undulations so as to maintain the OSW, as shown by Fig 5B.
A nondimensional number commonly used to characterize the swimming of body/caudal fin swimmers is the St (St = 2fa / U, where U is the swimming speed, 2a is the maximum peak to peak amplitude of the lateral fin excursion, and f is the frequency of tail beating). It has been reported that body/caudal fin swimmers cruise at an St between 0.2–0.4 [8,58]. It has also been reported, based on a meta-analysis of steady undulatory swimming, that the St varies over a much larger range of 0.2–1.8 [59]. Prior studies suggest that the basis of St between 0.2–0.4 in animals is high propulsive efficiency, which may relate to maintenance of attached flow [8] and hydrodynamic resonance [60], among other factors. Eloy [61], using Lighthill’s elongated body theory [62], has shown that an St between 0.15 and 0.8 minimizes energy expenditure and maximizes Froude efficiency. In swimming live knifefish, the St ranges between 0.3 and 0.6 (from data in [22]), the same range found in a study reporting data from a robotic knifefish as well as computational fluid simulations of knifefish fins [63].
A quantitative relationship between the St and the SW follows from their definitions:
StηwaveSW=γ,
(3)
where ηwave=UVw is the wave efficiency, which is the ratio of the velocity of a swimmer to the velocity of the traveling wave (frequency × wavelength) along the fin (see [64], where 1–ηwave is the “slip”). The wave efficiency can, in theory, take any value between zero and one. It has been reported that for high Froude efficiency, the wave efficiency should be in the range of 0.4–0.8 [61]. The wave efficiencies observed in swimming animals ranges from 0.1 to 0.9 [59]. The factor γ = 2a / ã, where 2a is the maximum peak to peak amplitude, while ã is the mean amplitude as previously defined, is due to the difference in how the amplitude parameter is defined in the formula for St and how we define it for the SW. γ ranges from four, for a rectangular fin similar to those of knifefish, to around eight for a deep triangular fin similar to those of some batoid fishes.
Given St and SW, the wave efficiency, ηwave, is determined according to Equation 3 for a given γ. For the range of γ from 4–8, a SW of 20 combined with the reported range of St of 0.2–0.4 [8,58] results in high wave efficiencies. As an example, the relationship from Equation 3 for γ = 6.7, that is typical of batoid fish, is shown graphically in Fig 6. The wave efficiencies are all above 0.6. Therefore, while St and SW are complementary, the observed values are also consistent with each other.
Although the St and the SW are related according to Equation 3, these two parameters are not equivalent. To appreciate this, three considerations should be noted. The first is that St and SW quantify different physical parameters of the problem. The St number quantifies information about the amplitude and frequency of undulations relative to the swimming speed, while the SW number quantifies the wavelength of undulations relative to amplitude. The difference between these two parameters is highlighted by situations in which median/paired fin swimmers undulate their fins while their swimming speed is zero to enhance stability and maneuverability [40], the phenomenon of counterpropagating waves mentioned above. In this situation, St approaches infinity, while the fish maintains the optimal SW.
The second consideration is that the optimal value of St (0.2–0.4) recommends the frequency that maximizes the propulsive efficiency (the ratio of thrust power to the power spent by the fin) of the fin [58] without any recommendation for the wavelength of undulations. A high propulsive efficiency case could be possible with large thrust or small thrust or other values. Of all the cases that have comparable propulsive efficiency, the case with maximum thrust is desirable since it will also lead to a fast swimming speed. For aquatic animals swimming by means of elongated median/paired fins, the OSW recommends an appropriate wavelength so that the propulsive thrust is also maximized. Thus, the St and the OSW are complementary and can help identify two physical parameters of the problem—frequency and wavelength—such that a median/paired fin animal swims with high propulsive efficiency and high thrust.
Finally, the third difference between the St and the OSW arises from consideration of how these two parameters can enter into the control of advanced underwater robots that use undulating fins. The St incorporates swimming velocity, the time derivative of position, which—like orientation—is termed a group variable in mechanics. The SW only refers to variables internal to the body (amplitude and wavelength), frequently called shape variables. Often in mechanical systems there exists full control authority over shape variables, meaning any time evolution of the variables can be specified to arbitrary precision. The group variables then evolve through nonlinear dynamics involving the shape variables [65]. In the case of swimming velocity, for example, these dynamics are governed by the Navier-Stokes equation at intermediate Reynolds numbers. Control of robotic undulators for advanced underwater robots in the future may involve the relatively simple control of proportionately changing wavelength with amplitude for maximizing force production by maintaining the OSW. However, maintaining the St within a specific range would require estimating swimming velocity with a sensor or forward model, accompanied by control which accounts for the nonlinear effects that the shape variables such as frequency and amplitude have on the velocity. To the extent that the neural or mechanical control of a fin in an aquatic animal proceeds along similar lines, it may also be the case that adherence to the OSW is a relatively direct function of variables under the control of the animal.
What is the physical basis of the optimal value of the SW? The SW can be related to the peak steepness (slope) of the surface of an undulatory fin. The value of the peak slope corresponding to the OSW will depend on the fin geometry and the shape of the undulatory wave. For a 2D sinusoidal wave at the OSW, the angle made by the fin surface at the location of the peak slope is 17.44°. Our results indicate that for a given amplitude, the propulsive force is maximum at an intermediate peak steepness of the surface of the fin: very steep and very shallow are both suboptimal. We hypothesize that the optimal peak steepness may be rationalized in terms of two competing mechanisms (Fig 7). A steeper surface (smaller SW) of the fin should be able to trap and transport the fluid backward more efficiently. The increase in backward fluid momentum should lead to a stronger propulsive force at small SW. We call this the “friction mechanism" in Fig 7. A shallower surface (larger SW) of the fin leads to larger wave velocity resulting in the fluid being propelled backward at higher speed. Faster fluid velocity should increase backward fluid momentum and should lead to a stronger propulsive force at large SW. We call this the “velocity mechanism” in Fig 7. Thus, we speculate that SW influences two separate physical mechanisms with opposite trends. These competing mechanisms would lead to an optimal value of propulsive force at an intermediate value of SW.
The experiments were carried out using an updated version of the “Ghostbot” (S4 Fig, panel A) used in a number of previous studies [20,39,66]. The Ghostbot generated propulsive force using an elongated fin inspired by the ventral fin of a knifefish. The fin was 32.6 cm long and 5 cm deep. The fin was composed of 32 discrete fin rays which were driven by individual motors housed in a cylindrical body. The rays were connected by a Lycra fabric which constitutes the fin membrane. A more detailed description of the Ghostbot can be found elsewhere [20].
The axial force generated by the ribbon fin was measured similar to the methods presented in our earlier work [20]. The robot was submerged into a custom built flow tunnel and was suspended from above on a frictionless air bearing system allowing near-frictionless motion (S4 Fig, panel B). The dimensions of the working area of the tunnel were 33 cm wide, 32 cm deep, and 100 cm long. The robot was oriented horizontally and fixed in place in the water. A calibrated single axis force transducer (LSB200, Futek, Irvine, CA, USA) was placed between the air-bearing platform and mechanical ground in the axial direction of the robot so that longitudinal forces could be measured. For each trial, voltages from the force transducer were recorded at 1,000 Hz over a period of approximately 20 seconds while the fin was undulating in stationary water. The last 8 s of data were averaged. A zero measurement was also taken while the robot’s fin was at rest (not undulating) and was subtracted from the average voltage measurement from each trial. Voltage data were then converted into force units based on the calibration, which had a maximum nonlinear error of 0.034%. Details of the experiment in which free swimming velocity of the robot was measured can be found in prior work [20].
The error bars in Fig 2 show STD of error from measurements of speed and force detailed in the following. The swimming speed of the robot was measured in five consecutive trials with the same kinematic parameters. The parameters chosen for these trials were f = 3 Hz, number of undulations = 2, and θmax = 30°. The time gap between each trial was such that the flow in the tank was equal to zero before the start of the next trial. Similarly, the force measurements were also measured in five consecutive trials. In each trial the fin was actuated with f = 3 Hz, number of undulations = 2, and θmax = 30°.
Numerical simulations of the ribbon fin were carried out using an implementation of the constraint-based immersed body (IB) method (cIB) [24,25,67–70]. The cIB involves the solution of the mass and momentum conservation equations, the Navier-Stokes equations, for the combined fluid and solid domain with a constraint force in the solid domain to impose the solid motion on the fluid. The Navier-Stokes equations for the fluid flow are solved in the entire computational domain in an Eulerian framework. The motion and kinematics of the immersed body (ribbon fin in the present work) are represented in a Lagrangian frame. Information between the Lagrangian frame of the IB and Eulerian frame of the fluid is exchanged through the discrete delta function operator [71]. Further details of the numerical technique can be found in previously published work [24,25,63]. The cIB method has been incorporated within the IBAMR software [24,25]. IBAMR is an IB method implementation with support for adaptive mesh refinement and distributed memory parallelism.
Two kinds of fin simulations were carried out: free swimming and stationary. In free swimming fin simulations, the translational degree of freedom of the swimmer being simulated is free. The swimming speed of the swimmer is determined by the solution of the numerical simulation. In the stationary fin simulations, all the translational degrees of freedom of the swimmer are locked and the swimmer is held stationary while it undulates its body as it would during free swimming. These simulations are used to compute the propulsive force generated by the swimmer’s undulations. Simulations were carried out on two sets of fins: a knifefish-sized fin (10 cm × 1 cm) and a smaller rectangular fin (2 cm × 0.4 cm). The computational cost of numerical simulations for the smaller fin is lower than that of knifefish sized fin. Because of this relatively lower computational cost, the smaller fin was used in the parametric study (Fig 4). Details of the parametric study based on the smaller fin are presented in S1 Methods Section 3. The physical properties of the fluid in the simulations were for water at 25°C: Density ρ = 0.9956×103 kg/m3, viscosity μ = 0.89×10−3 N∙s/m2.
In prior work, we have approximated the shape of the traveling waves along the fins of knifefish with a traveling sinusoid [19,20,39]. The sinusoidal approximation departs from measured wave shapes in several regards [22,55], but these differences appear to have negligible impact on our results. Consequently, the kinematics of the fin undulations in the experiments and numerical simulations are described by
θ=θmaxsin2π(x/λ−ft),
(4)
where θ is the angular displacement of a fin ray located at an axial distance x from the rostral end of the fin, θmax is the maximum possible angle of excursion, λ is the wavelength, and f is the frequency of undulation. The kinematic variables of the fin are shown in S1 Fig.
The computational resources for the numerical simulations presented in this work were provided by the Quest high performance computing facility at Northwestern University. The Quest high performance cluster is composed of 252 nodes of Intel Westmere X5650 processors with 48 GB memory/node, 68 nodes of Intel Sandybridge E2670 processors with 68 GB memory/node, and 110 Intel IvyBridge E5-2680 processors with 128 GB memory/node.
S1 Methods Section 1 discusses the measurement of SW across the species included in Fig 1, as well as the calculation of mean amplitude.
|
10.1371/journal.pgen.1002529 | A Natural System of Chromosome Transfer in Yersinia pseudotuberculosis | The High Pathogenicity Island of Yersinia pseudotuberculosis IP32637 was previously shown to be horizontally transferable as part of a large chromosomal segment. We demonstrate here that at low temperature other chromosomal loci, as well as a non-mobilizable plasmid (pUC4K), are also transferable. This transfer, designated GDT4 (Generalized DNA Transfer at 4°C), required the presence of an IP32637 endogenous plasmid (pGDT4) that carries several mobile genetic elements and a conjugation machinery. We established that cure of this plasmid or inactivation of its sex pilus fully abrogates this process. Analysis of the mobilized pUC4K recovered from transconjugants revealed the insertion of one of the pGDT4–borne ISs, designated ISYps1, at different sites on the transferred plasmid molecules. This IS belongs to the IS6 family, which moves by replicative transposition, and thus could drive the formation of cointegrates between pGDT4 and the host chromosome and could mediate the transfer of chromosomal regions in an Hfr-like manner. In support of this model, we show that a suicide plasmid carrying ISYps1 is able to integrate itself, flanked by ISYps1 copies, at multiple locations into the Escherichia coli chromosome. Furthermore, we demonstrate the formation of RecA-independent cointegrates between the ISYps1-harboring plasmid and an ISYps1-free replicon, leading to the passive transfer of the non-conjugative plasmid. We thus demonstrate here a natural mechanism of horizontal gene exchange, which is less constrained and more powerful than the classical Hfr mechanism, as it only requires the presence of an IS6-type element on a conjugative replicon to drive the horizontal transfer of any large block of plasmid or chromosomal DNA. This natural mechanism of chromosome transfer, which occurs under conditions mimicking those found in the environment, may thus play a significant role in bacterial evolution, pathogenesis, and adaptation to new ecological niches.
| All living species have the capacity to evolve in order to adapt to new and often hostile conditions. Horizontal gene transfer is a major route for rapid bacterial evolution. Some clearly identified mobile genetic elements (plasmids, phages, etc.) are by essence exchanged between bacteria. However, the mechanisms generating the bacterial core genome diversity are much less understood. In this study we have characterized in Y. pseudotuberculosis, a natural bacterial pathogen causing mesenteric lymphadenitis and enteritis, a mechanism of horizontal gene exchange that conveys the transfer of virtually any piece of chromosomal or plasmid DNA to a new bacterial host. This generalized mechanism of DNA transfer is optimal when the bacteria encounter conditions that might resemble those they met in their natural ecological niches. We demonstrate that this transfer mechanism is extremely powerful, as the presence on a conjugative replicon of an insertion sequence having a low specificity of insertion and transposing through replicative transposition is sufficient to drive the horizontal transfer of virtually any piece of chromosomal or episomal DNA. As such, this mechanism is much less constrained than the classical Hfr mechanism described in laboratory E. coli and could be used by a wide variety of bacterial species for gene exchange and evolution.
| Horizontal gene transfer (HGT) is a driving force for bacterial evolution, as it allows the dispersion of adaptive loci between closely related and also phylogenetically distant bacterial species. Well-characterized mobile genetic elements such as conjugative plasmids, transposons, Integrative conjugative elements (ICE), pathogenicity islands (PAI), or phages are associated with HGT of specific adaptive functions (antibiotic resistance, virulence, metabolic pathways) and participate to genome plasticity. However, exchanges of chromosomal regions that form the core genome and are not part of the mobile genetic pool are also observed. While their importance in bacterial evolution and speciation is now well established, the underlying mechanisms are often loosely described and remain hypothetical in many cases.
The Gram-negative enteropathogen Yersinia pseudotuberculosis carries a PAI termed High Pathogenicity Island (HPI) [1], which encodes the siderophore yersiniabactin [2]. The fact that this island is mobile within the genome of its host strain [3], and is present and often conserved both in terms of genetic organization and nucleotide sequence in various bacterial genera such as Escherichia coli (various pathotypes), Klebsiella or Citrobacter [4], suggested that it may have retained its ability to be horizontally transmitted to new bacterial hosts. Indeed, we evidenced the transfer of the HPI between natural Y. pseudotuberculosis isolates [3]. This phenomenon was observed only when the bacteria were incubated at low temperature (optimal at 4°C) and in broth, and was more efficient in an iron-poor medium [5]. However, this transfer did not require the integration/excision machinery encoded by the HPI, was RecA-dependent in the recipient strain, and involved not only the HPI but also adjacent sequences encompassing at least 46 kb of chromosomal DNA [3]. Similar results were recently obtained for the HPI of natural Escherichia coli isolates, using a multi locus sequence typing approach. The E. coli HPI was found to have been acquired simultaneously with the chromosomal flanking regions of the donor strains [6], indicating again that the island was transmitted as part of a larger chromosomal region. This phenomenon is not restricted to the HPI and to enterobacteria since it has been recently reported that movement of the Enterococcus faecalis PAI was invariably accompanied by transfer of flanking donor chromosome sequences [7].
The aim of this work was to characterize the mechanisms underlying horizontal chromosomal gene transfer in Y. pseudotuberculosis. We describe here a natural system of conjugative transfer, which may be used by a wide variety of bacterial species for gene exchanges, and which may represent a driving force for bacterial evolution.
Since we did not know whether the lateral transfer process previously observed was limited to the region encompassing the HPI or could involve any portion of the chromosome, two other loci (ureB and or5076) were labeled with a spectinomycin (Spe) and trimethoprim (Tmp) resistance cassette, respectively. These two genes were chosen because, based on the IP32953 sequence, they are predicted to be separated from each other and from the HPI (tagged with a kanamycin (Kan) cassette in the irp2 gene) by at least 1.5 Mb of chromosomal DNA (Figure S1). Moreover, the ureB gene, which is part of the urease locus, and or5076, encoding a putative toxin transporter [8] are not predicted to be involved in DNA transfer. After co-incubation of the donor 637-irp2K-ureBS-5076T and recipient 637ΔHPI-NalR strains (Table 1) under conditions (4 days at 4°C in LB-αα' with shaking) that we previously found to be optimal for HPI transfer [3], recipient strains having acquired either the irp2K (NalR, KanR, RifS), ureBS (NalR, SpeR, RifS) or or5076T (NalR, TmpR, RifS) antibiotic resistances were obtained. Acquisitions of the corresponding tagged loci were checked by PCR (Figure S1). Transfer frequencies were of the same magnitude for the three antibiotic-tagged loci (≈10−8, Figure 1). None of the transconjugants obtained had simultaneously acquired two of the antibiotic-tagged loci, indicating that the sizes of the chromosomal fragments transferred were inferior to 1.5 Mb.
We previously showed that horizontal transfer of the HPI occurs only at low temperatures [3]. The same temperature dependency was observed for ureBS and or5076T: transfer of each of the three antibiotic-tagged loci was detected only when the donor and recipient strains were co-incubated at temperatures below 20°C (Figure 1), and was more efficient at 4°C than at 12°C (≥13 fold higher), as previously observed for irp2K. Therefore, distantly located chromosomal loci can be transferred with similar efficiencies and temperature regulations.
Whether this transfer mechanism could also mediate horizontal transmission of episomal molecules was addressed by introducing the non-conjugative and non-mobilizable plasmid pUC4K (KanR) into the donor 637-RifR. Using the defined optimal growth conditions, transfer of pUC4K from the 637(pUC4K) to the recipient 637ΔHPI-NalR was obtained and confirmed by PCR with primers 210A/210B (Table S1). This transfer occurred at a frequency of 2.3 (±0.4)×10−7, which is at least 10 times higher than that of chromosomal loci. Therefore, the process of DNA transfer is not limited to chromosomal DNA but can also involve plasmid molecules.
Altogether our results demonstrate the existence of a mechanism that potentially allows transfer of any chromosomal or episomal DNA molecule at low temperature. This mechanism was thus named GDT4 (for Generalized DNA Transfer at 4°C).
The capacity of other Y. pseudotuberculosis strains to mediate GDT4 was studied by tagging the IP32953 and IP32777 strains with both a Kan and Spe cassettes inserted into the irp2 and ureB loci, respectively (Table 1). When these two recombinant strains were used as donors, no IP32637 transconjugants having acquired either irp2K or ureBS were obtained, indicating that GDT4 is not a property common to the entire Y. pseudotuberculosis species.
Strain IP32637 has the peculiarity of harboring an extra high molecular weight (≥100 kb) plasmid [9]. The role of this additional plasmid in chromosomal transfer was assessed by comparing GDT4 in IP32637 and its IP32637c plasmid-cured derivative [9]. Two tagged donor strains, 637c-irp2K and 637c-ureBS (Table 1), were generated and co-incubated with the 953-NalR recipient. No transconjugants were obtained, indicating a role of this plasmid in DNA transfer. The high molecular weight plasmid was thus designated pGDT4.
pGDT4 does not appear to be ubiquitous in the species Y. pseudotuberculosis as the genome sequences of IP32953 and of other Y. pseudotuberculosis strains available in databases did not evidence the presence of this plasmid. To get an insight into the frequency of pGDT4 carriage in this species, a 4 kb HindIII fragment of this episome, designated pGDT4.seq was cloned into pUC18, sequenced, and used to design primers (358A/B) for PCR screening. The analysis of a panel of 39 Y. pseudotuberculosis strains of serotypes I to V (Table S2) for the presence of the pGDT4 sequence identified two isolates (IP32699 and IP30215) that gave a PCR product of the expected size (Table S2). Both strains contained high molecular weight episomes whose HindIII-digestion patterns yielded some restriction fragments with a size similar to those of pGDT4, but the overall profiles of the three episomes were different (data not shown). Therefore, the plasmids found in IP32699 and IP30215 probably share some regions with pGDT4, but they are not identical to this plasmid. Since Yersinia pestis is a recent descent of Y. pseudotuberculosis [10], we also screened by PCR a panel of 51 strains of Y. pestis belonging to the three classical biovars (Antiqua, Medievalis and Orientalis) for the presence of the pGDT4-borne sequence. None of the strains tested yielded an amplification product (Table S2), suggesting the absence of vertical or horizontal transmission of pGDT4 to Y. pestis.
Plasmid analysis of transconjugants resulting from the co-incubation of the 637-irp2K-ureBS donor strain with the 637c-NalR recipient revealed that about half of them had acquired pGDT4 together with the chromosomal irp2 (8/20) or ureB (11/20) tagged region, thus indicating that pGDT4 is also transferable.
To further study pGDT4 transfer capacity, the plasmid was labeled by allelic exchange of the pGDT4-4kb segment with a Tmp cassette. When the 637-irp2K-ureBS donor strain carrying the tagged pGDT4T was co-incubated with the 637-NalR recipient, transconjugants harboring pGDT4T were obtained with a frequency approximately 103 times higher than that of chromosomal genes (Figure 2A and Table S3). The transfer frequency of pGDT4T increased 200 fold when 953-NalR instead of 637-NalR was used as a recipient (Figure 2B and Table S3), indicating that properties inherent to the recipient cells may influence their capacity to take up pGDT4. The difference in the ability of the two strains to acquire this plasmid could not be explained by a mechanism of surface exclusion, as the frequency of transfer of pGDT4T to IP32637 harboring or not harboring a resident pGDT4 was similar (Figure 2B and Table S3). Some unidentified intrinsic properties of the recipients such as a difference in their restriction/modification systems may be responsible for this difference. As observed with chromosomal DNA, no transfer of pGDT4T was detected when the bacteria were mated at temperatures ≥28°C (Figure 2C and Table S3). However, in contrast to chromosomal DNA [3], pGDT4T was transferable on nitrocellulose filters, at frequencies similar to those observed in a liquid medium (Table S3), but again, only at low temperature (Figure 2C). Therefore, transfer of pGDT4 is also temperature-dependent, but in contrast to chromosomal genes, it occurs at much higher frequencies, and both in liquid and on solid media.
Since the presence of pGDT4 is required for transfer, we wondered whether its presence could confer GDT4 properties to a strain that is naturally unable to mediate chromosomal transfer. For this purpose, a 953-ureBS transconjugant that had acquired pGDT4 simultaneously with chromosomal genes was used as donor and co-incubated with a 637ΔHPI-RifR recipient. While the parental 953-NalR strain was unable to transfer chromosomal DNA, the 953-ureBS(pGDT4) transconjugant gained the capacity to retransfer the acquired ureBS locus, though with a frequency 10 times lower (10−9) than that observed during the first transfer. These results further point at pGDT4 as a key element in the mechanism of chromosomal transfer.
To determine whether pGDT4 could encode its own transfer machinery, the plasmid was sequenced (EMBL accession number FM178282). The schematic map of the 94,967 bp circular plasmid molecule is shown on Figure 3. Of the 102 predicted coding sequences (cds) identified on pGDT4, 74 had homologs in databases (Table S4). Four major functional groups of genes were delineated on pGDT4:
Since pGDT4 carries a large set of genes predicted to be involved in conjugative transfer, we wondered whether this pGDT4-specific mobility function mediates GDT4. To investigate this potential role, the Mpf function was inactivated by allelic exchange of a large portion of the pil region (from pilL to pilV) of pGDT4 with a Tmp cassette in 637(pUC4K). After co-incubation of the resulting strain 637(pUC4K, pGDT4Δpil) with the 953-NalR recipient, no transconjugants having acquired pUC4K were obtained, indicating that the pilus-encoding region of pGDT4 is required for generalized DNA transfer.
As the Mpf region is predicted to encode a conjugative machinery, GDT4 most likely occurs by a mechanism of conjugation. To rule out other possible mechanisms of transfer, DNAse was added to the medium during the co-incubation period. The transfer frequency of the irp2 locus from the donor 637-irp2K-ureBS to the recipient 637ΔHPI-NalR was not affected, arguing against an acquisition of naked DNA through a transformation process. Cell-free filtrates of the supernatant of the donor strain incubated with the recipient strain did not allow DNA transfer, suggesting the absence of transferable DNA released from the bacteria but protected from the action of a DNAse (inside phage particles or membrane vesicles). These results argue against a transfer of DNA by transformation or transduction and further point at conjugation as the most likely mechanism. However, since this conjugative process was observed in liquid medium under agitation, we wondered whether a strong shaking of the culture would disrupt the pilus-mediated interactions between bacterial cells, and therefore decrease the transfer frequency. Surprisingly, when we increased the agitation of the medium containing the donor and recipient cells to 130 rpm (which was vigorous under our experimental conditions), the frequency of transfer of the irp2 locus was not affected (0.9×10−8). Electron microscopy analysis of IP32637 cells grown under conditions optimal for GDT4 did not reveal any pilus structures on the bacterial surface. In contrast, tightly aggregated bacilli that seemed to be connected by “bridges” were observed (Figure 4).
We noted that after pUC4K transfer, the plasmid sizes of pUC4K in 10 different transconjugants were variable. This was confirmed after digestion of the 10 plasmids with NdeI, an enzyme that has a single restriction site in pUC4K. Three plasmids (rpUC4K-1 to -3) had the expected pUC4K size, while the seven others (rpUC4K-4 to -10) had a size superior to that of the original molecule (data not shown), indicating that various types of rearrangements had occurred during plasmid transfer. Remarkably, a search for the potential transposition of pGDT4-borne IS (ISYps1, ISL3, ISYps2, ISNCY or ISYps3) on rpUC4K molecules by PCR (primers described in Table S1) showed that all seven larger size recombinant plasmids (rpUC4K-4 to -10) harbored ISYps1.
rpUC4K-5 to -9 had a size compatible with the acquisition of a single ISYps1 copy. Digestion with XhoI, an enzyme that cuts once in pUC4K and once in ISYps1, yielded two restriction fragments, thus confirming the presence of a single ISYps1 copy. However, two distinct restriction profiles were observed, one for rpUC4K-5 and -8 and one for rpUC4K-6, -7 and -9 (data not shown), indicating the occurrence of different genetic rearrangements. Sequencing of the regions encompassing the ISYps1 insertion site in one recombinant plasmid of each group (rpUC4K-5 and rpUC4K-6) demonstrated that the IS was inserted at two different sites, approximately 100 bp apart, and in opposite orientation (Figure 5a and 5b). The insertion generated an 8 bp duplication of the target sequence: AAAATAGG in rpUC4K-5 and TATTTGAA in rpUC4K-6.
rpUC4K-4 had a size superior to that of the above five plasmids. XhoI digestion revealed the presence of two ISYps1 copies on this plasmid (Figure 5c). To determine whether the region located between these two ISYps1 copies on rpUC4K-4 corresponded to a portion of pGDT4, a PCR amplification of three pGDT4 genes (the Ig-like domain, parF and traM), each located between two different ISYps1 copies on pGDT4 (Figure 5) was performed. No positive signal was detected, suggesting that the region located between the two ISYps1 is a duplicated portion of pUC4K (Figure 5c).
The rpUC4K-10 plasmid was different from all others since the PCR analysis showed that it carries all five pGDT4-borne IS, as well as the Ig-like domain and parF genes (but not traM, Figure 5d). rpUC4K-10 has thus most likely acquired the entire pGDT4 sequence located between the two ISYps3 transposons (Figure 5).
Altogether, these results show that most pUC4K transfers generated a variety of genetic modifications that were systematically accompanied by the transposition of the pGDT4-borne ISYps1 element.
ISYps1 belongs to the IS6 family, known to transpose through replicative transposition. This mode of transposition gives rise exclusively to replicon fusions (cointegrates), in which the donor and target replicons are separated by two IS copies in direct orientation. The cointegrate can be subsequently resolved by recombination between the two IS copies [17]. To determine whether ISYps1 transposes through this mechanism, this IS was cloned into the suicide mobilizable vector pSW23T and introduced into a replication-permissive E. coli strain, yielding ω7249(pSWYps1.1) (Table 1). After mating of this donor strain with ω4826, a non-replication permissive recA- recipient, ω4826::pSWYps1.1 transconjugants resulting from pSWYps1.1 integration into the recipient chromosome were obtained with a frequency of 8.5(±0.4)×10−6 (which corresponds to the frequencies of both conjugation and transposition, Table S5). Since the frequency of conjugation under these conditions was found to be 3.4(±0.9)×10−3 (Table S5), the transposition frequency of ISYps1 is thus approximately 2×10−3. The genomic DNA of eight independent ω4826::pSWYps1.1 colonies were digested with HindIII (which cuts once in pSW23T and not in ISYps1), and hybridized with an ISYps1 probe. All eight clones harbored, as expected, two integrated copies of ISYps1 (Figure S2). Of note, all clones exhibited different hybridization profiles.
To further determine whether association between ISYps1 and a conjugative plasmid allows cointegrate transfer, the IS was cloned on a non-mobilizable vector (pSW23) and introduced into an E. coli strain carrying the conjugative plasmid R388 (Table 1). After mating of the resulting pi3(R388, pSWYps1.2) donor strain with the ω4826 recipient that cannot sustain pSWYps1.2 replication, ω4826(R388::pSWYps1.2) transconjugants were independently selected on Cm (pSWYps1.2 tag) and Tmp (R388 tag) MH agar plates. CmR clones were found at a frequency of 9(±3)×10−5 (Table S5) and were all TmpR, while in the absence of an ISYps1 carried on the pSW23, no CmR transconjugants were obtained (Table S5). This further demonstrates that Yps1 drives the formation of cointegrates that can be subsequently transferred by conjugation. Under these conditions, R388 was transferred at a frequency of 2.2(±0.7)×10−1 (Table S5), indicating that transfer of these cointegrates occurs at high frequencies (≈2×10−5). To further characterize these events, the plasmid profiles of five independent R388::pSWYps1.2 cointegrates were analyzed after restriction with MfeI (10 sites in R388, one in pSWYps1.2). All five pSWYps1.2 insertions were in different locations on R388 (data not shown). The two MfeI junction fragments from one of these R388::pSWYps1.2 cointegrates were cloned into the EcoRI site of pUC18 and the precise cointegrate location was determined by sequencing. Transposition of pSWYps1.2 occurred in the orf5 cassette of the R388 integron [18] and led, as for the two pUC4K insertion events analyzed above, to an 8 bp duplication. The duplicated sequence (GATCCGAG) was different from the other two, further indicating the absence of a specific integration site.
Our results thus demonstrate that ISYps1 is able to transpose into a variety of insertion sites by replicative transposition through cointegrate formation, mediating the transfer of potentially any piece of non-mobilizable DNA molecule.
We have evidenced a mechanism of HGT that convey the conjugative transfer or virtually any piece of chromosomal or plasmid DNA in a natural isolate of Y. pseudotuberculosis. This mechanism shares some characteristics with those previously described, but has several novel and unique properties.
GDT4 is not observed at temperatures ≥20°C and its efficiency increases as the temperature decreases. Although some conjugative plasmids have been previously shown to be self-transferable at 14°C but not at 37°C [19], [20], to our knowledge no plasmid able to conjugate at 4°C has ever been described. Temperature-dependent plasmid transfer is primarily mediated by H-NS and Hha proteins, which can be both plasmid and/or chromosome encoded [20]. The pGDT4 sequence did not reveal any gene encoding such proteins, but it is known that chromosomally encoded Hha and YmoA (equivalent to H–NS) act as thermoregulators in Y. enterocolitica [21]. These proteins (also encoded by the Y. pseudotuberculosis genome [8]) may thus modulate pGDT4 transfer at cold temperatures. Interestingly, H-NS is an integral part of bacterial stress response pathways and its function is known to be sensitive to changes in environmental conditions such as temperature [22], [23]. A cold stress could thus be a signal for the bacteria to transfer their genetic material by GDT4. The low temperature and an iron-poor liquid environment may also induce changes in the bacterial membrane structure, as observed for the closely related organism Y. pestis, in which the transcription patterns of various genes encoding components of the bacterial membrane were modified during iron starvation [24] or growth at 10°C [25]. These modifications might facilitate the formation of pores through which chromosomal DNA could translocate. Indeed, large cell aggregates in which bacteria appeared to be connected by bridges were observed. Similar tight bacterial contacts, designated conjugative junctions [26] or conjugational junctions [27] have been observed during RP4 or F-mediated mating of E. coli, respectively. However, these physical properties do not seem to be pGDT4-mediated, as we also observed large bacterial aggregates and possibly intercellular channels with IP32953, a strain that does not harbor this plasmid (data not shown). These bacterial aggregates have some similarities with biofilms in which bacteria are also closely connected. Biofilm formation occurs under natural conditions in a variety of bacterial species [28], including Y. pseudotuberculosis [29]. It could thus be hypothesized that GDT4 may take place between bacteria residing within biofilms in their natural ecological niches. This observation also suggests that acquisition of new functions, including virulence factors, by Y. pseudotuberculosis takes place in the environment rather than in a mammalian host.
Another characteristic feature of GDT4 is that transfer of chromosomal DNA occurs only in a liquid medium. Like other conjugative processes, GDT4 requires a pilus-like mating system and a mating channel to occur, as demonstrated by the fact that inactivation of the pGDT4-borne pilus apparatus abolished this mechanism. While some plasmids transfer better on plates, some others encoding long flexible pili allow DNA transfer efficiencies of the same magnitude in liquid and on solid media [30], [31], and this applied to pGDT4. In contrast, the absence of transfer of chromosomal DNA on agar was unexpected. Also unexpected was the fact that the efficiency of transfer in broth was not affected by a strong agitation, as opposed to recent findings showing that a vigorous shaking negatively affected the transfer of several conjugative plasmids, including the F' plasmid that encodes long and flexible pili [32]. Actually, growth in a liquid medium at low temperature could create the conditions optimal for GDT4. Indeed, this environment might be more favorable for the formation of tight bacterial aggregates and inter-cellular bridges through which long stretches of chromosomal DNA could transit.
pGDT4 also triggered the conjugative transfer of the non-mobilizable plasmid pUC4K. Similarly, transfer of the non mobilizable plasmid pBR325 by an RP4::miniMu mobilizing plasmid was previously observed [33], but the mechanism underlying this genetic transfer was not characterized. cis-mobilization of non-mobilizable plasmid DNA can occur after integration of a conjugative plasmid into the genetic element to be transferred. Integration arises either by homologous recombination between identical elements, often two copies of the same IS located on each DNA molecule (as for the Hfr formation in E. coli [34]), or through the formation of cointegrates mediated by specific transposons or ISs [35]. Integration of pGDT4 into pUC4K could not occur via homologous recombination as the two replicons do not share any common IS or identical DNA sequences. However, pGDT4 carries ISYps1, an IS which is predicted to belong to the IS6 family (http://www-is.biotoul.fr/is.html). ISYps1 is the second IS of the IS6-type identified in the genus Yersinia [36]. Members of this family have the capacity to create cointegrates by replicon fusion in the absence of a homologous IS on the target DNA [37]. Furthermore Tn3, which also moves by replicative transposition, has been found to mediate cis-mobilization of non mobilizable plasmids by this mechanism [35]. The fact that several rpUC4K plasmids obtained after pGDT4-mediated transfer carried a copy of ISYps1 argues for a role of this IS in pGDT4 integration into its target. We have demonstrated that ISYps1 is indeed transposing through replicative transposition. ISYps1 has a low specificity of recognition of the target sequence, as attested by our observation that the three insertion sites sequenced (two on pUC4K and one on R388) were different. This model also predicts the formation of cointegrates carrying two copies of the IS element, each flanking the sequence of the donor plasmid. We do have observed the formation of cointegrates between R388 and pSWYps1 flanked by the expected ISYps1 copies. A resolution step, which occurs through homologous recombination between the two IS copies, is then required to separate the donor and target replicons, leaving a single IS copy in the target and restoring the donor plasmid. The rpUC4K-5 to -9 molecules that were found to carry one ISYps1 copy are most likely the results of such a resolution event. The presence on one recombinant plasmid (rpUC4K-10) of a portion of pGDT4 carrying ISYps1, ISYps2, ISYps3, ISNCY and ISL3 indicates that additional, complex rearrangements involving the Tn3-like transposon ISYps3 can also occur. Finally, the existence in some transconjugants of pUC4K plasmids with a size identical to that of the original molecule could be the result of the resolution of cointegrates containing pUC4K concatemers.
GDT4 thus represents a remarkable illustration and validation of the model of Tn3-mediated transmission of non conjugative plasmids proposed by Crisona et al. in the 1980's [35]. Following this model, the first step is the integration of pGDT4 into its target DNA by ISYps1-mediated replicon fusion during plasmid replication (Figure 6). As mentioned above, this generates a cointegrate which carries the two replicons separated on each side by an ISYps1 copy. This cointegrate then uses the conjugative machinery encoded by pGDT4 to promote its transfer to the recipient strain. The final step is the resolution of the cointegrate by homologous recombination between two ISYps1 copies or any other duplicated sequence present in the cointegrate. Since pGDT4 carries several ISYps1, the resolved molecules have different sizes and DNA composition.
GDT4 is also able to mediate the translocation of chromosomal DNA, most likely by integration into the bacterial chromosome and transfer in a Hfr-like manner. The Hfr mechanism is one of the earliest and best described examples of chromosomal transfer and is mediated by the F plasmid of E. coli [38]–[40]. F integrates stably into the E. coli chromosome through homologous recombination between IS copies present on both the F plasmid and the bacterial chromosome [38]–[42] to create Hfr strains, with transfer origins located at different chromosomal loci [43], [44]. In contrast to the classical Hfr mechanism, integration of pGDT4 into the chromosome probably occurs, as in pUC4K, via the ISYps1-mediated replication fusion mechanism. At least three pieces of evidence support this hypothesis: (i) no IS element is shared by pGDT4 and the IP32637 chromosome, in contrast to what is expected for the Hfr mechanism, (ii) in Y. pseudotuberculosis, three distantly located chromosomal loci (irp2K, ureBS and or5076T) were transferred with similar frequencies, and (iii) in all eight E. coli transconjugants analyzed, pSWYps1 was inserted at different sites on the chromosome. ISYps1 thus appears to have a very low specificity of recognition, allowing its insertion at multiple sites on bacterial plasmids and chromosomes. After mobilization of the chromosomal fragment adjacent to the pGDT4 integration site and transfer to a recipient strain, following the Hfr-type transfer model, homologous recombination between the incoming DNA and the chromosome is expected to take place, leaving no trace of pGDT4 in the chromosome of the transconjugant. Our previous observation that RecA activity is necessary in the recipient, but not in the donor strain for chromosomal transfer [3], and the results of the present study showing that pGDT4 is absent from some transconjugants that have acquired chromosomal genes further support this model of horizontal transfer. Our study thus validates the model proposed by Willets et al. in the 1980's for the mobilization of the E. coli chromosome via the formation of a cointegrate with the R68.45 plasmid during IS21 transposition [45]. Such cointegrate formations were widely used at that time to establish the genetic map of various bacterial species (see for instance [46], [47]). Most importantly our results show, without the need for heterologous plasmids like RP4 or R68.45, that this type of chromosomal conjugative transfer may occur under natural conditions in wild type bacterial pathogens carrying endogenous plasmids.
The capacity of wild type bacteria to naturally transfer large pieces of chromosomal DNA following the typical Hfr mechanism of homologous recombination between identical IS copies on the chromosome and the plasmids has been documented in a variety of bacteria, including extremophiles [48], Gram-positive cocci [49], and actinomycetes [50]. What we describe here is certainly a less constrained and more powerful mechanism, as it only requires the presence of an IS of the IS6 family on a conjugative replicon to generate cointegrates able to drive the horizontal transfer of any piece of DNA (chromosomal or episomal). It is remarkable that a high density of IS is commonly observed on plasmids. For instance the Shigella plasmid pWR100 carries 93 copies of complete or truncated IS belonging to 21 different types [51]. Thus, more than being IS depository, this location may reflect the broad selective advantage brought by plasmid/IS associations as a chromosomal transfer device. Such a ‘genetic symbiosis’, offers a means for the natural transfer of large blocks of genes conferring new metabolic properties or virulence functions. According to our model, GDT4 does not leave any signature in the recipient genome in most instances, and therefore its contribution to the numerous horizontal gene exchanges that shape bacterial genomes can hardly be quantified. However, according to the ISfinder database, approximately 5% of the known IS belong to the IS6 and Tn3 families, which use a replicative transposition mechanism. As they are found in all bacterial and archaeal phyla, the mechanism we describe here might be responsible for a substantial fraction of gene exchanges occurring among bacterial species.
Remarkably, this mechanism of DNA transfer was optimal when the bacteria were grown under conditions (low temperatures, iron poor medium, biofilm-like bacterial aggregates) that might be close to those met by these microorganisms in their normal ecological niches. This natural GDT mechanism may thus play a significant role in bacterial evolution, genetic polymorphism, pathogenesis and adaptation to new environmental conditions.
Bacterial strains used in this study are listed in Table 1 and Table S1. Wild type strains were taken from the collection of the Yersinia Research Unit (Institut Pasteur). Bacteria were grown in LB (Luria Bertani) or MH (Mueller Hinton) medium for 24 h at 28°C (Yersinia) or 37°C (E. coli) with agitation, or for 48 h on LB or MH agar plates. When necessary, kanamycin (Kan: 100 µg ml−1), rifampicin (Rif: 100 µg ml−1), nalidixic acid (Nal: 25 µg ml−1), spectinomycin (Spe: 50 µg ml−1), tetracycline (Tc: 15 µg ml−1), chloramphenicol (Cm: 25 µg ml−1), trimethoprim (Tmp: 20 µg ml−1), thymidine (dT: 0.3 mM) or the iron chelator αα'-dipyridyl (0.2 mM, Sigma) were added to the medium.
Spe (aadA) or Tmp (dfr) non-polar cassettes were PCR-amplified using primers described in Table S1, and pSW25 [52] or pGP704N-dfr [3] as templates, respectively. All allelic exchanges of chromosomal or plasmid genes by an antibiotic resistance cassette were done following the LFHR-PCR procedure [53]. The Spe and Tmp cassettes were introduced into the chromosomal ureB and or5076 genes, respectively, using primers that amplify upstream and downstream fragments of ureB and or5076, as shown on Figure S1 and Table S1. To label pGDT4, the plasmid was digested with HindIII and a 4 kb fragment (pGDT4-4kb) was purified and cloned into pUC18. Approximately 600 bp of each extremity of the cloned fragment were sequenced. These sequences were then used to design primers (358A/B and 359A/B, Table S1) that served for allelic exchange between the Tmp cassette and the target region of pGDT4 in strain 637-irp2K-ureBS. Correct insertion of the Tmp cassette was confirmed by PCR using primer pair 358A/359B. Mutagenesis of the pil region was done by replacing the pGDT4 region extending from pilL (pGDT4_0086) to pilV (pGDT4_0097) by a Tmp cassette, using primer pairs 773A/B and 774A/B (Table S1). The various antibiotic-tagged derivatives cured of pKOBEG-sacB were selected on sucrose plates.
Optimal conditions for chromosomal DNA transfer in Y. pseudotuberculosis have been previously described [3]. Briefly, the donor strain (usually RifR) harboring chromosomal loci labeled with antibiotic cassettes and the recipient strain (usually NalR) were grown overnight in LB at 28°C with agitation. Equal amounts (5×106) of donor and recipient cells were mixed in 25 ml of LB-αα' and grown at 4°C with mild rotary agitation (80 rpm) for 4 days. Donor and recipient bacteria were quantified on Rif and Nal plates, respectively, and transconjugants were selected on Nal plates containing the appropriate antibiotic. To ensure that the colonies were not spontaneous NalR mutants of the RifR recipient strain, the Rif susceptibility of the transconjugants was systematically checked. For every single DNA transfer experiment, 10 to 20 transconjugant colonies were analyzed by PCR for the acquisition of the corresponding antibiotic-tagged locus with primer pairs 233B/166, 92A/322B and 348B/346A (Table S1) as indicated on Figure S1. When the transfer of the irp2K locus was analyzed, the acquisition of the entire HPI by the recipient strain was further checked with primer pairs A10/144A and A9/143B (Figure S1 and Table S1). The frequency of DNA transfer was determined as the number of NalR (or KanR, SpeR, TmpR) RifS transconjugants per RifR donor cells. To determine whether free DNA molecules in the medium could mediate GDT4, the donor bacteria 637-irp2K-ureBS and the recipient 637ΔHPI-NalR were co-incubated in the presence of 100 U/ml of DNAse in the culture medium. The activity of the DNAse under these conditions was checked by adding 1 ug/ml of bacterial DNA to the culture medium and by observing that the added DNA was degraded.
Transfer of pGDT4T was studied after incubation of the donor (637-irp2K-ureBS(pGDT4T)) and various NalR recipient cells for four days at 4, 28 or 37°C in liquid or solid media. On solid medium, 2×108 donor and recipient cells were mixed on a 0.45 µm nitrocellulose filter (Millipore) and at the end of the incubation period, the bacterial mixture was suspended in 1 ml of MH. Donor and recipient cells were quantified on MH-Rif and MH-Nal plates, respectively. Transconjugants having acquired pGDT4T were identified as NalR/TmpR/RifS colonies. In each transfer experiment, 10 transconjugants were analyzed by PCR for the presence of pGDT4T with primer pair 358A/346B (Table S1). Finally, the pGDT4T transfer frequency was calculated as the number of NalR/TmpR/RifS transconjugants per donor cells.
One transconjugant resulting from the co-incubation of the 637-irp2K-ureBS donor strain with the 637c-NalR recipient was used to obtain a plasmid extract which contained only pGDT4. Sequencing was performed using the whole genome shotgun strategy [54]. A 2–3 kb insert library was generated by random mechanical shearing of pGDT4 DNA and cloning into pcDNA-2.1 (Invitrogen). Recombinant plasmids were used as templates for cycle sequencing reactions consisting in 35 cycles (96°C for 30 s; 50°C for 15 s; 60°C for 4 min) in a thermocycler, using the Big dye terminator kit (V3.1, Applied Biosystems). Samples were precipitated and loaded onto a 96-lane capillary automatic 3700 DNA sequencer (Applied Biosystems). In an initial step, 1000 sequences from the library were assembled into 5 contigs using the Phred/Phrap/Consed software [55], [56] (8-fold sequence coverage). Consed was used to predict links between contigs. PCR products amplified from the pGDT4 template were used to fill gaps and to re-sequence low quality regions using primers designed by Consed. Physical gaps were closed using combinatorial PCR. The correctness of the assembly was confirmed by ensuring that the deduced restriction map was identical to the one obtained experimentally. The traX and traY genes fusion into a single traXY gene was checked by re-sequencing this locus on the original pGDT4 DNA preparation.
ISYps1, ISYps2 and ISYps3 designation were attributed by the ISfinder database (http://www-is.biotoul.fr/). The nucleotide sequence of pGDT4 has been submitted to the EMBL database under accession number FM178282. Details and properties of the different ISYps characterized in this work are accessible through the ISFinder web site.
Bacteria were negatively stained with 2% uranyl acetate onto glow discharged copper grids. The samples were observed in a Jeol 1200EXII and/or a JEM 1010 (Jeol) equipped with a Keenview camera (Eloise) at 80-kV accelerating voltage. Images were recorded with an Analysis Pro Software version 3.1 (Eloise).
The sequence corresponding to the ISYps1 copy carried on rpUC4K-6, flanked by its 113 bp upstream and 138 bp downstream regions, was amplified using primers 1039/1040 and cloned as an EcoRI-BamHI insert into the suicide mobilizable vector pSW23T [52], giving rise to pSWYps1.1. This plasmid was then introduced into E. coli ω7249 [57], a strain allowing pSW23T replication and conjugative transfer. Conjugation between this donor strain and E. coli ω4826 was performed as previously described [57], The frequency of conjugation–transposition frequency was calculated as the number of CmR transconjugants (ω4826::pSWYps1.1) per total number of recipients (TcR). The conjugation frequency was established in parallel by conjugation from the same donor ω7249(pSWYps1.1) to a ω4826 pir+ derivative (obtained through transformation with plasmid pSU38Δpir which expresses pir [52]). The frequency of illegitimate recombination of the pSW23T which can lead to CmR transconjugants was established by conjugation between donor ω7249(pSW23T) and ω4826, and found to be 4.6(±1.7)×10−8. Genomic DNA from 8 independent ω4826::pSWYps1.1 colonies were extracted using QIAGEN Genomic Tips and buffer set, digested with HindIII, and hybridized with a probe internal to ISYps1 (generated by PCR amplification with primers 1041/1044 and labeled with α-32P dCTP, using the Random Primed labeling kit (Roche)).
The EcoRI-BamHI fragment carrying ISYps1 was transferred from pSWYps1.1 to the non-mobilizable version of pSW23 [52], giving rise to pSWYps1.2. This plasmid was then introduced into the E. coli pi3 pir+ strain that harbors the IncW conjugative plasmid R388, which does not carry any IS (GenBank BR000038), giving rise to pi3(R388, pSWYps1.2). Conjugation of this donor strain with ω4826 yielded ω4826(R388::pSWYps1.2). The frequency of cointegrate formation after mating was calculated as the number of CmR transconjugants per total number of recipients harboring R388 (TmpR). The ability of R388 to form transferable cointegrates with pSW23 in the absence of ISYps1 was assessed in the same conditions by replacing pSWYps1.2 by pSW23 in the pi3(R388) donor, and found to be inferior to 10−9.
|
10.1371/journal.pgen.1004074 | Tissue Specific Roles for the Ribosome Biogenesis Factor Wdr43 in Zebrafish Development | During vertebrate craniofacial development, neural crest cells (NCCs) contribute to most of the craniofacial pharyngeal skeleton. Defects in NCC specification, migration and differentiation resulting in malformations in the craniofacial complex are associated with human craniofacial disorders including Treacher-Collins Syndrome, caused by mutations in TCOF1. It has been hypothesized that perturbed ribosome biogenesis and resulting p53 mediated neuroepithelial apoptosis results in NCC hypoplasia in mouse Tcof1 mutants. However, the underlying mechanisms linking ribosome biogenesis and NCC development remain poorly understood. Here we report a new zebrafish mutant, fantome (fan), which harbors a point mutation and predicted premature stop codon in zebrafish wdr43, the ortholog to yeast UTP5. Although wdr43 mRNA is widely expressed during early zebrafish development, and its deficiency triggers early neural, eye, heart and pharyngeal arch defects, later defects appear fairly restricted to NCC derived craniofacial cartilages. Here we show that the C-terminus of Wdr43, which is absent in fan mutant protein, is both necessary and sufficient to mediate its nucleolar localization and protein interactions in metazoans. We demonstrate that Wdr43 functions in ribosome biogenesis, and that defects observed in fan mutants are mediated by a p53 dependent pathway. Finally, we show that proper localization of a variety of nucleolar proteins, including TCOF1, is dependent on that of WDR43. Together, our findings provide new insight into roles for Wdr43 in development, ribosome biogenesis, and also ribosomopathy-induced craniofacial phenotypes including Treacher-Collins Syndrome.
| Here, we describe the identification and characterization of a novel zebrafish craniofacial mutant, fantome (fan), caused by a point mutation in the wdr43 gene. Although previously characterized as UTP5 in yeast, a nucleolar protein functioning in ribosome biogenesis, here we show that Wdr43 also regulates early zebrafish development, including NCC specification and differentiation. Mutations in nucleolar proteins have been found to be causative for a variety of human craniofacial syndromes including Treacher-Collins Syndrome (TCS), often caused by mutations in TCOF1, which also plays important roles in ribosome biogenesis. However, the underlying mechanisms linking ribosomal biogenesis and NCC specification and differentiation into pharyngeal arch cartilages remains poorly understood. Here we describe the fan/wdr43 mutant phenotype, and present functional characterizations of Wdr43 in craniofacial development. We show that WDR43 is required for the proper nucleolar localization of a variety of nucleolar proteins, including TCOF1/Treacle. These studies provide new insight into ribosomal protein function in early zebrafish development, with focus on NCC derived craniofacial development, as a model for human craniofacial neurocristopathies.
| Neural crest cells (NCCs), a transient cell type that is unique to vertebrates, originate from the dorsal aspect of the neural tube during embryogenesis. After undergoing an epithelial-to-mesenchymal transition (EMT), NCCs migrate along well defined pathways, and eventually inhabit peripheral destinations where they differentiate into diverse derivatives, including melanocytes, craniofacial cartilage and bone, smooth muscle, and neuronal lineages. In the head region, cranial neural crest cells (CNCC) give rise to nearly all craniofacial structures, including the facial skeleton and the vast majority of facial connective tissues [1], [2]. Defects in CNCC development are associated with craniofacial malformations, one of the most common of human birth defects [3].
Treacher-Collins syndrome (TCS), an autosomal dominant congenital disorder of craniofacial development, characterized by mandibulofacial dysostosis including cleft palate and hypoplasia of the facial bones, is most commonly associated with mutations in the TCOF1 gene [4]. Treacle, the protein encoded by the TCOF1 gene, is a nucleolar phosphoprotein [5], [6] that plays a key role in ribosome biogenesis via involvement in both rDNA methylation and rRNA transcription [7]–[10]. Extensive research in the mouse model has shown that mutations in Tcof1 disrupt ribosome biogenesis, resulting in impaired proliferation and subsequent apoptosis of neuroepithelial and NCC precursors, which in turn results in reduced numbers of NCCs migrating into the developing craniofacial complex [10]. Interestingly, inhibition of p53 function can rescue craniofacial abnormalities in mouse Tcof1 mutants, without rescuing ribosome biogenesis defects [11].
The question of how ribosome biogenesis defects can preferentially affect NCC proliferation and differentiation remains to be elucidated. In eukaryotic cells, ribosome biogenesis begins with the transcription of rRNA from rDNA located in the nucleolus, the most prominent visible structure in the nucleus. Ribosome biogenesis is extremely complex, requiring the accurate processing of pre-rRNAs into four different ribosomal RNAs (28S, 18S, 5S and 5.8S in vertebrates) and complex formation with about 80 constituent ribosomal proteins. In addition, more than 200 nucleolar ribosome biogenesis factors are required to complete the entire ribosome biogenesis process. All ribosomal RNAs, except the 5S rRNA, are initially transcribed as a 47S polycistronic precursor, which subsequently becomes cleaved, folded and modified into the 28S, 18S and 5.8S mature rRNAs prior to being incorporated into functional ribosomes [12], [13].
The cleavage and modification of rRNA is directed by small nucleolar RNAs (snoRNAs). U3, one of the most extensively studied snoRNA, is an essential component of the small subunit (SSU) processome, a large ribonucleoprotein (RNP) complex that is required for the maturation of the 18S rRNA and formation of the ribosomal small subunit (40S) [14]. The SSU processome can be further subdivided to three sub-complexes, UTPB, UTPC and UTPA/t-Utp [14]. The t-Utp complex contains seven proteins (Utp4, Utp5, Utp8, Utp9, Utp10, Utp15 and Utp17) [15], [16]. In yeast, depletion of individual t-Utp members commonly is associated with both pre-rRNA synthesis and processing defects [14], [16], [17], although another group reported that the t-Utp subcomplex plays a role in pre-rRNA stabilization rather than transcription [18]. Although the functions of t-Utp components appear to be conserved in eukaryotes, some human UTP orthologs have not yet been identified [19], indicating that Utp proteins in higher eukaryotes may have evolved specific functions. Recently, a new metazoan specific protein, NOL11, has been characterized as a hUTP4 interacting partner via yeast two hybrid (Y2H) analysis [20].
Although ubiquitously expressed in virtually all eukaryotic cells, mutations in ribosome biogenesis proteins often result in tissue-specific developmental defects [21]. For example, hUTP4/Cirhin is associated with North American Indian childhood cirrhosis (NAIC) [22]. Mutation of zebrafish Bap28, the ortholog of human UTP10, results in excess apoptosis primarily in the central nervous system [23], while mutation in WDR36/UTP21, a modifier protein to human primary open angle glaucoma (POAG), results in mouse embryonic lethality [24]. Mutation of Wdr36 in zebrafish doesn't produce any obvious defects in the first three days of development, while later developmental defects include small eyes and head combined with upregulation of the p53 stress-response pathway [25]. Developmental defects in other organs have also recently been reported [26]–[28].
Here, we report a novel zebrafish mutant, fantome (fan), characterized by a variety of early developmental defects including eye, hindbrain, forebrain, cardiac, neurocranium, fin, and NCC derived pharyngeal arch cartilage development. NCCs in fan mutants fail to differentiate, and NCC precursors undergo p53 mediated cell apoptosis. fan mutants contain a point mutation in the zebrafish wdr43 gene, which encodes Wdr43, the ortholog of yeast Utp5. We demonstrate that, similar to the yeast ortholog, zebrafish and human WDR43 localize to the nucleolus. We also show that the C-terminal of Wdr43, truncated in fan mutants, mediates its localization to nucleoli, and is both necessary and sufficient to mediate its interaction with other t-UTP subcomplex members. Interestingly, blocking WDR43 expression in HeLa cells results in nucleolar maturation defects, together with abnormal localization of other nucleolar proteins, including TCOF1. Together, our data suggest that loss function of Wdr43 results in ribosome biogenesis defects that induce the p53 signaling pathway, triggering cell death of NCC precursors. We introduce the fan mutant as a valuable model to provide insight into a variety of human craniofacial neurocristopathy diseases.
The fan mutant, identified in a large scale ENU chemical mutagenesis screen conducted by the Yelick Lab [29], was notable by its distinctive lack of pharyngeal arch cartilages at 4 days post fertilization (dpf) (Fig. 1D′, arrow). Subsequent developmental analyses showed that fan mutants are first identifiable at 16 hpf by a distinct area of necrotic cells present in the neural ectoderm of the presumptive eye (Fig. 1A′, arrow). At 24 hpf, a larger area of necrosis was detected mainly in neural tissues (Fig. 1B′, bracket), while at 30 hpf, fan mutants displayed distinct hydrocephaly in hind brain ventricles (Fig. 1C′, arrow), incomplete closure of the choroid fissure, and lack of pigmented retinal epithelium in the ventral eye (Fig. 1C′, arrowhead). Preliminary whole mount in situ hybridization (WISH) analyses of NCC marker gene expression revealed that fan mutants exhibited reduced snail2 and dlx2a expression at 12 and 24 hpf, respectively, as compared to age matched wild type sibling (Fig. 1E′, I′ vs. E, I). At 96 hpf, Alcian blue staining revealed that homozygous fan mutants lacked virtually all pharyngeal arch cartilages (Fig. 1D′, arrow) as compared to age matched wild type siblings (Fig. 1D). Early lethal homozygous recessive fan mutants die at approximately 5–6 dpf, while heterozygous fan embryos and adults appear normal and are viable and fertile.
In normal zebrafish development, NCCs originating from the dorsal aspect of the neural tube migrate ventrally to the pharyngeal pouches and give rise to a variety of structures including pharyngeal arch cartilages [30] [31] [32]. To more carefully characterize the pharyngeal arch phenotype observed in fan mutants, we used WISH to examine the developmental expression of additional NCC markers, including: sox9a, essential for proper morphogenesis and differentiation of pharyngeal arch cartilages [33], [34]; the pan-NCC marker crestin, expressed in pre-migratory and migratory NCCs [35]; hand2, expressed in branchial arch mesenchyme [36]; and dlx2a, which is expressed in migrating CNC that contribute to the pharyngeal arches [37]. We detected reduced expression of all NCC marker expression in fan mutants as compared to age matched wild type sibling embryos (Fig. 1F–K′). To better visualize and study NCC defects in fan mutants, we also created a Tg(fli1a:EGFP)/fan mutant reporter line, which expresses EGFP in the derivatives of the cranial neural crest until at least 7dpf, and in developing vasculature [38]. We found that fan mutants exhibited fewer GFP-positive NCCs, and abnormal NCC migration and pharyngeal arch formation, as compared to age matched wild type siblings (Fig. 1L–L′).
Using bulk segregant analyses, we determined that the fan locus mapped between SSLPs z8774 and z9831 on zebrafish Linkage Group (LG) 17 (Fig. 2A), to an interval of 3.93 Mb containing 11 genes. Further analyses of cDNA and genomic DNA sequences of these genes identified a cytosine to thymidine mutation at nucleotide 1066 of the wdr43 gene in this interval, which introduced a premature stop codon at amino acid 356 in exon 9 (Arg356Stop) (Fig. 2A). This mutation was confirmed by sequencing full length wdr43 cDNA amplified from fan mutant mRNA, and sequence analysis of exon 9 of the wdr43 gene in PCR amplified genomic DNA isolated from six individual homozygous fan mutants. This mutation was not present in wild type sibling cDNA or genomic DNA, and was detected along with wild type sequence in heterozygous fan family members. Full length zebrafish Wdr43 contains 650 amino acid and is well conserved from yeast to human (Data not shown). Domain analysis of the zebrafish Wdr43 protein reveals that it is composed of three WD40 repeats and one Utp12 domain (http://pfam.sanger.ac.uk) (Fig. 2A). The truncated form of Wdr43 encoded in the fan mutant lack the C-term 294 aa, including the Utp12 domain (Fig. 2A).
The identified gene mutation in fan/wdr43 mutants was confirmed using two approaches. We used single cell injections of titrated amounts of wdr43 antisense morpholino oligomers (MOs) to test whether targeted depletion of Wdr43 in wild type embryos resulted in embryos that phenocopied the fan mutant. We first confirmed the functional targeting of anti-sense wdr43 MOs by demonstrated quenching of the wdr43-GFP mRNA construct fluorescence in vivo (Fig. S1). We next injected wdr43 MOs into clutches of fan mutants and wild type single cell stage embryos, which were then raised to 3–5 dpf and stained with Alcian blue to examine pharyngeal arch cartilage formation. Our results showed that wild type embryos injected wdr43 MOs exhibited early neural tissue necrosis similar to that observed in fan mutants (Fig. 2C, E, arrows). When wdr43 MOs were injected into single cell stage fan mutant embryos, we observed no apparent exacerbation of the fan mutant phenotype, suggesting that the fan mutation is a functional null (data not shown). WISH analyses of wdr43 MO injected embryos revealed similar reduction in NCC marker gene expression, as observed in fan mutants (Figure S2).
Secondly, we performed rescues by injecting full length wild type wdr43 mRNAs into single cell stage fan mutant and wild type sibling embryos. Analyses of injected embryos at 5 dpf via Alcian blue staining revealed rescue of the pharyngeal arch cartilage formation, although full rescue was not observed (Fig. 2, I versus G, F). Together, these data provide strong evidence that the identified wdr43 gene mutation results in the fan mutant phenotype.
We next examined the developmental expression pattern of wdr43 mRNA via whole mount and sectioned in situ hybridization (ISH). wdr43 is maternally expressed, and maintains a fairly ubiquitous expression pattern during the first 24 hours of development (Fig. 3A–J). wdr43 expression becomes restricted to neural and pharyngeal arch tissues between 24 and 48 hpf (Fig. 3K), consistent with the pharyngeal arch defects observed in fan mutants. Sectioned ISH demonstrated discrete wdr43 mRNA expression in neurepithelium and pharyngeal arches at 48 and 72 hpf (Fig. 3, N, N′, P, P′, arrows), consistent with the observed hindbrain and pharyngeal arch defects observed in fan mutants. Strong expression was also observed in the gut epithelium (Fig. 3, N, P). We examined the expression of fan mutant wdr43 mRNA using RT-PCR analysis of developmentally staged fan mutant and wild type siblings followed by digestion with Dde I, a unique restriction site introduced by the fan allele. These analyses showed that wild type wdr43 was detectable at all stages examined, and also that mutant wdr43 was detectable in fan mutants at 48 and 72 hpf, and thus was apparently not targeted for nonsense mediated decay (Fig. 3, M).
To better characterize tissue necrosis and cell proliferation in fan mutants, TUNEL assay and phosphohistone H3 (pH3) IF analyses were performed, respectively, on developmentally staged fan mutant and wild type sibling embryos (Fig. 4). TUNEL revealed significantly upregulated apoptosis in fan mutants at all developmental staged examined (Fig. 4A, arrows). In contrast, cell proliferation, indicated via pH3 antibody staining, was decreased in fan mutants as compared to age matched wild type siblings (Fig. 4B). Quantification of TUNEL and pH3 immunofluorescence showed significantly increased apoptosis in fan mutants as compared to age matched wild type siblings at all stages examined, and significantly decreased cell proliferation at 48 hpf (Fig. 4C). Together, these results are consistent with the observed lack of NCC derived pharyngeal arch tissues in fan mutants.
To better understand the molecular nature of mutant Wdr43 protein, we next examined the subcellular localization of wild type and fan mutant zebrafish Wdr43 in human cells. First, we performed immunofluorescence analyses of cultured HeLa cells using the anti-human WDR43 antibody and demonstrated that endogenous human WDR43 localized to nucleoli, as shown by co-localization with the nucleolar marker protein, B23 (Fig. 5, A1–A4). Next, we generated N-terminal EGFP-tagged wild type and truncated fan mutant zebrafish wdr43 constructs driven by the CMV promoter, which we transfected into cultured HeLa cells, and then visualized chimeric fusion protein using anti-GFP antibody. The EGFP-tagged wild type Wdr43 protein showed perfect overlapping expression pattern with B23 (Fig. 5, B1–B4), consistent with the previously characterized nucleolar localization of the yeast ortholog for Wdr43, Utp5 [39]. In contrast to the full length EGFP-Wdr43 protein, EGFP-tagged fan mutant Wdr43 (amino acids 1–364) lost its exclusive nucleolar localization, and rather exhibited expression throughout the nucleus (Fig. 5, C1–C4). To correlate these in vitro results to in vivo expression studies in zebrafish, the same EGFP-tagged zebrafish wild type and fan mutant wdr43 constructs were injected into single cell stage zebrafish, which were then analyzed via confocal microscopy (Fig. S3). These analyses also showed that while full length Wdr43 co-localized with mCherry-B23 to nucleoli (Fig. S3, A–C), truncated fan mutant Wdr43 remained dispersed throughout the nucleus (Fig. S3, D–F).
Utp5, the yeast ortholog of Wdr43, has been shown to function in the yeast t-Utp subcomplex, which mediates both pre-ribosomal RNA (rRNA) transcription and processing [16] [17]. Previously, it has been shown that yeast Utp5 interacts with Utp4 and Utp15 (Freed & Baserga 2010) [40]. We used yeast two-hybrid analyses (Y2H) to examine the protein-protein interactions between yeast and zebrafish wild type and fan mutant Wdr43 with other t-Utp complex proteins. We found that both yeast and zebrafish full length Wdr43 interacted with Utp4 and Utp15 (Fig. 5D and Fig. S4), consistent with previously published yeast Utp5/Wdr43 binding studies [17]. We also determined that the C-terminal portion of Wdr43 protein is both necessary and sufficient to mediate this interaction. Zebrafish and yeast truncated fan mutant Wdr43 did not bind to either Utp4 or Utp15, while the C-terminal fragment of Wdr43 alone was able to bind to both Utp4 and Utp15 (Fig. 5D zebrafish and Fig. S4 yeast). Together, these data revealed the conserved interaction of yeast and zebrafish full length Wdr43 proteins with Utp4 and Utp15, and also suggest that the C-terminal portion of Wdr43, which contains the Utp12 domain, is required for protein interaction of Wdr43 with other t-UTP subcomplex member proteins.
Based on our results, and those of previously published reports, we anticipated that Wdr43 would play an important role in ribosome biogenesis. We therefore investigated pre-rRNA synthesis and processing in 30 hpf and 50 hpf fan mutant and wild type sibling embryos via Northern blot analysis, using a probe specific for the 5′ external transcribed spacer (5′ETS) region of the pre-rRNA at the start site of transcription (Fig. 6A). These analyses showed reduced levels of the primary transcript (labeled a) in fan mutants (M) at both 30hpf and 50 hpf as compared to that of wild type (W) siblings, consistent with a defect in pre-rRNA transcription in fan mutants. Quantification of the ratio of full length primary transcript (a) to the processed pre-18S rRNA (b) showed reduced pre-rRNA in fan mutants relative to age matched wild type siblings. These results are consistent with previously published results showing that siRNA knockdown of human UTP5 resulted in defects in pre-rRNA transcription and processing [19], and suggest conserved functions for vertebrate Wdr43 and yeast Utp5/Wdr43 in pre-rRNA transcription [17].
Having defined an important role for zebrafish Wdr43 in ribosome biogenesis, we further examined its function in cultured human HeLa cells, which we transfected with human WDR43 small interfering RNA (siRNA) (Sigma MISSION esiRNA) to silence WDR43 expression. GFP esiRNA was used as a negative control for these studies. Analysis of WDR43 protein and mRNA expression in siRNA treated cells using both Western blot (Fig. 6B) and qRT-PCR (Fig. 6C) analyses, respectively, confirmed that endogenous WDR43 expression was significantly reduced with WDR43 siRNA treatment. We next examined how WDR43 depletion affected the localization of another t-UTP complex Wdr43 interacting protein, UTP15. Due to the lack of available antibody for UTP15, we transfected N-terminal mCherry tagged zebrafish UTP15 into HeLa cells, and monitored its localization via fluorescent confocal microscopy. Consistent with a role for UTP15 in pre-rRNA processing, we found that mCherry tagged UTP15 localized to nucleoli in control GFP siRNA treated cells (Fig. 6D). In contrast, in HeLa cells depleted of WDR43 using WDR43 siRNA, mCherry-tagged UTP15 did not localize to nucleoli, but rather appeared exhibited a perinuclear expression pattern (Fig. 6E). These results indicate that WDR43 is required for entry into the nucleus, as well as for proper nucleolar localization of UTP15.
It was intriguing to us that many of the observed phenotypes observed in fan mutant zebrafish have also been reported in humans (and mice) with mutations in TCOF1/Treacle, the gene commonly mutated in Treacher-Collins Syndrome (TCS). TCS results in aberrant NCC specification and differentiation, craniofacial dysmorphologies, increased cell apoptosis, and upregulated p53 signaling [10], [11]. Based on these common characteristics, we investigated whether the localization of TCOF1 and other nucleolar proteins was affected in human HeLa cells depleted of WDR43 protein. For reference, we examined the expression of the nucleolar proteins Mpp10, Nucleolin and Fibrillarin, which have been associated with distinct nucleolar functions. Mpp10 is normally found in the dense fibrillar component (DFC) and in the boundary between the DFC and the fibrillar center (FC), sites of rDNA transcription and pre-rRNA splicing and modification by snoRNPs, while Fibrillarin is normally found in association with snoRNAs throughout the DFC [41]. Nucleolin/C23 is normally localized to the outer layer of nucleoli with fainter expression at the center [42]. Our investigation of the expression of these nucleolar proteins, and TCOF1, in control and WDR43 siRNA treated HeLa cells (Figure 7) showed that TCOF1 exhibited a reduced and perinucleolar expression pattern in WDR43 depleted cells (Fig. 7 B vs. A, arrows). Mpp10 expression appeared reduced, but was expressed throughout the smaller nucleoli (Fig. 7 B vs. D, arrows). Nucleolin, although barely detectable in WDR43 siRNA expressing cell lines as compared to control GFP siRNA treated cells, was also expressed in a perinucleolar fashion, similar to that of TCOF1 (Fig. 7 F vs. E, arrows). In contrast, Fibrillarin expression appeared relatively less affected in WDR43 depleted cells, and was detected throughout the smaller nucleoli (Fig. 7H vs. G, arrows). These results are indicative of disrupted nucleolar organization and rRNA transcription, consistent with the observed defects in pre-rRNA transcription observed in fan mutants. To confirm and more reliably study these observations, we next examined TCOF1 localization in stable HeLa cell lines expressing stable short hairpin RNAs (shRNAs) for GFP control and WDR43 (Fig. 8). We first tested five shRNAs against WDR43 gene and found that all of them showed somewhat reduced WDR43 expression using qRT-PCR analysis (Fig. S5). Western blot analyses showed that two of the shRNA WDR43 cell lines, sh9 and 2a1, exhibited the most efficient inhibition of protein levels. As observed in WDR43 siRNA treated cell lines, stable WDR43 shRNA expressing cell lines sh9 and 2a1 exhibited mislocalized TCOF1 expression at the periphery of nucleoli, as compared to control shRNA stable cell lines (Fig. 8, E, I, N, O vs. A, M, arrows). Together, these results suggest that WDR43 expression is required for proper nucleolar organization and the subnucleolar localization of a variety of nucleolar proteins including TCOF1, and for optimal pre-rRNA transcription.
We also found that WDR43 depletion had an effect on nucleolar size and shape. WDR43 depleted cells had larger numbers of mini nucleoli as compared to control cells, which exhibited fewer numbers of larger sized, mature nucleoli. To quantitate this observation, we monitored nucleolar number and size in control and WDR43 depleted cultured HeLa cells immunostained for TCOF1 (Fig. 8). We found that in normal and control shRNA cultured HeLa cells, nucleoli reassembled after mitosis, with several small nucleoli fusing into ∼1–4 larger, mature nucleoli per HeLa cell. In contrast, TCOF1 expressing nucleoli failed to fuse together in WDR43 shRNA expressing HeLa cells, but rather remained as small unfused mini nucleoli, or “nucleolar caps”, which also appeared spherical shape as compared to the irregular shaped nucleoli present in control HeLa cell cultures (Fig. 8 F, J vs. B). We also found that the average number of nucleoli was increased, and the number of fused nucleoli was reduced, in WDR43 shRNA expressing HeLa cells as compared to control shRNA treated HeLa cells, as shown using the nucleolar marker B23 (Fig. 8, F, J vs. B). Quantification of these results revealed that WDR43 depleted cells exhibited increased numbers of smaller nuclei as compared to control cells (Fig. 8, Q, R). Together, these results demonstrate that depletion of WDR43, an essential ribosome biogenesis factor, affects nucleolar maturation and assembly.
Ribosome biogenesis defects, such as those observed in fan mutants, have been reported to be associated with upregulation of the p53 signaling pathway, and cellular apoptosis [11], [25] [43] [44] [45] [46]. Based on the increased apoptosis observed in fan mutants, we next examined p53 signaling pathway gene expression in developing fan mutant and wild type sibling embryos. Immunohistochemical analysis using the zebrafish p53 antibody (kind gift of David Lane) [47] revealed high levels of expression of p53 in fan mutants, while p53 was virtually undetectable in wild type sibling embryos at 5 dpf (Fig. 9, B vs. A). Next, we used qRT-PCR analyses to show that the expression of p53 downstream target genes, including the N-terminal truncated p53 isoform delta113p53, mdm2, and cyclin G1 were all upregulated in 24, 48 and 72 hpf fan mutants as compared to wild type sibling controls (Fig. 9, C, D and data not shown). We next tested whether knockdown of p53 signaling via injection of p53 anti-sense MOs could rescue the fan mutant phenotype. Similar to previous reports in the Treacher-Collins mouse model [11], we found that the fan mutant pharyngeal, neural and eye defects were largely rescued in p53 MO injected fan mutants (Fig. 9, G vs. F, and H), and NCC marker gene expression was also rescued in fan mutants (Fig. S2). TUNEL analyses revealed rescue of apoptosis in p53MO injected fan mutants (Fig. 9. K vs. J, I), indicating that increased apoptosis observed in fan mutants was mediated via upregulated p53 signaling pathways.
Proper ribosome biogenesis is required for the production of functional ribosomes, the primary site of protein synthesis. Most if not all ribosomal proteins (RPs) are thought to be essential for ribosome biogenesis and cell survival. It is therefore surprising that ribosome biogenesis defects caused by mutations in certain RPs can lead to variable and seemingly tissue-specific defects in vertebrate development. For example, mutations in several RPs are associated with human congenital hypoplastic Diamond-Blackfan anemia (DBA), and similar DBA phenotypes were observed when DBA associated RP mutations were expressed in zebrafish [48]–[51]. In addition, several reports on zebrafish ribosome biogenesis protein mutants describe a variety of diverse phenotypes, ranging from tumors, to central nervous system degeneration, to organogenesis defects [23], [25], [52], [53].
Here we present data showing that defects in the zebrafish ribosome biogenesis protein Wdr43 result in early developmental defects in a variety of tissues, including neural, eye and heart and pharyngeal arches, while later developmental defects appear fairly localized to NCC derived pharyngeal arch cartilages. These observations raise the question of how a defect in what is thought to be a universally required ribosomal biogenesis protein, Wdr43, can result in a rather specific craniofacial tissue-specific phenotype? One possibility is that there are tissue specific, developmental requirements for ribosome biogenesis proteins. For example, certain ribosome biogenesis factors may have tissue specific, developmental expression patterns. In fact, we show here that the expression of zebrafish wdr43 mRNA becomes localized to neural and pharyngeal arch tissues starting at ∼24 hpf, which is consistent with the observed fan/wdr43 mutant phenotype. It is possible that additional ribosomal proteins that regulate cell cycle or apoptosis may similarly exhibit tissue specific expression patterns. Another theory is that ribosome biogenesis defects and subsequent anticipated reduced protein translation efficiency will most significantly affect those tissues exhibiting a high demand for protein synthesis. This may include NCC and erythropoiesis progenitor cells, although recent data does not find evidence for increased translation in NCC [21]. Finally, decreased efficiency of cellular translation machinery may affect a wide and varied spectrum of translation products in different cell types due to mRNA competition for timely translation, which could result in diverse readouts in different cell types.
The craniofacial phenotype exhibited by zebrafish fan mutants resembles the craniofacial malformations observed in individuals with Treacher-Collins Syndrome. The fact that NCC specification and differentiation are similarly affected by mutations in nucleolar proteins – TCOF1/Treacle and a common subunit of RNA polymerases I and III in Treacher-Collins Syndrome [54] [55], and Wdr43 in fan mutants - leads to the intriguing question of why NCCs may be more sensitive to ribosome biogenesis defects as compared with other tissues. Based on our data presented here and on the published reports of others, we hypothesize that high protein translation levels must be maintained by progenitor and differentiating NCCs in order to support their extensive cell proliferation, migration and differentiation. In Treacher-Collins Syndrome, TCOF1 mutant induced defects in ribosome biogenesis are characterized by stimulation of the nucleolar stress response, which in turn activates the p53 apoptosis pathway, resulting in the depletion of the neural crest precursor pool [10]. We observe a similar upregulation of p53 signaling and depletion of NCCs in fan mutants. Although beyond the scope of the present study, it will be interesting in future studies to compare pre-rRNA transcription, ribosome biogenesis and protein translation efficiency in developmentally staged NCC versus non-NCC populations harvested from fan mutant and wild type siblings.
Our Northern blot results indicated that pre-rRNA levels are significantly decreased in developmentally staged zebrafish fan/wdr43/utp5 mutants, consistent with the previously characterized role for yeast Utp5 in pre-rRNA transcription [16], [19]. We suggest that a variety of nucleosomal proteins are required for optimal pre-rRNA transcription. Novel findings from this report include the fact that blocking WDR43 function in HeLa cells resulted in the distinct mislocalization of nucleolar proteins including UTP15, Mpp10, nucleolin and to a lesser extent fibrillarin, suggesting that Wrd43/UTP5 is required for proper subnucleolar organization and function. Interestingly, TCOF1 also mislocalized to the outer periphery of nucleoli, rather than exhibiting its normal distribution throughout the nucleolus, suggesting that WDR43 may also be required for proper TCOF1 subnucleolar localization and function. Although we have not detected direct binding between TCOF1 and WDR43/UTP5 using Y2H, we have detected interactions between WDR43/UTP5 and other rDNA transcription component proteins (data not shown). Together, these results suggest roles for Wdr43/UTP5 in ribosomal protein sub-nucleolar localization and function of other ribosome biogenesis factors, and raise the intriguing possibility that manipulation of WDR43 expression could be used to correct the localization and improve the function of TCOF1 in Treacher-Collins Syndrome patients.
Nucleolar mis-localization phenotypes have also been observed in HeLa cells treated with actinomycin D, an inhibitor of RNA Pol I [56], which is a TCOF1/Treacle interacting protein [57]. It is possible that WDR43 may also function together with TCOF1 and Nopp140 to recruit proteins to the nucleolar organizer regions (NORs) and the upstream binding factor (UBF), an RNA PolI transactivator [19]. Wdr43 could also mediate rRNA transcription by binding to rDNA and UBF directly, as shown by other Utps [58]. These functions for Wdr43 remain to be elucidated.
We also used both siRNA and shRNA WDR43 silencing methods to confirm the function of WDR43 in nucleolar fusion in cultured HeLa cells. At the present time, mechanisms regulating nucleolar fusion remain poorly understood. It has been shown that after mitosis, multiple small nucleoli form around transcriptionally active NORs, and as cells progress through the cycle, these small nucleoli fuse to form larger nucleoli [59] [60]. Although the mechanism of WDR43 function in nucleolar fusion is not clear, preventing nucleolar fusion may not be common to all ribosome biogenesis protein mutations based on the fact that inhibition of NOL11 resulted in the formation of one large (not small) nucleolus [20]. One possible explanation is that WDR43 depletion may result in structural changes to rDNA, which in turn could interfere with nucleolar fusion [61]. Such a proposed function for WDR43 may be dependent or independent of its function in the t-Utp complex. A recent study using Xenopus oocytes showed that nucleoli exhibit fluid dynamics similar to that of liquid droplets, and that nucleolar fusion requires dynamic exchange between nucleoli and the nucleoplasm [62]. In future studies, it will be interesting to determine whether WDR43 is also involved in this process.
Recent reports emphasize the apparent tissue specific functions for ribosomal proteins previously thought to exhibit functions in all cells and tissues. The results presented here suggest previously unrecognized roles for Wdr43/UTP5 in craniofacial development. The fact that Wdr43/UTP5 is needed for proper formation of nucleoli and for sub-nucleolar organization and function indicates important roles for Wdr43 as a key participant in ribosome biogenesis. As such, the zebrafish mutant fantome provides a valuable vertebrate developmental model and tool to continue in depth functional studies of RPs and ribosome biogenesis factor proteins in NCC differentiation, including the identification of effective tools for reducing the incidence of craniofacial birth defects.
AB and WIK fantome/wdr43 mutant and wild type zebrafish were raised in the Tufts Zebrafish Facility at 28.5°C and developmentally staged as previously described (Westerfield, M., 1995). For whole mount in situ hybridization analyses, pigmentation was inhibited by treating embryos with 1-phenyl-2-thiourea (PTU) at a final concentration of 0.2 mM as previously described (The Zebrafish Book, U. Oregon Press).
All experimental procedures on zebrafish embryos and larvae were approved by the Tufts University Institutional Animal Care and Use Committee (IACUC) and Ethics Committee.
The fan mutant was identified in a large-scale ENU-mutagenesis screen conducted by the Yelick Laboratory [29]. Genetic mapping strains were created by crossing identified heterozygous fan mutants to polymorphic WIK wild type zebrafish. Embryos were collected from pairwise matings of mapping strain fan/WIK heterozygotes, and scored at 48 hpf for fan specific phenotypes. Genomic DNA was extracted from individual fan mutant and wild type embryos, and bulk segregant analyses were performed using primers designed to amplify SSLP markers from the Massachusetts General Hospital Zebrafish Server website (http://zebrafish.mgh.harvard.edu). The fan mutation mapped to zebrafish linkage group 17 (LG17), to a region spanning 11 genes. Nucleotide sequence analyses of all 11 genes identified a premature stop codon in the wdr43 gene of all fan mutant embryos that was not present in wild type siblings.
Whole-mount and sectioned in situ hybridizations were performed as previously described (Thisse et al., 1993), using a probe generated via PCR using the following primers (wdr43-forward: 5′- CAGTGCAACAAAAGTTGGTGA-3′; wdr43-reverse: 5′- AAAGTTCTGGTTGGCTGCA-3′). All other probes were obtained from zfin.org. Embryos were analysed using Zeiss Axiophot and M2Bio microscopes, and imaged using Zeiss Axiophot Imager digital camera (Munich, Germany). Digital images were processed using Adobe Photoshop software.
Antisense morpholino oligonucleotides (MOs) targeted to the initiation of translation codon of wdr43 mRNA (5′TCCGTCCGCCGCCATCTTACCGTTC3′) were injected into the yolk of 1 cell stage wild type or fan mutant embryos. 2 nL of MO at a concentration of 10 ng/µL was used to knockdown wdr43 translation.
Total RNA was extracted from 20 wild type and 20 fan mutant embryos at 24, 48 and 72 hpf, or from HeLa cells 48 hours after transfection using RNeasy Plus Kit (Qiagen, Valencia, CA). DNA was removed using the DNA-free DNase Treatment & Removal Kit (Ambion,Life Technologies, Grand Island, NY) to remove genomic DNA contamination. cDNA was synthesized using a SuperScript III First-Strand Synthesis System (Invitrogen, Life Technologies, Grand Island, NY) with random primers. Gene expression was quantified by qRT-PCR using QuantiTect SYBR Green PCR Master Mix (Qiagen, Valencia, CA) and real-time cycler Mx3000P (Stratagene, Agilent Technologies, Santa Clara, CA). Primers for zebrafish p53 isoforms, mdm2, and cyclinG1 were used as described [63]. The following primers were used to amplify the human WDR43 gene: Forward: CCTTCCGCGCACCTCAGTGGTAC; Reverse: AACTGGCGTTGCATGTCCTGTGA. Primers for β-actin, used to normalize the expression levels, were as described [64].
Yeast two-hybrid assays for interaction between yeast and zebrafish Utp proteins were performed as previously described (Freed and Baserga, 2010). Briefly, yeast utp5 cDNA encoding full length, N-terminal (1–343aa) and C-terminal (344–643aa) proteins were cloned into the pGADT7 prey vector. Additional yeast UTP genes of the t-Utp subcomplex (Utp8, Utp9, Utp10, Utp15 and Utp17) cloned into the bait vector were as previously described (Freed and Baserga 2010). Both bait and prey vectors were transformed into AH109 yeast strain and interactions were identified based on the ability of transformants to grow on AHTL dropout medium after 3–5 days of incubation at 30°C. To test the interaction between zebrafish Wdr43/Utp5, Utp4 and Utp15, full length zebrafish utp4 and utp15 cDNAs were purchased (Openbiosystems, Lafayette, CO) and cloned into pGADT7 prey vector. Zebrafish wdr43 cDNAs encoding full length, truncated fan mutant Wdr43 (1–356aa) and C-terminal portion of Wdr43 (357–650aa) proteins were cloned into the pGABT7 bait vector. Interactions were tested by growth in triple dropout medium after 3–5 days of incubation at 30°C. The Y2H studies of zebrafish Wdr43 protein interactions were tested in both bait and prey constructs.
To determine the subcellular localization of wild type and mutant Wdr43 proteins, GFP cDNA was cloned onto the N-terminal end of full length or fan mutant zebrafish wdr43 cDNA under the direction of the CMV promoter, using multi-site Gateway reactions [65]. These constructs were then transfected into HeLa cells using Lipofectamine 2000 reagent (Invitrogen, Life Technologies, Grand Island, NY). After 36 hours of growth at 37°C, transfected cells were fixed with 4% PFA, and subjected to standard immunofluorescence (IF) analyses using the anti-B23 antibody (1∶200, Santa Cruz Biotechnology, Inc., Santa Cruz, CA). The rabbit polyclonal anti-WDR43 antibody (1∶100, Abcam, Cambridge, MA) was used to detect the endogenous WDR43. To check the expression in zebrafish embryos, the same constructs were injected into single cell zebrafish embryos, which were harvested at 24hpf, and analyzed for GFP expression using a Leica TCS SP2 confocal microscope.
For siRNA experiments, HeLa cells were transfected with WDR43 MISSION esiRNA (Sigma, EHU004691) using Lipofectamine 2000 reagent (Invitrogen, Life Technologies, Grand Island, NY). GFP esiRNA (Sigma, EHUEGPF) was used as a negative control. Treated cells were harvested 48 hours after transfection for Western blotting and immunofluorescence experiments. shRNA constructs were purchased from Openbiosystems and used to establish stable shRNA expressing HeLa cells following the manufacture's protocol. The following shRNA constructs were used: pGIPZ-WIPI1-2, RHS4430-98853022; pGIPZ-non-targeting control RMS4348.
Total RNA was extracted from 20 wild type and 20 fan mutant embryos at 30 and 50 hpf using standard Trizol protocol for RNA isolation. Northern blot analysis was carried out as described in Freed et al, 2012 [20]. For each sample, 2 µg of RNA was separated by electrophoresis on a 1% agarose/1.25% formaldehyde gel in Tricine/Triethanolamine buffer and transferred to a nylon membrane (Hybond-XL, GE Healthcare). Pre-rRNA species were detected by methylene blue staining and hybridization with a 32P-radiolabelled oligonucleotide probe to the 5′ETS: CGAGCAGAGTGGTAGAGGAAGAGAGCTCTTCCTCGCTCA. Quantification of pre-rRNA processing was performed using Image J (National Institutes of Health).
Developmentally staged wild type and fan mutant embryos, fixed and processed for cryosectioning, were sectioned at 10 µm. Apoptosis TUNEL assay was performed using the In situ cell death detection kit, Fluorescein (Roche Applied Science, Indianapolis, IN, USA). Cell proliferation was assayed with phospho-Histone H3 immunofluorescence analysis, using anti- phospho Histone H3 (Ser10) antibody (Cell Signaling, Danvers, MA) and anti-rabbit goat antibody conjugated with Alexafluor 594 (Life Technologies, Grand Island, NY).
|
10.1371/journal.ppat.1006746 | Stably expressed APOBEC3H forms a barrier for cross-species transmission of simian immunodeficiency virus of chimpanzee to humans | APOBEC3s (A3s) are potent restriction factors of human immunodeficiency virus type 1/simian immunodeficiency viruses (HIV-1/SIV), and can repress cross-species transmissions of lentiviruses. HIV-1 originated from a zoonotic infection of SIV of chimpanzee (SIVcpz) to humans. However, the impact of human A3s on the replication of SIVcpz remains unclear. By using novel SIVcpz reporter viruses, we identified that human APOBEC3B (A3B) and APOBEC3H (A3H) haplotype II strongly reduced the infectivity of SIVcpz, because both of them are resistant to SIVcpz Vifs. We further demonstrated that human A3H inhibited SIVcpz by deaminase dependent as well independent mechanisms. In addition, other stably expressed human A3H haplotypes and splice variants showed strong antiviral activity against SIVcpz. Moreover, most SIV and HIV lineage Vif proteins could degrade chimpanzee A3H, but no Vifs from SIVcpz and SIV of gorilla (SIVgor) lineages antagonized human A3H haplotype II. Expression of human A3H hapII in human T cells efficiently blocked the spreading replication of SIVcpz. The spreading replication of SIVcpz was also restricted by stable A3H in human PBMCs. Thus, we speculate that stably expressed human A3H protects humans against the cross-species transmission of SIVcpz and that SIVcpz spillover to humans may have started in individuals that harbor haplotypes of unstable A3H proteins.
| Cellular cytidine deaminases of the APOBEC3 (A3) family are potent restriction factors that are able to inhibit retroviruses, this A3 restriction is counteracted by lentivirus Vif proteins. Human APOBEC3H (A3H) represents the most evolutionarily divergent A3 gene; it includes seven haplotypes and several splice variants. The polymorphism of human A3H has relevance for HIV-1 infection and AIDS progression. HIV-1 originated from cross-transmission of SIVcpz to humans. However, little is known about how human A3s affect the replication or transmission of SIVcpz. In this study, we comprehensively analyzed the anti-SIVcpz activity of chimpanzee and human A3s. Human A3H haplotype II was identified as strong inhibitor against SIVcpz regardless of Vif. In addition, other stably expressed human A3H haplotypes and splice variants showed strong antiviral activity against SIVcpz. Moreover, based on the recent Great Ape Genome Project, we found that the polymorphism of chimpanzee A3H is lower compared with the diversity of human A3H. And chimpanzee A3H haplotypes identified in this study showed similar anti-SIVcpz activity and Vif sensitivity. Our results provide a model that stably expressed human A3H protects humans against the cross-species transmission of SIVcpz and that SIVcpz spillover to humans may have started in individuals that harbor haplotypes of unstable A3H proteins.
| Simian immunodeficiency virus (SIV) naturally infects many species of African Old-World monkeys, such as African green monkeys, mandrills and red-capped mangabey [1,2]. However, these viruses appear to be nonpathogenic in their natural hosts [2,3]. Chimpanzees (cpz), which are the evolutionarily closest extant primate to Homo sapiens, are infected by SIVcpz [4]. The common chimpanzee includes four subspecies, only two of which, Pan troglodytes troglodytes (Ptt) and Pan troglodytes schweinfurthii (Pts), are infected by SIVcpz (SIVcpzPtt and SIVcpzPts, respectively) [4]. Genome analysis of SIVcpz indicates that SIVcpz originates from the cross-species transmission and recombination of three different SIV strains: SIVrcm from the red-capped mangabey (rcm), SIVgsn/mus/mon from the greater-spot-nosed (gsn), mustached (mus), and mona monkeys (mon), respectively, and a currently unidentified SIV [5–7]. SIVcpzPts is thought to be the origin of SIVcpzPtt after intra-chimpanzee transmission [5].
SIVcpz is of particular interest because it is the ancestor of human immunodeficiency virus (HIV)-1. HIV-1 M and N groups originated from zoonotic transmission of SIVcpzPtt from west-central Africa [8,9]. Additionally, recent studies indicate that SIVgor from gorillas (gor) is the origin of HIV-1 groups O and P [10,11]. The HIV-1 M group is the pandemic virus, whereas viruses of groups N and P are only found in a few infected individuals [12,13]. The HIV-1 O group is mainly distributed in west-central Africa and has a low prevalence rate (less than 1% of global HIV-1 infections) [14,15]. The other HIV lentivirus, HIV-2, resulted from cross-species transmission of SIV from the sooty mangabey monkey (SIVsmm) [14].
Human intrinsic cellular antiviral factors may have direct relevance for the zoonotic infection of humans and the human-to-human spread of SIVs. Several restriction factors have been identified that repress lentiviral replication [16–18]. The family of human APOBEC3 (A3) restriction factors is formed by seven different proteins, A3A–D and A3F–H. Virion encapsidated APOBEC3D (A3D), A3F, A3G, and A3H inhibit HIV-1 that lacks the gene vif (HIV-1Δvif) by deaminating cytidines in the viral single-stranded DNA that is generated during reverse transcription, thereby introducing G-to-A hypermutations in the coding strand [19]. To achieve productive infections, lentiviral Vif proteins directly interact with A3s and recruit them to an E3 ubiquitin ligase complex to induce A3 degradation by the proteasome [20–22]. Several studies have investigated how A3G serves as a barrier for cross-species transmission of lentiviruses [23–25]. Human A3H represents the most evolutionarily divergent A3 gene; it includes seven haplotypes and several splice variants [26–28]. The protein stability of human A3H is one determinant of its antiviral activity [29–31]. The human A3H haplotype (hap) II, in contrast to A3G, is only sensitive to specific HIV-1 Vifs and adaptation of HIV-1 Vif to A3H hapII has been described [32–35]. Thus, the polymorphism of human A3H has relevance for HIV-1 infection and AIDS progression [36,37].
To investigate how human A3s may affect the replication of SIVcpz, we generated novel luciferase reporter viruses based on two SIVcpz strains (SIVcpzPtsTAN1 and SIVcpzPttMB897). This system revealed that SIVcpz transmission to humans may have been significantly affected by the presence of stable A3H.
To test SIVcpz, we first generated SIVcpz nanoluciferase (NLuc) reporter viruses using two SIV strains (SIVcpzPttMB897 and SIVcpzPtsTAN1; S1 Fig). SIVcpzPttMB897 was isolated from wild chimpanzee (Pan troglodytes troglodytes) in southern Cameroon in 2007 [9,38], and this strain is regarded as the ancestor of the pandemic HIV-1 M group [14]. SIVcpzPtsTAN1 was derived from chimpanzee subspecies Pan troglodytes schweinfurthii, and this strain does not cause sustained infections of humans [39]. The SIVcpz reporter constructs were generated by replacing most of the nef gene with NLuc. Additionally, the vif gene of SIVcpz was inactivated (S1 Fig). SIVcpz-NLuc reporter viruses pseudotyped with the glycoprotein of the vesicular stomatitis virus (VSV-G) were produced by plasmid transfection of 293T cells. When infected with the VSV-G pseudotyped viruses, SIVcpzPttMB897-NLuc and SIVcpzPtsTAN1-NLuc, 293T cells showed high luciferase counts while very low nanoluciferase activity was detected when the viruses were not VSV-G pseudotyped (Fig 1). The luciferase activity of SIVcpzPtsTAN1-NLuc was around 10-fold less than SIVcpzPttMB897-NLuc, even when equal amounts of virions normalized for reverse transcription activity were used for infection (Fig 1). Thus, these two novel SIVcpz reporter viruses transmitted the luciferase enzyme activity via glycoprotein-dependent infection.
Four SIV-luciferase reporter viruses based on SIV of macaques (SIVmac), African green monkeys (SIVagm) and chimpanzees (SIVcpzPts, and SIVcpzPtt) were used to investigate the antiviral activity of chimpanzee A3s. We found that cpzA3C, D, F, G, and H reduced the infectivity of SIVmacΔvif (Fig 2A). SIVmac Vif fully antagonized restrictions of cpzA3C, G, and H, and to a large extent overcame cpzA3F, but it did not inhibit the restriction of cpzA3D (Fig 2A). Chimpanzee A3s showed a similar restriction pattern against SIVagmΔvif, but SIVagm Vif only abolished the restriction of cpzA3C and partly inhibited the restriction of cpzA3H. Even in the presence of SIVagm Vif, cpzA3D, F, and G significantly reduced the infectivity of SIVagm (Fig 2A). The expression of the cpzA3C, F, G, and H was detectable by immunoblotting using HA-tagged specific antibodies, while cpzA3D was not detectable using our immunoblotting system (S2A and S2C Fig).
In the absence of Vifs, cpzA3C reduced the infectivity of SIVmac and SIVagm by 5–10 fold and weakly inhibited SIVcpzPttMB897 and SIVcpzPtsTAN1 by 1–2 fold (Fig 2C and 2D). cpzA3D, F, and H inhibited SIVcpzΔvif by 10–15-fold, while cpzA3G reduced the infectivity of both SIVcpzPttMB897Δvif and SIVcpzPtsTAN1Δvif to an even greater extent (Fig 2C and 2D). SIVcpzPttMB897 and SIVcpzPtsTAN1 Vifs are able to counteract all cpzA3s, but not all in cases to the same level, e.g. cpzA3F, the full viral infectivity (vector control) was restored, consistent with previous study [40] (Fig 2C and 2D). Taken together, these data indicate that chimpanzee A3s, such as cpzA3D and cpzA3G, can protect chimpanzees from infection with SIVs of rhesus macaques and African green monkeys.
Our data and a previous study indicate that chimpanzee A3s, especially cpzA3D, play an important role as a barrier to cross-species transmission of SIVs from monkeys to chimpanzees (Fig 2A and 2B) [40]. Next, we asked whether human A3s (hA3s) form a barrier to SIVcpz infection of humans. Thus, we analyzed the anti-SIV activity of human A3s by using the four SIV reporter systems. Similar to chimpanzee A3s, hA3C, D, F, G, and H (including hapI and hapII) inhibited SIVmacΔvif and SIVagmΔvif infections, and SIVmac Vif abolished most of these restrictions but was only weakly active against hA3D (Fig 3A). SIVagm Vif only significantly overcame the restriction of hA3C, hA3H hapI, and hA3H hapII (Fig 3B). hA3D, hA3F and hA3G displayed resistance to SIVagm Vif counteraction, indicating that these three factors may protect humans against infection by SIVagm (Fig 3B). Consistent with a previous study, hA3B strongly reduced the infectivity of SIVmac and SIVagm regardless of Vif [41]. However, hA3A showed only a low-level inhibition of SIVmac and SIVagm and this restriction was resistant to both SIVmac and SIVagm Vifs (Fig 3A and 3B). The expression of hA3s in transfected 293T cells was detected by immunoblotting (S2B and S2C Fig). In the absence of Vif, hA3D, F, and G reduced the infectivity of SIVcpz, while Vif proteins from both SIVcpzPttMB897 and SIVcpzPtsTAN1 antagonized these hA3s (Fig 3C and 3D). In contrast to the experiments with SIVmacΔvif and SIVagmΔvif, no antiviral activity of hA3C was seen against SIVcpzΔvif (Fig 3C and 3D). Interestingly, two human A3s (hA3B and hA3H hapII) showed strong inhibition of SIVcpz regardless of Vif expression (Fig 3C and 3D). While hA3H is expressed in primary CD4+ lymphocytes and has the ability to inhibit HIV-1 [26,35], hA3B is not found in HIV target cells [35]. Together our data indicate that hA3H hapII may block SIVcpz cross-species transmission to humans.
To characterize the interaction between SIVcpz and A3H in more detail, the incorporation of cpzA3H and hA3H hapII into SIVcpz viral particles was analyzed by immunoblotting. In the absence of Vif, both cpzA3H and hA3H hapII were encapsidated into SIVcpzPttMB897 and SIVcpzPtsTAN1 (S2D Fig). Vif from both SIVcpz strains reduced the cpzA3H protein level in the cell lysate by depletion and decreased the cpzA3H incorporation into viral particles (S2D Fig). In agreement with the infectivity data of SIVcpz with cpzA3H (Fig 2C and 2D), SIVcpzPtt Vif was more active against cpzA3H than SIVcpzPts Vif [40]. However, the steady-state expression and particle encapsidation of hA3H hapII did not change in the presence of SIVcpz Vif, which corresponds with hA3H’s hapII antiviral activity against wild-type SIVcpz (Fig 3C and 3D and S2D Fig). Furthermore, we investigated whether the cytidine deaminase activity is required for hA3H hapII inhibiting SIVcpz. We introduced the E56A mutation in the cytidine deaminase domain of hA3H hapII, which was previously reported to completely abolish the protein’s deaminase activity [42]. The E56A mutant of A3H lost significantly anti-viral activity compared with wild-type hA3H hapII, but remained a 10-fold inhibitory activity against SIVcpzΔvif (Fig 4A). Next we analyzed the presence of G-to-A mutations indicative of A3 deamination in the viral genome by amplifying a 700-bp fragment of the viral genome 12 h post-infection. Viruses prepared without co-expression of A3 showed no detectable G-to-A mutations. However, in the presence of hA3G, we found a hypermutation rate of around 2.8% in the SIVcpz genome (Fig 4B). Viruses made in the presence of hA3H hapII contained a mutation rate of around 0.9%, while hA3H hapII E56A did not edit the SIVcpz genome (Fig 4B). The sequence plots, confirmed that hA3G preferred GG motif (mutated G is underlined, which is CC in the deaminated minus strand), while hA3H hapII mutated a GA motif (TC in the minus strand) predominantly (Fig 4B), which is consistent to previous studies [42–44]. Taken together, these data indicate that hA3H hapII inhibits SIVcpz by both deaminase dependent and independent mechanisms.
To further characterize the level of anti-SIVcpz activity mediated by hA3H, different amounts (5–200 ng) of hA3H hapI or hA3H hapII expression plasmids were co-transfected with SIVcpzPttMB897 wild-type or Δvif reporter constructs and the viral infectivities were determined. The results indicate that the anti-SIVcpz activity of hA3H hapII increased with the dose of transfected hA3H hapII plasmid regardless of Vif expression (Fig 5A and 5B). Even a low level of hA3H hapII (5 ng) displayed around 10-fold inhibition of SIVcpzPttMB897Δvif and Vif was not able to overcome this restriction. We also found that hA3H hapI showed around 20-fold inhibition of SIVcpzPttMB897, when 200 ng hA3H hapI expression plasmid was transfected (Fig 5A). 100 ng hA3H hapI and 10 ng hA3H hapII plasmids displayed similar protein expression levels, and they showed similar strength of inhibition of SIVcpzPttMB897 (Red box in Fig 5A and 5B), which is consistent with previous studies [45,46]. These results indicate that the protein expression level of A3H is one of the key determinants for its antiviral activity.
Human A3H has seven haplotypes and several splice variants, and the A3H protein stability determines the antiviral activity [29–31]. In the absence of Vif, cpzA3H, hA3H hapII, hA3H hapV, hA3H hapVII, and four hA3H hapII splice variants (SV182, SV183, and SV200) strongly inhibited SIVcpzPttMB897 (Fig 5C). However, Vif only counteracted the restriction of cpzA3H and was inactive against all the tested hA3H variants (Fig 5C). Corresponding immunoblotting results of lysates of the transfected cells confirmed that SIVcpzPttMB897 Vif only reduced the protein level of cpzA3H and protein levels of the hA3Hs were not changed by Vif co-expression (Fig 5D).
To learn more about the strength of hA3H’s antiviral activity, the spreading replication of SIVcpz in human T cells (SupT11) that stably expressed hA3H hapII [47] was investigated. To facilitate replication of CCR5-tropic SIVcpz, we modified SupT11 cells to express human CCR5 (S3A and S3B Fig). The spreading replication was tested with full-length unmodified viruses (SIVcpzPttMB897, SIVcpzPttGab1, and SIVcpzPtsTAN1). SIVcpzPtsTAN1 did not replicate in the SupT11-hA3H hapII and SupT11-vector cells, regardless of the input of virus (1 ng reverse transcriptase (RT) activity or 50 ng RT) for the initial infection (Fig 5E and S3C Fig). Both SIVcpzPttMB897 and SIVcpzPttGab1 replicated efficiently in SupT11-vector cells, while no virus spreading was observed in SupT11-hA3H hapII cells (Fig 5E). These data indicate that hA3H hapII is a strong inhibitor of infection of SIVcpz in human T cells. We conclude, therefore, that stably expressed hA3H variants are Vif-resistant restriction factors of SIVcpz.
Both cpzA3H and hA3H hapII displayed strong anti-SIVcpz activity, while they had different sensitivities to SIVcpz Vif counteraction (Figs 2C, 2D, 3C and 3D). One recent study demonstrated that residue 97 of cpzA3H and hA3H hapII determines the sensitivity to HIV-1 clone NL4-3 Vif [48]. Thus, we tested whether residue 97 would also be important for SIVcpz Vif inhibition of A3H. The Q97K and K97Q mutations were introduced into cpzA3H and hA3H hapII, respectively. The results showed that cpzA3H Q97K and hA3H hapII K97Q retained their anti-SIVcpz activity in the absence of Vif (Fig 6A and 6C). While SIVcpzPttMB897 Vif almost fully overcame the inhibition of wild-type cpzA3H, it only partially antagonized cpzA3H Q97K and, similarly, SIVcpzPtsTAN1 Vif did not counteract cpzA3H Q97K (Fig 6A and 6C). Additionally, hA3H hapII showed resistance to SIVcpzPttMB897 Vif, but this resistance was partially lost when the K97Q mutation was introduced (Fig 6A). In contrast to SIVcpzPttMB897 Vif, both wild-type hA3H hapII and its K97Q mutant showed resistance to SIVcpzPtsTAN1 Vif (Fig 6C). Furthermore, we analyzed the protein expression level of these A3H mutants in the presence of SIVcpz Vifs. hA3H E121K was included as a control mutant that could not be degraded by HIV-1 Vif [48,49]. SIVcpzPttMB897 Vif slightly reduced the protein level of cpzA3H Q97K compared to the no-Vif control, which is consistent with the infectivity data (Fig 6A and 6B). SIVcpzPtsTAN1 Vif did not affect the expression of cpzA3H Q97K (Fig 6D). hA3H hapII K97Q was depleted by co-expression of SIVcpzPttMB897 Vif, while the presence of SIVcpzPtsTAN1 Vif did not affect hA3H protein levels (Fig 6B and 6D). We conclude that Vifs from SIVcpzPttMB897 and SIVcpzPtsTAN1 have distinct interaction properties with hA3H hapII.
To find out how diverse A3H is in chimpanzees, we analyzed the deep-sequencing reads from the recent Great Ape Genome Project [50]. We mapped reads to the hA3H region (hg19, chr22:39496284–39498576) and the exons of A3H were isolated. The coding regions of A3H from 61 chimpanzees (10 Pan troglodytes ellioti, Pte; 16 Pan troglodytes schweinfurthii, Pts; 22 Pan troglodytes troglodytes, Ptt; 13 Pan troglodytes verus, Ptv) were analyzed. We found four single-nucleotide polymorphisms (SNPs) of cpzA3H (nucleotide positions 50, 359, 402, and 481; Table 1, Fig 7A and S4 Fig). Two of them (SNP_50 and SNP_359) were only present in Ptv with an overall frequency of 6.5% and 9.8%, respectively. SNP_402 was only found in Pts with a frequency of 9%. However, SNP_481 was detected in Pte, Pts, and Ptt with a frequency of 8.2%, 19.6%, and 34.4%, respectively. The detailed SNP and zygosity information is described in Table 1. These four SNPs including the reference cpzA3H were named from haplotype I (hapI) to haplotype V (hapV) (Fig 7A). In addition, we performed a phylogenetic analysis of A3H from apes (rhesus macaque A3H was also included). The results showed that gibbon, rhesus macaque, and orangutan A3H were classified into one clade. Gorilla A3H formed a separate clade, and human and chimpanzee A3H were classified into two clades, respectively (Fig 7B). Bonobo A3H was classified into the clade of chimpanzee A3H.
The protein stability differs in human A3H haplotypes and it is one of the determinants of its antiviral activity [29–31]. Thus, the expression of five cpzA3H haplotypes in 293T cells was tested by immunoblotting. All cpzA3H haplotypes produce stable proteins and had similar expression levels (Fig 7C). Moreover, these five cpzA3H haplotypes displayed similar anti-SIVcpz activities and were all sensitive to Vifs from SIVcpz lineages (Fig 7D and 7E). These data indicate that the polymorphism of cpzA3H does not affect its protein stability or antiviral activity.
There have been four independent transmissions from different SIVcpz/gor strains to the human population, which caused HIV-1 groups M, N, O, and P, respectively [11,14]. Thus, we tested the sensitivity of cpzA3H and hA3H hapII to Vifs from several SIVcpz/HIV-1 lineages. The immunoblots of co-expressing cells indicated that cpzA3H was depleted by all the tested SIVcpz Vifs, and it was also depleted by Vifs from HIV-1 B-LAI (M group), N-116, and O-127, but was not degraded by HIV-1 F-1 Vif (Fig 8A). hA3H hapII was resistant to depletion of all SIVcpz Vifs tested, including SIVgor Vif (Fig 8A). However, HIV-1 B-LAI, F-1, and N-116 Vifs induced the degradation of hA3H hapII (Fig 8A). Unexpectedly, HIV-1 O-127 Vif, which protein expression was not detectable was inactive against hA3H hapII (Fig 8A).
By testing chimeras of SIVcpzPttMB897 and HIV-1 LAI Vif, we identified that the Vif N-terminal region (residues 40–70) is essential for hA3H hapII depletion (Fig 8B and 8C). A previous study described the importance of HIV-1 Vif residues F39 and H48 for antagonism of hA3H hapII [33]. F39 is present in SIVcpzPttMB897 Vif, but at the 48 position, an asparagine (N) is found (Fig 8D). However, introducing an N48H mutation (construct M1) in SIVcpzPttMB897 Vif did not promote degradation of hA3H hapII (Fig 8E). However, the local area of residue 48 of SIVcpzPttMB897 Vif was important as an additional mutation revealed that changing residues 47EN48 to 47PH48 (construct M2) facilitated hA3H hapII depletion (Fig 8E). Furthermore, a replication-competent SIVcpzPttMB897_EN-PH with this substitution showed spreading replication in hA3H hapII-containing SupT11 cells (Fig 8F).
A previous study showed that HIV-1 Vif from a homozygous hA3H haplotype II patient had greater activity against hA3H hapII compared to other laboratory HIV-1 Vifs, which correlated with the presence of four amino acid substitutions (60GDAK63 to 60EKGE63) [32]. This substitution was introduced into SIVcpzPttMB897 Vif and led to enhanced hA3H hapII depletion (Fig 8E, Vif M2 compared to M3 and M4). Based on a recent HIV-1 Vif-hA3H hapII co-structure model [49], the co-structure of SIVcpzPttMB897 Vif-hA3H hapII was modeled. From the structure, we found that residues 47EN48 and 60GDAK63 of SIVcpzPttMB897 Vif were in close contact with hA3H hapII (Fig 8G). Both regions are diverse in Vifs from distinct SIVcpz and HIV-1 lineages (Fig 8H).
Next, we tested the replication of SIVcpz in human PBMCs from donors with different hA3H genotypes. We identified three donors who were homozygous for A3H hapI, hapIV and hapII, respectively. The protein expression of A3H in stimulated PBMCs was detected by immunoblotting, demonstrating highest protein levels in the PBMCs of hapII, moderate levels in PBMCs of hapI and weak levels in cells of hapIV (Fig 9A). However, the A3G expression in these PBMC was identical (Fig 9A). The viral replication experiments indicated that SIVcpzPttMB897 replicated fastest in PBMCs from the donor with haplotype IV, and moderately in PBMCs from the donor with haplotype I (Fig 9B and 9C). However, the replication of SIVcpzPttMB897 was inhibited in PBMCs from donor with haplotype II (Fig 9B and 9C). In summary, we speculate that stable hA3H forms a barrier for zoonotic transmission of SIVcpz to humans and Vif adaptation to stable hA3H would be needed for high-level infection of humans with this haplotype (Fig 10).
SIVcpz originated from the cross-species transmission and recombination of three different SIVs [5,6]. After lentiviral transmission to a new host that differs in one or many A3 proteins, Vif adaptation is expected at the interface of both proteins [25,51]. In our study, all tested SIVcpz Vifs had the ability to counteract cpzA3Hs (Figs 2 and 7). Lucie Etienne et al. found that SIVrcm Vif acts like SIVcpz Vifs and can neutralize cpzA3H, while SIVmus Vif could not antagonize the restriction of cpzA3H [40]. Overcoming the restriction of cpzA3H may be one explanation for SIVcpz selectively acquiring the 5’ region (including vif) from SIVrcm during recombination, and acquiring the 3’ region (including vpu, env, and nef) from SIVgsn/SIVmus/SIVmon may have facilitated the counteraction of other restriction factors, such as Tetherin or Serinc3/5 [52–54]. Here, we also found that cpzA3D, F, and G were resistant to SIVagm Vif and similarly, cpzA3D was resistant to SIVmac Vif, confirming a previous report [40]. This observation indicates that cpzA3D and cpzA3G can protect chimpanzees from infection with SIVs of rhesus macaques and African green monkeys. On the other hand, SIVcpz Vifs could counteract all the tested cpzA3s. However, cpzA3F showed a moderate level of resistance to degradation induced by SIVcpz Vif (Fig 2C and 2D), possible suggesting that cpzA3F may provide some repression of SIVcpz infection. Human and chimpanzee A3D, F, and G display a similar sensitivity to SIVcpz Vif, indicating that the inhibitory activity against cpzA3s by SIVcpz may be a prerequisite for the cross-species transmission of SIVcpz to the human population.
Here, we found that hA3C and hA3H hapI display a strong restriction against SIVmacΔvif and SIVagmΔvif; however, no antiviral activity was observed against SIVcpzΔvif (Fig 3) or HIV-1Δvif [26,41,55]. These data suggest that the viral sensitivity to hA3C and hA3H hapI was lost in the evolution of SIV lineages and not during the evolution of HIV-1. We cannot determine whether this happens during the creation of SIVcpz due to the lack of information regarding the antiviral activity of hA3C and hA3H hapI against SIVrcm/SIVgsn/SIVmus/SIVmon. We speculate that some SIVs similar to HIV-1 have the ability to escape hA3C and hA3H hapI restriction by a Vif-independent mechanism [55].
cpzA3H appears to be much less polymorphic than hA3H. However, A3F and A3G in chimpanzee are more diverse than the human orthologs [40]. Although our chimpanzee sample number was limited (61 chimpanzees), the results suggest that cpzA3H is relatively conserved among chimpanzees. Residues 15 and 105 of hA3H determine the protein stability and anti-viral activity [28]. However, no variability was identified at these two positions in cpzA3H, which is in agreement with the comparable protein stability and anti-viral activity of the currently recognized five cpzA3H haplotypes (Fig 7). Vifs of different SIVcpz isolates degrade all haplotypes of cpzA3H indicating that cpzA3H is not a restriction factor for inter-subspecies transmission of SIVcpz. Compared to cpzA3H, human A3H is more diverse and includes seven haplotypes and several splice variants [28,30,31]. In our study, the stably expressed hA3H haplotypes were identified as Vif-resistant inhibitors against SIVcpz, indicating that these active hA3Hs are strong barriers to prevent SIVcpz infection of humans. After the zoonotic transmission of SIVcpz to humans expressing unstable A3H haplotypes, the very early human-to-human transmission was likely to be severely affected by humans expressing the A3H haplotypes with a stable protein (Figs 9 and 10). A possible mutation that would enhance SIVcpz Vif adaptation was investigated by replacing residues 47EN48 of SIVcpzPttMB897 Vif with 47PH48 (Fig 8). It is possible—but unlikely—that there are currently not identified viruses circulating in chimpanzees with vif genes encoding 47PH48 residues enhancing SIVcpz cross-species transmission to humans. In fact, the 47PH48 motif is also found in HIV-1 patients who harbor hA3H hapII [32,34]. The frequency of active hA3H varies significantly between populations, with the highest frequency in Africans (around 50% harbor stable A3H) [28,30]. This observation may be the result of a selective sweep caused by exposure to a retrovirus such as SIV or HTLV or other A3H-sensitive pathogens [56,57]. Several previous studies described a positive and balancing selection of human and chimpanzee MHC loci, caused by HIV-1/SIVcpz infections [56,58–60].
In addition to hA3H hapII, human tetherin is also a strong barrier against SIVcpz transmission to humans. SIVcpz Nef recognizes the cytoplasmic domain of chimpanzee tetherin and inhibits its restriction, but it cannot overcome the restriction of human tetherin due to a deletion in this domain [54]. However, the virus adapts to this restriction by regaining Vpu-mediated inhibition of tetherin after transmission of SIVcpz to humans [54]. In fact, other unknown restriction factors may exist to control the cross-transmission of SIVcpz to humans. For example, a recent study found that introducing a M30R/K mutation in the Gag matrix could enhance SIVcpz replication fitness in human tonsil explant cultures [38].
Overall, our study suggests that the stable active human A3Hs can protect humans against the spillover of SIVcpz, and SIVcpz cross-species transmission to humans may have started in those that harbored unstable A3H proteins.
Chimpanzee APOBEC3 (A3) expression plasmids (A3D, A3F, A3G and A3H) were provided by Michael Emerman [40], chimpanzee A3C plasmid was described recently [61]. Human A3s (A3A-A3H) were expressed by PTR600 vector with a carboxyl-terminal triple hemaggutinin (HA) tag [33]. Human A3H haplotype V, VII, splice variants and E56A of haplotype II expression plasmids with a carboxyl-terminal flag tag were provided by Viviana Simon [31]. Human A3H haplotype II with an N terminal HA tag was re-cloned into PTR600 vector by using standard PCR. All human and chimpanzee A3H mutants were generated by site direct mutagenesis and confirmed by sequencing. The MLV packaging construct pHIT60 was kindly provided by Jonathan Stoye, which encodes the gag-pol of MoMLV [62]. The Plasmid of pBABE.CCR5 that encodes human CCR5 was obtained from NIH AIDSREPOSITORY [63]. SIVmac-Luc (R-E-), SIVmac-Luc (R-E-)Δvif and SIVagm-Luc (R-E-) and SIVagm-Luc (R-E-)Δvif were provided by N. R. Landau [64]. The replication competent SIVcpzPtt clones MB897, EK505, Gab1 were kindly provided by Frank Kirchhoff [38,65]. SIVcpzPts clones TAN1.910 and TAN2.69 and SIVgor clone CP2139 were obtained from NIH AIDSREPOSITORY [10,66]. To generate the Nanoluciferase reporter virus of SIVcpzPttMB897, the nef gene was replaced (the first 7 amino acids of Nef remained) by nanoluciferase gene by overlapping PCR using NheI and XhoI restriction sites. Additionally, two stop codons were inserted amino-terminal of Vif (amino acid position 40 and 44) by overlapping extension PCR using PshAI and NheI restriction sites. The same method was performed to create nanoluciferase reporter virus of SIVcpzPtsTAN1, and the restriction sites are shown in S1 Fig. Simply, the nef gene was replaced (the first 7 amino acids of Nef remained) by nanoluciferase gene by overlapping PCR using AclIII and XbaI restriction sites. Additionally, two stop codons were inserted at the amino-terminal of Vif (amino acid position 40 and 44) by overlapping extension PCR using PshAI and AclI restriction sites. All constructs were verified by sequencing analysis. To generate the SIV Vif expression plasmids, Vif fragments from the following molecular clones: SIVcpzPtt EK505 (DQ373065), Gab1 (X52154), MB897 (EF535994) and SIVcpzPts TAN1 (AF447763), TAN2 (DQ374657) and SIVgor CP2139 (FJ424866) were amplified and inserted into pCRV1 by EcoRI and NotI. Vif expression plasmids of HIV-1 LAI, F-1, N-116 and O-127 were provided by Viviana Simon [33,61]. All SIVcpzPttMB897 Vif mutants were generated by overlapping PCR and cloned into pCRV1 without any tag, verified by sequencing.
HEK293T (293T, ATCC CRL-3216) cells were maintained in Dulbecco’s high-glucose modified Eagle’s medium (DMEM, Biochrom, Berlin, Germany) supplemented with 10% fetal bovine serum (FBS), 2 mM L-glutamine, penicillin (100 U/ml), and streptomycin (100 μg/ml). SupT11 cells containing empty control and hA3H hapII were kindly provided by Reuben S. Harris and cultured in RPMI supplemented with 10% fetal bovine serum (FBS), 2 mM L-glutamine, penicillin (100 U/ml), and streptomycin (100 μg/ml) [47]. SupT11 cells with expression of hCCR5 were generated by MLV transduction. Simply, 1x106 SupT11 cells were transduced by MLV vector (produced by transfecting pBABE.CCR5, pHIT60 and VSV-G expression plasmid into 293T cells). 3 days after transduction, the SupT11 cells were selected for 3 weeks by using 1 μg/ml puromycin. For producing the single round infection of SIV reporter virus, 3×105 293T cells in 24-well plates were co-transfected with 300 ng SIVmac-Luc (R-E-), or SIVagm-Luc, or SIVcpzPttMB897-NLuc, or SIVcpzPtsTAN1-NLuc; or the corresponding delta Vif versions, 30 ng human A3s or 200 ng chimpanzee A3s expression plasmids and 50 ng VSV-G (pMD.G), and pcDNA3.1(+) (Thermo Fisher Scientific) was used instead of A3 expression plasmids. Human A3s were expressed in plasmid PTR600, while chimpanzee A3s were expressed in plasmid pcDNA3.1(+). 30 ng of PTR600-human A3s constructs had comparable expression levels with 200 ng pcDNA3.1(+)-chimpanzee A3s plasmids. Transfections were performed by using Lipofectamine LTX (Thermo Fisher Scientific) according to manufacturer’s instruction. The viral supernatants were collected 48 h post transfection. The reverse transcriptase (RT) activities of viruses were quantified by using the Cavidi HS lenti RT kit (Cavidi Tech, Uppsala, Sweden). For SIVmac and SIVagm infections, 5×104 293T cells were seeded in 96-well plates one day before transduction, and 50 pg RT of viruses were used for infection. After 48 h, firefly luciferase activity was measured with Steady-Glo Luciferase system (Promega) according to the manufacturer’s instructions on a MicroLumat Plus luminometer (Berthold Detection Systems, Pforzheim, Germany). For SIVcpz-NLuc, we observed high nanoluciferase enzyme activity in cell supernatant of transfected cells. 293T cells in 96-well plates were infected with 20 pg of SIVcpzPttMB897-NLuc or SIVcpzPtsTAN1-NLuc. To eliminate the effect of contaminating nanoluciferase in the supernatant of virus producer cells, we changed the medium 8 h post infection. 48 h after transduction, the cells were carefully washed by PBS once, and the nanoluciferase activity was measured with Nano-Glo Luciferase system (Promega) on a MicroLumat Plus luminometer (Berthold Detection Systems). Each sample was analyzed in triplicates; the error bar of each triplicate was shown. Infections in which the VSV-G glycoprotein was omitted served as control for nanoluciferase enzyme background enzyme activity.
1 x 106 293T cells were infected with DNase I (Thermo Fisher, Germany) treated SIVcpzΔVif-Nluc produced in 293T cells together with hA3G, hA3H hapII, hA3H hapII E56A or pcDNA3.1(+). At 12 h post-infection, cells were washed with PBS, and DNA was isolated using a DNasy DNA insolation kit (Qiagen, Germany). A 700-bp fragment of the SIVcpz-Nluc (200-bp C terminal of env plus 500-bp nanoluciferase gene) was amplified using DreamTaq DNA polymerase (Thermo Fisher, Germany) with primers: 5’-attctccagtattggggacaagag-3’ and 5’-ttacgccagaatgcgttcgcac-3’. The PCR parameters were: 95°C for 5 min; 30 cycles with 88°C for 30 s, 57°C for 30 s, 72°C for 1min; 10 min at 72°C. PCR products were cloned using CloneJET PCR cloning kit (Thermo Scientific). Seven to ten clones were sequenced for each sample. A3 induced hypermutations were analyzed with the Hypermute online tool (http://www.hiv.lanl.gov/content/sequence/HYPERMUT/hypermut.html). The overall mutation rate was calculated by using the total number of G-A mutations divided by the total analyzed nucleotides.
To analyze CD4 and CCR5 expression level of SupT11 cell lines, 5×105 cells were stained by α-hCD4 PE mouse IgG1k (Dako, Hamburg, Germany) and α-hCCR5 FITC (BD Bioscience, Heidelberg, Germany) separately according to the manufacturer’s instruction. The mouse IgG1/RPE isopeptidase was used as negative antibody control for CD4 staining. The measurement was carried out by BD FACSanto (BD Bioscience). Data analysis was done with the Software FlowJo version 7.6 (FlowJo, Ashland, USA).
Buffy-coats obtained from anonymous blood donors were obtained from University Hospital Düsseldorf blood bank. Whole blood was obtained from healthy and de-identified African donors that signed an informed consent. The research has been approved by the Ethics Committee of the Medical Faculty of the Heinrich-Heine-University Düsseldorf (Reference No 4767R - 2014072657) and performed according to the principles expressed in the Declaration of Helsinki.
Cellular RNA from PHA stimulated human PBMCs was isolated by using QIAGEN RNA extraction kit (Qiagen). 1 μg of total cellular RNA was used for reverse transcription with the RevertAid H Minus First Strand cDNA synthesis kit (Thermo Scientific). Human A3H cDNA was amplified with Q5 High-Fidelity DNA Polymerase (New England Biolabs) using primers: 5’-atggctctgttaacagccgaaacattcc-3’ and 5’-ggactgctttatcctgtcaagccgtcgc-3’. PCR products were cloned using CloneJET PCR cloning kit (Thermo Scientific). Six to ten clones were sequenced for each donor.
To produce SIVcpz, 1×106 293T cells in 6-well plate were transfected with 2 μg SIVcpz molecular clone plasmids (SIVcpzPtsTan1, SIVcpzPttMB897 and SIVcpzPttGab1). 2 days after transfection, the viral supernatants were collected and centrifuged at 12,000 rpm for 10 min to remove cell debris. Then the viral supernatants were concentrated through 20% sucrose cushion at 14,800 rpm 4 h, followed by resuspension in RPMI. The viral stock was quantified by using the Cavidi HS lenti RT kit (Cavidi Tech, Uppsala, Sweden). 5×105 cells of each SupT11 cell lines (SupT11-vector-hCCR5 and SupT11-hA3H hapII-hCCR5) were infected with 1 ng or 5 ng RT activity of SIVcpz in a 24-well plate (in 500 μl) and cells were washed with PBS 1 day post-infection. Each second day, 200 μl supernatant was collected, clarified, and stored at 80°C, and cultures were supplemented with fresh media. The replications were performed in two independent experiments, and each infection was performed in duplicates.
3 x 105 PHA stimulated PBMC from three donors were infected overnight with SIVcpzPttMB897 representing either 1 ng RT activity or 5 ng RT activity in the presence of 30 U/ml Interleukin-2 (IL-2) in 96-well round-bottom plates (total volume 200 μl). After infection, cells were washed three times and maintained in complete RPMI with 30 U/ml of IL-2 for 15 days. 100 μl culture supernatant was collected every 2–3 days, and cultures were supplemented with fresh media. The RT activities of viruses in culture supernatant were quantified by using the Cavidi HS lenti RT kit (Cavidi Tech, Uppsala, Sweden).
A total of 3×105 293T cells in 24-well plates were co-transfected with 200 ng chimpanzee A3H expression plasmid or 50 ng hA3H haplotype II in PTR600 expression vector and 300 ng pCRV1 Vif expression plasmids, pcDNA3.1(+) (Thermo Fisher Scientific) was used to fill up the total transfected plasmid DNA to 500 ng. Transfections were performed by using Lipofectamine LTX (Thermo Fisher Scientific). 48 h post transfection, cells were lysed and clarified by 14,000 rpm/30 mins centrifugation. The expression of A3H and Vif were analyzed by immunoblots.
Transfected 293T cells were lysed in radioimmunoprecipitation assay (RIPA) buffer (25 mM Tris-HCl [pH 8.0], 137 mM NaCl, 1% NP-40, 1% glycerol, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate [SDS], 2 mM EDTA, and protease inhibitor cocktail set III [Calbiochem, Darmstadt, Germany]). To pellet virions, culture supernatant were centrifuged at 12,000 rpm for 10 min followed by centrifugation through 20% sucrose cushion at 14,500 rpm 4 h and resuspended in RIPA buffer, boiled at 95⁰C for 5 min with Roti load reducing loading buffer (Carl Roth, Karlsruhe, Germany) and resolved on a SDS-PAGE gel. The expression of A3s and SIV/HIV Vifs were detected by mouse anti-hemagglutinin (anti-HA) antibody (1:7,500 dilution, MMS-101P; Covance, Münster, Germany), rabbit anti-HA antibody (1:1,000 dilution, C29F4, cat. 3724, Cell Signaling, USA) and rabbit anti-Vif polyclonal antibody (1:1,000 dilution, NIH AIDSREAGENTS, cat. 2221) [67]; tubulin and SIVcpzPttMB897 capsid protein was detected using mouse anti-α-tubulin antibody (1:4,000 dilution, clone B5-1-2; Sigma-Aldrich, Taufkirchen, Germany) and mouse anti-capsid p24/p27 MAb AG3.0 (1:50 dilution) separately [68], followed by horseradish peroxidase-conjugated rabbit anti-mouse or donkey anti-rabbit antibodies (α-mouse or rabbit-IgG-HRP; GE Healthcare, Munich, Germany), and developed with ECL chemiluminescence reagents (GE Healthcare). The expression of A3H in SupT11 cell lines was detected by using anti-hA3H (1:1,000) antibody [35] followed by horseradish peroxidase-conjugated rabbit anti-mouse and developed with ECL chemiluminescence reagents. 8 x 106 human PBMCs from three donors were lysed in 100 μl RIPA buffer with protease inhibitor cocktail set III [Calbiochem, Darmstadt, Germany]). The expression of A3G, A3H and tubulin were detected by using anti-hA3H (1:1,000) [35], anti-hA3G (1:10,000, NIH AIDSREAGENTS, cat. 9906) [69] and anti-tubulin (1:4,000 dilution, clone B5-1-2; Sigma-Aldrich, Taufkirchen, Germany) antibodies, respectively.
The primate A3H sequences were obtained from GenBank, the accession numbers are: EU861357, EU861358, EU861359, EU861360, EU861361, DQ408606 and DQ507277. CpzA3H SNPs were described in this study. The A3H sequences were aligned using the ClustalW in Mega 7 software. The phylogenetic analysis was performed in Mega 7 by using bootstrap neighbor joining method. Test parameters were estimated using 500 bootstrap replicates.
To analyze the interaction surface between SIVcpz Vif and hA3H hapII, the structure of SIVcpz Vif was modeled using HIV-1 Vif (4N9F) [70] as template by using SWISS-MODEL online server (http://www.swissmodel.expasy.org/). The recent crystal structure of hA3H hapII (6B0B) was also used to model the structure of cpzA3H. The SIVcpz Vif and hA3H hapII co-structure was modeled based on the recent HIV-1 Vif-A3H interaction surface analysis [49]. The graphical visualization was constructed using PyMOL (PyMOL Molecular Graphics System, version 1.5.0.4; Schrödinger, Portland, OR).
Data are represented as the mean with SD in all bar diagrams. Statistically significant differences between two groups were analyzed using the unpaired Student’s t-test with GraphPad Prism version 5 (GraphPad software, San Diego, CA, USA). A minimum p value of 0.05 was considered as statistically significant: P value < 0.001 extremely significant (***), 0.001 to 0.01 very significant (**), 0.01 to 0.05 significant (*), >0.05 not significant (ns).
|
10.1371/journal.ppat.1000269 | Genomic Analysis Reveals a Potential Role for Cell Cycle Perturbation in HCV-Mediated Apoptosis of Cultured Hepatocytes | The mechanisms of liver injury associated with chronic HCV infection, as well as the individual roles of both viral and host factors, are not clearly defined. However, it is becoming increasingly clear that direct cytopathic effects, in addition to immune-mediated processes, play an important role in liver injury. Gene expression profiling during multiple time-points of acute HCV infection of cultured Huh-7.5 cells was performed to gain insight into the cellular mechanism of HCV-associated cytopathic effect. Maximal induction of cell-death–related genes and appearance of activated caspase-3 in HCV-infected cells coincided with peak viral replication, suggesting a link between viral load and apoptosis. Gene ontology analysis revealed that many of the cell-death genes function to induce apoptosis in response to cell cycle arrest. Labeling of dividing cells in culture followed by flow cytometry also demonstrated the presence of significantly fewer cells in S-phase in HCV-infected relative to mock cultures, suggesting HCV infection is associated with delayed cell cycle progression. Regulation of numerous genes involved in anti-oxidative stress response and TGF-β1 signaling suggest these as possible causes of delayed cell cycle progression. Significantly, a subset of cell-death genes regulated during in vitro HCV infection was similarly regulated specifically in liver tissue from a cohort of HCV-infected liver transplant patients with rapidly progressive fibrosis. Collectively, these data suggest that HCV mediates direct cytopathic effects through deregulation of the cell cycle and that this process may contribute to liver disease progression. This in vitro system could be utilized to further define the cellular mechanism of this perturbation.
| Chronic HCV infection is associated with progressive liver injury and subsequent development of fibrosis/cirrhosis. The cellular mechanisms by which HCV replication, and subsequent virus–host interactions, may mediate liver injury are unclear. Microarray experiments were performed to characterize the host transcriptional response to HCV infection of cultured hepatocytes in an attempt to gain insight into the mechanism of HCV-associated cell death. Analysis of the gene expression data revealed that many differentially regulated genes function to induce apoptosis in response to cell cycle arrest, possibly in response to DNA damage and oxidative stress. Labeling of dividing cells in culture followed by flow cytometry also demonstrated the presence of significantly fewer cells in S-phase in HCV-infected cultures relative to mock cultures, suggesting HCV infection is associated with delayed cell cycle progression. Finally, many of the cell-death–related genes whose expression changes in response to HCV infection of cultured hepatocytes were also differentially regulated in liver tissue from HCV-infected patients with histological evidence of fibrosis. In summary, HCV may mediate direct cytopathic effects through perturbation of the cell cycle which potentially contributes to liver disease progression.
| Hepatitis C virus (HCV), a member of the Flaviviridae family, is a blood-borne pathogen which currently infects approximately 170 million people worldwide. Exposure to HCV typically results in a persistent infection and approximately 30% of chronically infected patients will develop progressive liver disease including fibrosis, cirrhosis and hepatocellular carcinoma (HCC) [1]. The majority of pathology associated with chronic infection is believed to occur via a HCV-specific cell-mediated immune response [2]. However, in light of this, it is somewhat perplexing that liver disease progression is accelerated in immuno-compromised individuals. Specifically, HCV/HIV-coinfected patients and liver transplant patients receiving immuno-suppressive drugs tend to develop fibrosis/cirrhosis at a much faster rate than immuno-competent individuals [3]. A recent study found that HCV-specific CD8 T cells were actually associated with areas of low hepatocellular apoptosis and weak fibrosis. It is thought that these cells are protective of liver damage through production of IL-10 [4]. Furthermore, characterization of the host response to HCV infection in the SCID-Alb/uPA mouse model demonstrated histological evidence of hepatocyte apoptosis in a manner similar to that observed during acute HCV infection in patients [5]. HCV infection in this model was associated with perturbations in cellular pathways, including lipid metabolism and oxidative stress, which have the potential to be cytopathic. The inability of these animals to generate a virus-specific immune response raises the intriguing possibility that HCV replication is capable of directly mediating hepatocyte apoptosis.
It is now thought that both direct cytopathic effects and immune-mediated processes likely play a role in HCV-associated liver injury [3]. The cellular mechanisms by which HCV replication, and subsequent virus-host interactions, may mediate liver injury are unclear. Progress in this area has been hindered by the lack of appropriate model systems in which to investigate the role of viral factors in liver disease progression. Currently, studies focused on defining the mechanisms of HCV-associated liver injury are primarily restricted to limited analysis of patient samples, including liver biopsy tissue. While such studies have provided significant insight into the role of steatosis, oxidative stress and death-receptor signaling in liver disease, there are obvious limitations with respect to conducting more mechanistic studies, in particular the role of viral proteins.
There is a wealth of literature describing experiments in which one more HCV proteins are over expressed in cultured hepatocytes. The results indicate a wide range of, and often conflicting, effects of viral protein expression on cellular functions, including apoptosis (Reviewed in Fischer et al, 2007). Core protein, in particular, has been reported to have both pro- and anti-apoptotic effects on death-ligand mediated hepatocyte apoptosis, including TNF-α, CD95Ligand and TRAIL-induced apoptosis, via a variety of mechanisms. The HCV envelope protein E2 has been found to both inhibit TRAIL-induced apoptosis and also to induce mitochondria-related/caspase-dependent apoptosis in the same hepatoma cell line. Perturbations of apoptotic pathways have also been demonstrated with the non-structural proteins. The NS3 protease inhibits pro-apoptotic RIG-I signaling via cleavage of the adaptor protein Cardif and also induces apoptosis of hepatocytes via caspase-8. NS5A has been found to inhibit apoptosis through multiple mechanisms, including sequestering of p53, activation of NFkB, increased expression of bcl-XL and p21 as well as activation of the P13-kinase-AKt/PKB survival pathway. While intriguing, many of these studies involve the expression of a single HCV protein, often at very high levels which do not accurately represent those seen in naturally infected livers. These experiments also fail to study the impact of potentially crucial interactions between the different HCV proteins.
A significant breakthrough in HCV research was achieved by the discovery of a specific HCV strain that efficiently infects and replicates in the cultured hepatoma cell line Huh-7.5 [6],[7],[8],[9]. This strain, termed JFH-1, was isolated from a Japanese patient who suffered fulminant hepatitis following exposure to the virus [6],[7]. Subsequent inoculation of clonal JFH-1 into chimpanzees and SCID-Alb/uPA mice resulted in productive infections in the absence of fulminant hepatitis, suggesting that the host response to infection played a key role in the severe form of hepatitis observed in this patient [8]. This model system provides the opportunity to study the impact of viral protein expression and replication on host cell function during a productive HCV infection and to potentially investigate the role of viral factors in liver injury. In the current study, microarray experiments were performed to characterize the host transcriptional response to HCV infection in an attempt to gain insight into the mechanism of HCV-associated cell death. Both the presence of activated caspase-3 and induction of cell death-related genes indicated that HCV infection was associated with a direct cytopathic effect. Gene ontology analysis suggests a role of cell cycle perturbation, possibly in response to oxidative stress and/or TGF-β1 signaling, in HCV-mediated apoptosis.
To study the host response to infection during the early phase of acute infection, Huh-7.5 cells were infected at a relatively high MOI (1–2 virions/cell) with HCV genotype 2a chimeric virus, J6/JFH (HCVcc) [7]. The virus used to infect cells was a pool of cell culture adapted virions generated by multiple passages over naïve Huh 7.5 cells (see Materials and Methods). For infection controls, cells were inoculated with either UV-inactivated HCVcc (to distinguish effects due to virus binding and virus replication) or conditioned media (mock). Conditioned media was used for the mock as it has been shown that factors present in the media of cultured cells can induce transcriptional changes (Walters, unpublished data). Cells were incubated with HCVcc for approximately 8 hrs, after which the cells were washed and fresh media added. Following infection, HCV (+) cells were visualized using an anti-NS5A antibody. As shown in Figure 1A, the majority of the cells expressed viral antigen by 48 hrs post-infection and continue to do so for the remainder of the study. No HCV RNA or viral protein expression was detected in cells exposed to UV-inactivated HCVcc (data not shown). Samples were harvested at 24, 48, 72, 96, and 120 hours post-infection and cellular RNA isolated for measuring intracellular HCV RNA levels and for global gene expression profiling.
Similar to what has been reported previously, a cytopathic effect was observed in the cultures of cells infected with HCVcc starting around 72 hrs post-infection (Figure 1B). This effect was not observed in the cells exposed to either UV-inactivated HCVcc or conditioned media, indicating that it is induced by active viral replication. The cytopathic effect became more prominent at 96 hrs and appeared to include the majority of the cells by 120 hrs post-infection (Figure 1B). Immuno-histochemistry specific for cleaved caspase-3 demonstrated activation of a terminal pathway involved in apoptosis that appears to cause the cytopathic effect in culture. As shown in Figure 1C, cleaved caspase-3 was present in cells infected with HCVcc beginning around 72 hrs, suggesting that the mechanism of cell death was apoptosis. It was not observed in cells exposed to either conditioned media (data not shown) or UV-inactivated virus. The initial presence of activated caspase-3 also coincided with peak levels of intracellular viral RNA (Figure 2B), suggesting a causative link between the level of HCV replication and cell death.
Microarray experiments were performed to characterize the host transcriptional response to HCV infection in an attempt to gain insight into the mechanism of HCV-associated cell death. For these experiments, mRNA samples isolated from cells exposed to either HCVcc or UV-inactivated HCVcc were compared to mRNA isolated from time-matched mock-treated cells. Figure 2A shows the global gene expression profiles of cells infected with UV-inactivated HCVcc and HCVcc at 24, 48, 72, 96, and 120 hrs post-infection. Similar to what was observed in previous genomic studies using the chimpanzee and SCID-Alb-uPA mouse models, the overall effect of HCV infection on cellular gene expression was subtle in the early phases of infection, with less than 50 differentially regulated genes at 24 hrs post-infection. Overall, 860 genes showed a 2-fold or higher change in expression (P value≤0.05) in at least one experiment (Figure 2A). The primary sequence name and fold-change of these genes are shown in Table S1. In contrast, cells exposed to the UV-inactivated virus showed very little, if any, regulation of genes (2-fold change, P value≤0.05) throughout the time-course, indicating that the process of virus attachment and entry into cells does not significantly impact host cell gene expression. A similar lack of differential regulation was observed in global transcriptional profiling of cells containing the HCV full-length replicon (data not shown). As shown in Figure 2B, there was a clear association between intracellular HCV RNA levels and number of differentially expressed genes, with the maximum regulation of cellular genes coinciding with peak intracellular HCV RNA levels (72 hrs post-infection). The reason for the decreasing HCV RNA levels following 72 hrs is unclear but may be related to a decrease in the number of cells capable of producing high levels of virus. It is interesting to note that the majority of transcriptional changes involve increased expression of host genes. Few differentially expressed genes showed decreased expression in the HCV-infected relative to the mock-infected cultures (Figure 2A), although the significance of this finding is unclear.
Gene ontology analysis was used to identify the cellular processes represented by the changes in steady-state abundance of transcripts associated with HCV infection. Notably, many of the differentially expressed genes belong to functional categories of cell death, cell cycle and cell growth/proliferation. Indeed, cell death genes comprised approximately half of all annotated differentially regulated genes at each time-point, with the exception of 24 hrs post-infection which showed negligible regulation of cellular genes. Figure S1 demonstrates the expression profiles of 118 genes associated with cell death in HCV-infected cells. Differential expression occurred beginning at 48 hrs post-infection at which time there was no significant visual evidence of cytopathic effect or apoptosis. Maximum differential expression of cell death genes occurred at 72 hrs post-infection whereas the level of apoptosis, as measured by caspase-3 cleavage, continued to increase until 120 hrs post-infection.
Both the differential expression of cell death-related genes and detection of cleaved caspase-3 in HCV-infected cells indicated that the observed cytopathic effect is apoptosis. Ingenuity Pathway Analysis identified a large number of cell death-related genes that function in cell cycle checkpoint/arrest, suggesting a potential role of cell cycle perturbation in apoptosis of infected cells (Figure 3). Consistent with this, a significant number of genes were identified that are associated with the DNA damage/oxidative stress response, many of which belong to the NRF2-mediated oxidative stress response pathway. This suggests that HCV replication is associated with generation of reactive oxygen species (ROS). Interestingly, the expression of two members of this pathway (CAT and EPHX1) was decreased, suggesting the ability of the cells to deal with excess ROS may be impaired. Apoptosis associated with cell cycle arrest is thought to be mediated through the mitochondria and p53 pathway [10]. In support of this, there was regulation of genes that have been linked to cytochrome c release from mitochondria (BBC3, BIK, BMF, PMAIP1, GSN, HRK). Many of these genes, along with others (DAPK3, CASP4, RIPK2), are also known to specifically regulate caspase activation. Interestingly, Ingenuity Pathway Analysis revealed that p53 signaling was significantly effected during HCV infection and this could provide the important link between oxidative stress/DNA damage, cell cycle arrest and apoptosis. Indeed, many of the differentially expressed genes associated with HCV infection that are involved in cell cycle arrest (TP53INP1, GADD45A, GADD45B, KLF6, UHRF1) and DNA damage/oxidative stress response (PMAIP1, ATF3, BBC3, FOXO3A, NOXA, CAT, UHRF1) regulate apoptosis via interaction with p53.
The functional categories cell cycle and cell growth/proliferation were also significantly enriched among genes showing differential expression during HCV infection. However, the majority of the differentially expressed genes associated with cell cycle regulation were involved with cell cycle checkpoint/arrest and subsequent induction of apoptosis, rather than actual progression through the cell cycle (Figure 4A). This likely explains the significant overlap between cell cycle genes and those associated with cell death as described above. Many of the genes associated with checkpoint/arrest involved the G1/S phase transition, suggesting that this checkpoint is the main area of cell cycle regulation by HCV replication. Similar to what was observed with apoptosis-associated genes, many cell cycle genes function to induce cell arrest in response to DNA damage and cellular stress. Genes previously identified as transcriptional targets of p53-induced growth arrest and apoptosis (BBC3, PMAIP1, AURKB, MKi67, RRM2, MCM4, and MCM6) were also differentially regulated in HCV-infected cells. Again this suggests that perturbations in the p53 signaling pathway play a key role in perturbation of cell cycle and induction of apoptosis during HCV infection [11],[12]. Increased expression of genes encoding proteins which function to either decrease p53 levels or serve as a protective effect against p53-dependent apoptosis (TYMS, JUND, and UBD) suggests the cell is attempting to counteract the activation of the p53 signaling pathway. Alternatively, the cell may be trying to undergo apoptosis and the virus is trying to counteract the process. A much smaller set of genes associated with mitosis and cell cycle progression (positive regulators of cell proliferation) were also differentially regulated. Interestingly, the expression of MK167, Mcm4 and Mcm6, common markers of cell proliferation, are decreased during HCV infection. This is consistent with the observation that proliferation of HCV-infected Huh-7.5 cells was slower than naïve cells and provides further support that HCV-infection delays cell cycle progression (data not shown). Quantitative PCR analysis of a number of the genes shown in Figure 4A demonstrated a good correlation with the gene expression data from the microarrays (Figure 4B), although the ratios calculated from RT-PCR generally exceeded those obtained using microarrays.
Flow cytometry analysis was performed to determine if HCV infection was associated with alterations in the cell cycle. Specifically, the number of cells progressing through S-phase of the cell cycle was determined by pulse-labeling the cells with the nucleoside EdU (5-ethynyl-2′-deoxyuridine), followed by a copper-catalyzed covalent reaction to fluorescently detect the DNA-incorporated nucleoside analog (see Materials and Methods). At 72-hours post-infection, approximately 21% of the HCV-infected population showed evidence of EdU incorporation/DNA synthesis, compared to 61% of the UV-inactivated control cells (Figure 4C). This significant reduction of labeled cells in the HCV-infected population suggests reduced cellular proliferation, or a block in cell cycle progression prior to S-phase, due to the presence of the virus. Comparable results were obtained when performing the analysis by immuno-fluorescence on fixed/attached cells (Figure 4D). Reduced cell proliferation due to HCV could also be seen at earlier time-points (24 and 48 hours post-infection), but the difference was not as dramatic, although it was progressive (data not shown).
HCV infection was also associated with differential expression of genes associated with cytokine/growth factor signaling, with the highest induction again at 72 hours post-infection (Figure 5). These genes included pro-inflammatory cytokines which are chemotactic for specific immune cells (e.g. CCL4-macrophages, CXCL1-neutrophils, IL8/CXCL2/CXCL3-PMNs, CX3CL1-macrophages, NK, lymphocytes, and CCL20-dendritic/lymphocytes). Of particular interest, some of the cytokines induced by HCV infection in vitro (including CCL4, CXCL1, IL32, TGFβI, TNFRSF12A, SOCS3 and TNFSF14) were identified by ANOVA analysis as being significantly (P value<0.01) associated with fibrosis progression in these transplant patients (data not shown). The fact that they are induced in HCV-infected Huh-7.5 cells suggests that hepatocytes themselves are an important source of these cytokines in an HCV-infected liver. The induction of SOCS2 and SOCS3, negative regulators of cytokine signaling, may indicate that the hepatocytes are actually trying to attenuate the expression of cytokines, possibly because they are negatively impacting cell viability. Interestingly, despite the lack of both TLR3 and a functional RIG-I in Huh-7.5 cells, induction of known interferon stimulated genes (ISGs), including ISG15 and ISG20, was also observed. Comparison of the gene expression profiles of HCV-infected and IFN-treated Huh-7.5 cells revealed significant overlap in differentially regulated genes, suggesting that HCV infection is associated with activation of Type 1 IFN signaling (data not shown). As the cultures were nearly 100% infected, the expression of these genes is likely coming from HCV-infected hepatocytes.
TGF-β1 is the most likely candidate for exerting effects on hepatocytes that are consistent with the gene expression data indicating cell arrest and apoptosis. It is a potent inhibitor of cell growth of many cell types, including hepatocytes, and growth arrest occurs by blocking the cell cycle at middle and late G1 phase of cell cycle [13]. Although TGF-β1 itself was not induced, there was increased abundance of a significant number of genes associated with TGF-β1 signaling, particularly at 72 hrs post-infection. Many of these genes are either associated with the TGF-β1 signaling pathway (BMP2, TGIF1, SMAD7 and 9, MRAS, FOS, ROR1, PDGRFA, JUN, INHBA) and/or are known to be regulated by TGF-β1. The increased expression of Smad7, which provides a TGF-β1-induced negative feedback loop by inhibiting nuclear translocation of SMAD proteins, suggests that TGF-β1 is produced and interacting with receptors present on hepatocytes [14]. Interestingly, KLF10 (also known as TIEG) is a gene induced by TGF-β1 that induces the generation of ROS and the loss of mitochondrial membrane potential prior to death. This, together with the fact that p53 plays a key role in TGF-β1-induced growth arrest [15], may provide an important link between the three most significant pathways affected by HCV replication: TGF-β1 signaling, p53 signaling and the NFR2-mediated oxidative stress response (Figure 6). As indicated by the network analysis, there is extensive interaction between genes associated with these pathways.
As part of a separate study examining the progression of fibrosis in liver transplant patients with re-current HCV, microarray experiments were performed comparing individual patient liver biopsy tissue (n = 25) to a pool of normal, uninfected liver tissue. To determine the clinical relevance of transcriptional changes observed in HCV-infected Huh-7.5 cells, the expression of the cell death-related genes regulated at 72 hrs post-infection was assessed in liver tissue from these HCV-infected patients. As shown in Figure 7A, many of the genes that were induced during HCV infection in cell culture were also regulated during re-current HCV infection in liver transplant patients. Interestingly, the increased expression of a subset of these genes appeared to be associated with liver disease progression as they were, in general, more highly induced in patients who developed rapidly progressive fibrosis post-transplant (indicated in red text) than in patients who did not (black text). The fact that only a subset of the cell death genes regulated in vitro were regulated in liver tissue can likely be attributed to the much lower incidence of hepatocyte apoptosis and multiple cell types in the livers of chronically infected patients. A similar scenario was observed when the expression of cytokine signaling genes differentially regulated during HCV infection of Huh-7.5 cells was assessed in the patient cohort. A subset of these genes was more highly expressed in patients who develop recurrent liver disease (indicated in red text in Figure 7B). Significantly, some were identified by ANOVA (comparing patients with and without re-current disease post-transplant) as being statistically (P value<0.05) associated with fibrosis development. Collectively, these data demonstrate that, despite JFH-1 being somewhat of an atypical HCV, transcriptional changes which occur in HCV-infected Huh-7.5 cells parallel those which occur specifically during fibrosis development in HCV-infected patients.
This study represents the first report of global transcriptional profiling of HCV-J6/JFH-infected cultured human hepatoma cells. It is unique in that it examines the host response to in vitro infection during the early acute phase of infection. The host transcriptional response corresponds closely to the levels of HCV replication, with the most gene expression changes coinciding with peak intracellular viral load (72 hours). No changes in gene expression were observed in cells treated with UV-inactivated HCVcc, indicating that viral attachment/entry does not significantly impact host gene expression. It also indicates that the changes observed in the HCV-infected cells are dependent on HCV replication and not the interaction of secreted cellular factors present in the inoculum. In contrast, both replication-dependent and -independent transcriptional changes are observed during acute influenza virus infection [16]. This discrepancy may be a reflection of inherent differences in host response to acute versus chronic viruses. Chronic viruses such as HCV may have evolved to cause minimal impact upon entry into cells in an effort to delay cellular changes that may trigger an immune response, as evidenced by the minimal effect on host gene expression even at 24 hrs post-HCV infection. Similar results were obtained in transcriptional profiling of acute HBV infection in chimpanzees, where viral entry and expansion occurred in the absence of host gene regulation [17].
Significantly, the results of this study indicate that HCV has the potential to mediate direct cytopathic effects, suggesting that not all liver injury during chronic HCV infection is immune-mediated. The initial appearance of cytopathic effect, activated caspase-3 and the highest induction of cell death-related genes all coincided with peak viral loads, suggesting that intrahepatic HCV RNA levels play a role in hepatocyte cell death. While the role of viral load in HCV-liver disease remains controversial, there is evidence to suggest that higher replication rates are associated with more severe liver disease, particularly in the liver transplant setting. Significantly higher pre- and post-transplant serum HCV levels has been associated with cholestatic fibrosis, a severe form of hepatitis [18]. Similarly, elevated serum HCV pre-transplant is associated with accelerated HCV-induced allograft injury [19]. High levels of intrahepatic HCV in biopsies taken at early times post-transplant was found to be an independent predictor of progression to chronic active hepatitis [20]. In the non-transplant setting, in situ hybridization demonstrated an association between the number of hepatocytes harboring replicating HCV and severity of fibrosis [21]. Collectively, these studies suggest that elevated viral replication may cause increased liver injury.
Differences in viral load may actually provide an explanation for the discrepancy in the level of hepatocyte cell death that occurred in vitro, which involved the majority of cells, and during chronic HCV infection. The lack of important dsRNA signaling molecules in the hepatoma cell line Huh-7.5 used for this study allows higher levels of HCV replication than what is typically observed in the liver of chronic HCV patients (Walters, unpublished data), presumably due to decreased activation of an intracellular innate antiviral response. The balance of IFN-mediated suppression of HCV replication and HCV-mediated regulation of host innate antiviral pathways likely varies from cell to cell in infected livers. If the balance shifts more toward HCV control over innate antiviral signaling, then HCV levels within that cell may increase to a level that is incompatible with cell survival. HCV infection in Huh-7.5 cells likely reflects the extreme of this situation. Indeed, levels of HCV replication in Huh-7 cells are much lower than in Huh-7.5 cells and is associated with a delay in cell death. It is possible that apoptotic hepatocytes in infected livers also have higher levels of HCV replication than non-apoptotic cells, although this would be technically challenging to demonstrate.
A significant proportion of differentially expressed genes during in vitro HCV infection were associated with cell cycle checkpoint/arrest and subsequent induction of apoptosis, which may have been in response to oxidative stress/DNA damage. Also, the observed delayed growth kinetics of HCV-infected cells and flow cytometry analysis demonstrating fewer cells in S-phase, suggests that delayed cell cycle progression may be involved in HCV-mediated cytotoxicity. This is particularly interesting in light of recent studies where immuno-histochemistry of patient liver biopsies demonstrated that few hepatocytes which have entered the cell cycle go beyond G1 phase during chronic HCV infection [22],[23]. The G1 arrest observed in patient livers was associated with increased expression of p21, a cdk-cyclin inhibitor which causes G1 arrest after DNA damage. A correlation was observed between p21 expression and fibrosis severity, suggesting a link between delayed cell cycle progression and liver injury. Such results suggest the delay in cell cycle progression observed in HCV-infected Huh-7.5 cells is physiologically relevant. Consistent with these in vivo studies, many of the genes associated with arrest during in vitro HCV infection are involved in G1 arrest or transition from G1 to S phase. Interestingly, a link between perturbations in cell cycle control and pathogenesis has been observed in SIV infection in non-human primates, another chronic viral infection that can have different disease outcomes. Perturbation of the cell cycle within CD4 (+) T lymphocytes is characteristic of pathogenic HIV/SIV infection [24],[25],[26] and, similar to what is proposed in the current study, increased T lymphocyte susceptibility to apoptosis correlates with cell cycle perturbation [26]. Unlike the AIDS field, there is no animal model of HCV-associated liver disease in which to validate the biological significance of events which occur in cell culture models. To circumvent this challenge, gene expression data from in vitro HCV infection was integrated with an extensive database of patient liver microarray experiments. The intriguing finding that a higher induction of a subset of these genes was observed in HCV-infected patients with rapidly progressive fibrosis post-transplant, but not those HCV-infected patients lacking histological evidence of fibrosis, also suggests a potential role of cell cycle perturbations in HCV pathogenesis. However, it is important to note that this patient cohort is immuno-compromised and so it is uncertain if HCV-associated CPE causes significant liver injury in immuno-competent individuals.
Both gene expression profiling and flow cytometry analysis suggest that HCV-mediated apoptosis of Huh-7.5 cells is linked to perturbations in cell cycle progression. However, it is difficult to determine from the gene expression data if the cell cycle arrest is directly linked to apoptosis or if there other factors that are driving the arrested cells to undergo apoptosis. It is also not clear what factors are responsible for inducing the delay in cell cycle progression. Perturbation of the cell cycle may be mediated directly by HCV proteins, as has been observed in other viral infections including HIV (vpr) and HBV (x protein) [27],[28]. Due to the association between chronic HCV infection and development of hepatocellular carcinoma, there has been keen interest in the impact of HCV on cell cycle regulation. Core protein in particular has been implicated in impairment of G1 to S phase transition through multiple mechanisms, including induction of p21 expression and concomitant decrease in cdk2 activity [29], direct interaction and suppression of CAK activity [30], and stabilization of cell cycle inhibitor p27 [31]. Delayed progression through S-phase has been shown to be mediated both by NS2-mediated down-regulation of cyclin A [32] and NS5B induction of IFN-B [33]. NS5A-induced chromosome instability has been linked to aberrant mitotic regulation, including impaired mitotic exit [34]. It is unclear whether delaying cell cycle progression would be beneficial or detrimental to HCV replication. The liver is normally a quiescent organ and so hepatotropic viruses which establish chronic infections, such as HBV and HCV, may have evolved to replicate efficiently under such non-proliferating conditions. In support of this, production of infectious HCV in cultured hepatoma cells does not seem to be significantly impacted during growth arrest induced by either serum starvation or DMSO [35],[36].
The differential regulation of numerous genes associated with DNA damage/oxidative stress response, including many associated with the NRF2-oxidative stress response, suggests this as a possible mechanism of arrest. Oxidative stress has long been thought to play a key role in HCV pathogenesis and a potential link between HCV-associated perturbation of lipid metabolism genes and oxidative stress was observed in the SCID-Alb/uPA mouse model [5]. Specifically, HCV-infected animals showing induction of genes functioning in cholesterol biosynthesis, peroxisome proliferation, and β-oxidation also showed induction of genes which function in antioxidant cell defense, presumably due to generation of reactive oxygen species (ROS) during β-oxidation of fatty acids. In the current study, oxidative stress appears to be independent of lipid metabolism. No significant regulation of genes associated with cholesterol synthesis or enzymes involved in β-oxidation was observed at any time-point of infection. It is possible that the presence of viral gene products or replication directly results in the generation of ROS. Core protein has been implicated in perturbations in mitochondrial function, including release of cytochrome c and loss of membrane potential, as well as production of ROS in a variety of systems [37],[38],[39]. Consistent with the current study, expression of full-length HCV open reading frame was found to cause marked growth inhibition and increased intracellular ROS [40]. Preliminary global quantitative proteomic data showed multiple perturbations in the host proteome indicative of HCV-associated metabolic stress and ROS generation as early as 24 h post infection (Diamond et al, manuscript in preparation). Consistent with this idea, these perturbations were accompanied by a concomitant increase in proteins functioning in antioxidant cell defense.
Another possible mechanism of cell arrest could be through activation of TGF-β1 signaling. TGF-β1 is a potent inhibitor of cell growth and apoptosis of many cell types, including hepatocytes, and growth arrest occurs by blocking cell cycle at mid and late G1 phase [13],[14]. This is consistent with the gene expression data showing regulation of numerous genes involved in G1 arrest. Significant correlations between TGF-β1 polymorphisms, intensity of hepatocyte-specific TGF-β1 staining, serum TGF-β1 levels and degree of fibrosis have consistently been demonstrated in HCV-infected patients [41],[42],[43]. However, its role in hepatocyte apoptosis during HCV infection remains unclear. The possibility that TFG-β1 may be directly involved in HCV-associated cell cycle arrest is intriguing. While increased expression of TGF-β1 was not observed in the current study, the expression of numerous genes associated with TGF-β1 signaling pathway, and those known to be induced by TGF-β1, was elevated during infection. It is possible that very low levels of the cytokine are needed to activate the pathway, a scenario similar to what is observed with Type 1 IFN and ISGs. Furthermore, in an infected liver other cell types, including hepatic stellate cells, are an important source of TGF-β1. Chronic HCV infection is universally associated with the induction of ISGs in the liver but usually in the absence of detectable increased expression of IFN-α/β [5],[44]. TGF-β1 has been proposed as a potential therapeutic target for treatment of viral hepatitis and so a clear understanding of the role it plays in HCV pathogenesis is crucial [45]. Further study is warranted to determine if TGF-β1 is inducing cell arrest and also whether this is directly linked to apoptosis or simply sensitizing cells to apoptosis through alternative mechanisms, including the effects of oxidative stress.
Alternate mechanisms of HCV-induced cytopathic effects, such as through induction of ER stress, have been proposed [46]. However, there was little evidence for regulation of genes associated with either ER stress or the unfolded protein response in the current study. This is consistent with a recent study demonstrating that HCV-JFH-1 mediates apoptosis through a mitochondrion-mediated, caspase-3 dependent pathway in the absence of ER stress [47]. Although ER stress is associated with transcriptional regulation of a subset of genes, it is unclear if gene expression profiling would accurately detect ER stress and so it should not be ruled out as a possible contributor of apoptosis in the current study. Apoptosis has also been found to be mediated through the induction of the death ligand, TRAIL, and its receptors [48]. No increase in the expression of TRAIL, or its receptors, was observed in the current study, making it unlikely to be involved in apoptosis of Huh-7.5 cells. This is possibly related to the absence of functional RIG-I and TLR3 in these cells, which are key components of dsRNA signaling pathways. It is important to note that these alternative mechanisms of HCV-mediated cell death are not necessarily mutually exclusive, and multiple mechanisms of cytotoxicity may be involved in liver injury during chronic HCV infection.
Collectively, the gene expression data and flow cytometry analysis suggests that HCV infection is associated with perturbation of the cell cycle which may sensitize cells to apoptosis. Significantly, the integration of the in vitro and patient liver gene expression data also suggests that this process contributes to liver disease progression. These results suggest that despite the fact that HCV typically establishes persistent infections, events which occur during the very acute phase of infection of individual hepatocytes can determine the ultimate fate of the cell. During the time this manuscript was in preparation, a report was published describing altered expression of cell cycle and apoptotic proteins, as demonstrated using immunohistochemistry, in liver biopsies from chronic HCV patients [49]. However, the delay in cell cycle progression and apoptosis were observed in separate cell types (hepatocytes versus sinusoidal cells, respectively). The results of the current study demonstrate that cell cycle perturbation and apoptosis occur in the same cell, suggesting a direct link, and also provide additional insight into potential mechanisms of cell cycle perturbation, including oxidative stress and TGF-β1 signaling. Further study is warranted to more clearly define the mechanism of HCV-associated cell cycle perturbation. This may provide significant insight into the pathogenesis of HCV infection with the possibility of identifying novel therapeutic targets.
Huh 7.5 cells (human hepatoma) were electroporated with 1 µg of in vitro transcribed RNA from the chimeric HCV genome J6/JFH (see reference 7); five identical electroporations were performed. Cells were expanded three times following electroporation and supernatants were pooled and used to infect naïve Huh 7.5 cells. Following a single expansion of infected cells, virus containing supernatants were collected, pooled, and used to again infect naïve Huh 7.5 cells. This process was repeated a total of five times and resulted in the generation of a relatively high titer (2×105 TCID50/ml) and large volume (∼500 mls) of stock virus (HCVcc). Approximately one-half of the HCVcc stock was exposed to UV light for 60 seconds, using a Stratalinker UV light box, and served as a non-infectious control (UV-HCV). In addition, during the multiple passages of HCVcc on naïve cells, a mock control sample was generated by passing conditioned media along in parallel. For infections, Huh 7.5 cells were seeded at a density of 3×106 cells/plate on p150 plates and treated for ∼8 hours with 20 mls of supernatant containing virus (HCVcc), UV-inactivated virus (UV-HCV), or conditioned media (mock). This amounted to a moi of ∼1.3. Following initial infection, the supernatant was replaced with fresh media and incubated until harvest at 24, 48, 72, 96, and 120 hours post-infection.
Following removal of supernatant, cells were washed once with PBS and then scraped from the plate in ice cold PBS. RNA was isolated from approximately 106 cells using RNeasy mini prep kit with an on-column DNase treatment, following the manufacturer protocol (Qiagen).
Cells were fixed in −20°C methanol or 1% paraformaldehyde (in PBS at room temperature) for 15–20 minutes. Following a series of PBS rinses, the cells were blocked in 1% BSA/0.2% skim milk in PBS for 30–60 minutes at room temperature. Cells were incubated in primary antibody (diluted in 0.5% Tween-20 in PBS) overnight at 4°C; mouse anti-NS5A (1∶2000, Clone 9E10, ref. 21) and rabbit anti-activated caspase-3 (1∶500, Cell Signaling). Click-iT EdU chemistry was performed following manufacturer protocol (Invitrogen); 10 µM EdU labeling for 3 hours and 30 minute reaction with AlexaFluor 488 azide (1∶400 dilution). Secondary antibodies are AlexaFluor conjugated and used at 1∶1000 dilution; goat anti-mouse AlexaFluor488 and goat anti-rabbit AlexaFluor594.
Cells were trypsinized, washed twice with ice cold PBS and fixed 1% paraformaldehyde for 20 minutes. Following a series of PBS rinses, the cells were blocked and permeabilized in 0.1% FBS/0.1% saponin in PBS. HCV infected cells were detected using an anti-NS5A antibody (clone 9E10) directly conjugated with the AlexaFluor647 fluorophore (according to manufacturer protocol; Invitrogen, A20173). Proliferating cells were detected using Click-iT EdU chemistry, as described above, following manufacturer guidelines for flow cytometry analysis (Invitrogen); AlexaFluor 488 azide (1∶400 dilution). Flow cytometry was performed on a BD FACSCalibur machine with analysis done using FlowJo software (version 8.7.1).
Core needle liver biopsies were collected from patients at the University of Washington All patients gave informed consent to protocols approved by the Human Subjects Review Committee at the University of Washington. Normal, uninfected liver tissue (n = 10) was obtained from donor livers that were considered unacceptable for liver transplantation. These uninfected samples were pooled to create a standard normal liver reference that was used for all microarray experiments using patient tissue. Fibrosis was graded by a single liver pathologist using the Batts-Ludwig grading system [50].
Microarray format, protocols for probe labeling, and array hybridization are described at http://expression.microslu.washington.edu. Briefly, a single experiment comparing two mRNA samples was done with four replicate Human 1A (V2) 22K oligonucleotide expression arrays (Agilent Technologies) using the dye label reverse technique. This allows for the calculation of mean ratios between expression levels of each gene in the analyzed sample pair, standard deviation and P values for each experiment. Spot quantitation, normalization and application of a platform-specific error model was performed using Agilent's Feature Extractor software and all data was then entered into a custom-designed database, Expression Array Manager, and then uploaded into Rosetta Resolver System 7.0 (Rosetta Biosoftware, Kirkland, WA) and Spotfire Decision Suite 8.1 (Spotfire, Somerville, MA). Data normalization and the Resolver Error Model are described on the website http://expression.viromics.washington.edu. This website is also used to publish all primary data in accordance with the proposed MIAME standards [51]. Selection of genes for data analysis was based on a greater than 95% probability of being differentially expressed (P≤0.05) and a fold change of 2 or greater. The resultant false positive discovery rate was estimated to be less than 0.1% (Walters, unpublished data). Ingenuity Pathway Analysis (IPA) software and Entrez Gene (www.ncbi.nlm.nih.gov/sites) were used for gene ontology analysis.
Quantitative real-time PCR (RT-PCR) was used to validate the gene expression changes and measure intrahepatic HCV RNA. Total RNA samples were treated with DNA-free DNase Treatment and Removal Reagents (Ambion, Austin, TX). Reverse transcription was performed using random hexamer primers and Taqman RT reagents (Applied Biosystems, Foster City, CA). Real-time PCR was performed using an ABI 7500 Real Time PCR system and Taqman chemistry. Each target was run in quadruplicate with Taqman 2× PCR Universal Master Mix and a 20 µL total reaction volume. Primer and probe sets for relative quantification were selected from the Assays-on-Demand product list (Applied Biosystems) including two endogenous controls, GAPDH and 18 S ribosomal RNA. Quantification of each gene, relative to the calibrator, was calculated by the instrument, using the equation 2−ΔΔCT within the Applied Biosystems Sequence Detections Software version 1.3. Probes used for analysis (Applied Biosystems): Human genes: eukaryotic 18S rRNA (Catalogue No. Hs99999901_s1); ATF3 (catalogue No. Hs00910173_ml), MKi67 (catalogue No. Hs01032443_ml), MCM4 (catalogue No. Hs00381539_ml), MCM6 (catalogue No. Hs00195504_ml), TGIF1 (catalogue No. Hs00545014_ml), CAT (catalogue No. Hs00156308_ml), SMAD7 (catalogue No. Hs00998193_ml), GADD45A (catalogue No. Hs00169255_ml), GADD45B (catalogue No. Hs00169587_ml)
Primer and probe sets for absolute quantification of intrahepatic viral load were designed based on sequences of HCV 1a armored RNA (Ambion Diagnostics, Austin, TX) using Primer Express (version 3). A standard curve was made from six serial dilutions of HCV 1a armored RNA (Ambion Diagnostics) with a known viral copy number. The PCR efficiency was determined by the slope of the standard curve Standard curve analysis and viral load was determined using the Applied Biosystems SDS Software 1.3 (Applied Biosystems, CA). Total RNA was DNase treated prior to cDNA synthesis via reverse transcription and all samples were processed with equal mass amounts of total RNA [52]. All measurements were taken in quadruplicate with negative and non-template controls. Primer and probe sets consisted of F: CAC TCC CCT GTG AGG AAC TAC TG, R: GCT GCA CGA CAC TCA TAC TAA CG, and P: 6FAM-TTC ACG CAG AAA GC-MGBNFQ and were designed from the 5′UTR using Primer Express 3.0 (Applied Biosystems, CA). Quantification of HCV RNA levels was performed on the same total RNA sample that was used for the microarray experiments.
|
10.1371/journal.pbio.0050105 | A Dominant, Recombination-Defective Allele of Dmc1 Causing Male-Specific Sterility | DMC1 is a meiosis-specific homolog of bacterial RecA and eukaryotic RAD51 that can catalyze homologous DNA strand invasion and D-loop formation in vitro. DMC1-deficient mice and yeast are sterile due to defective meiotic recombination and chromosome synapsis. The authors identified a male dominant sterile allele of Dmc1, Dmc1Mei11, encoding a missense mutation in the L2 DNA binding domain that abolishes strand invasion activity. Meiosis in male heterozygotes arrests in pachynema, characterized by incomplete chromosome synapsis and no crossing-over. Young heterozygous females have normal litter sizes despite having a decreased oocyte pool, a high incidence of meiosis I abnormalities, and susceptibility to premature ovarian failure. Dmc1Mei11 exposes a sex difference in recombination in that a significant portion of female oocytes can compensate for DMC1 deficiency to undergo crossing-over and complete gametogenesis. Importantly, these data demonstrate that dominant alleles of meiosis genes can arise and propagate in populations, causing infertility and other reproductive consequences due to meiotic prophase I defects.
| About 10%–15% of couples are infertile due to defects in meiosis (the process by which egg or sperm cells containing a single copy of each chromosome are produced). Because studying the genetics of meiosis in humans is difficult, we performed genetic screens in mice and identified a novel mutation in Dmc1 that causes male-specific infertility due to defects in meiosis. Dmc1 encodes a key protein required for meiotic recombination; the mutation causes a single amino acid change that prevents genetic exchange, or crossing-over, in males, abolishes its recombination activity, and abrogates the production of sperm. Though heterozygous females are fertile, they have fewer oocytes due to a high incidence of meiosis I abnormalities, and show susceptibility to premature ovarian failure. Importantly, these data demonstrate that dominant alleles of meiosis genes can arise and propagate in populations, and produce meiotic prophase I defects that cause infertility and other reproductive abnormalities.
| Genetic recombination occurs in all organisms and is critical for repair of DNA damage, proper chromosome segregation during meiosis, and genetic diversification. Recombination in yeast and mice is initiated by the formation and processing of double-strand breaks (DSBs). Meiotic DSBs are repaired by proteins that mediate homologous strand exchange, mismatch repair, and resolution of recombination intermediates. As these activities are occurring, homologous chromosomes undergo pairing and synapsis, which are completed by the pachytene stage of meiosis. The ability of germ cells to complete meiosis, and to undergo proper segregation of chromosomes in the subsequent meiotic divisions, hinges on the fidelity of these events. Defects in recombination and meiosis have been shown to underlie aneuploidy syndromes such as Downs [1] and azoospermia in men [2].
Our understanding of the genetic control of meiotic recombination in mammals has depended largely on studies of model organisms such as yeast. Mice with null mutations in orthologs of recombination genes often have meiotic defects that are very similar to yeast. However, it is clear that mammals have many genes required for meiosis that do not have orthologs in yeast, and that there are substantial differences between the sexes in the response to mutations in meiotic genes [3–5]. Additionally, mouse null mutations generated by gene targeting of yeast orthologs do not model deleterious hypomorphic or dominant alleles that may occur in human populations.
Because of these issues, and because the genetics of meiosis is difficult to address in humans, we and others have undertaken a forward genetic approach to the identification of novel mutant genes causing infertility in mice [6,7]. An example of the power of this approach was the identification of Mei1, a novel vertebrate-specific meiosis gene subsequently implicated in human male infertility [8]. Here, we describe the isolation of an allele of Dmc1 (Dmc1Mei11) that uncovers remarkable sex-specific properties, unlike a null allele.
Dmc1 is a meiosis-specific RecA/Rad51 homolog required for recombinational repair of meiotic DSBs. Escherichia coli RecA promotes strand transfer between homologous DNA molecules in an ATP-dependent manner [9], and human DMC1 has intrinsic ATP-dependent strand invasion activity that is stimulated in the presence of the HOP2-MND1 complex [10]. In Saccharomyces cerevesiae and mice of both sexes, DMC1-deficient mutants arrest in late zygonema/early pachynema of meiotic prophase I, with an accumulation of DSBs and defective synaptonemal complex (SC) formation [11–13]. In contrast, Dmc1Mei11 has the unusual property of causing autosomal dominant male sterility. Our studies of this allele provide insight into the biochemical properties of DMC1, underscore stark sexual dimorphism in meiotic recombination and response to errors thereof, and provide experimental proof that autosomal dominant meiotic mutations can arise, propagate, and cause mammalian infertility.
In previous work, phenotype-driven screens were conducted on pedigrees of mice derived from chemically mutagenized embryonic stem cells [14]. During analysis of a pedigree that was founded with a first-generation (G1) daughter of a chimeric male, a fully penetrant, dominant male-specific sterility trait was discovered and mapped to Chr 15 [6] (see Materials and Methods). Based on phenotypic analyses described herein, we labeled this mutation Meiosis defective 11 (Mei11). Since Mei11 can be transmitted only through females, yet the pedigree was initiated by a male chimera, it is possible that the mutation arose not in embryonic stem cells but in the G1 female or one of her descendants.
We found that spermatogenesis in males heterozygous for Mei11 in the C3HeB/FeJ (C3H) strain background (≥N3) was arrested in meiotic prophase, with an absence of postmeiotic spermatids (Figure 1A–1C). The most advanced type of germ cells were spermatocytes with nuclear chromatin characteristic of zygotene/pachytene-stage cells (Figure 1B), present in testicular seminiferous tubules at epithelial stage IV, a developmental point at which many meiotic mutants arrest [15]. Failure to progress further led to a massive degeneration of spermatocytes (Figure 1C, labeled tubule). This histological phenotype is indistinguishable from that displayed by Dmc1−/− knockout mice [11,12,16]. Interestingly, the meiosis I arrest phenotype was partially modified by genetic background. Heterozygous males in the C57BL/6J (B6) background (N3–N8) were universally sterile, and the vast majority of seminiferous tubules were also arrested at epithelial stage IV (Figure 1D and 1E). However, we observed rare seminiferous tubules containing mid-pachynema spermatocytes (with XY bodies; Figure 1E, arrows) and postmeiotic round and elongating spermatids (Figure 1F, blue and black arrows, respectively).
We genetically mapped (see Materials and Methods) Mei11 to a ~2 Mb region containing the meiotic gene Dmc1 (Figure S1A). Despite the recessive, non–sex-specific nature of null alleles, we sequenced Dmc1 and identified two consecutive base changes (TG to CC) in all affected Mei11/+ males and carrier females (Figure S1B). These correspond to nucleotides 968–969 of the mouse Dmc1 transcript and cause an alanine to proline change (discussed below). To verify this mutation causes the mutant phenotype, we generated mice containing Mei11 in trans to a Dmc1 null allele. This resulted in female sterility concomitant with ovarian defects (see below). Furthermore, we found that B6-Mei11/Dmc1− testes underwent complete prophase I meiotic arrest equivalent to that in Dmc1−/− males (not shown), without signs of postmeiotic differentiation as in B6-Mei11/+ males (Figure 1E and 1F). The failure to complement the Dmc1 null indicates that Mei11 is a defective allele of Dmc1 and is likely causative for the infertility phenotype of male heterozygotes. Accordingly, we labeled the Mei11 mutation Dmc1Mei11.
To characterize the underlying cause of meiotic arrest in mutant spermatocytes, we examined the behavior of meiotic chromosomes throughout prophase I using antibodies directed against diagnostic markers of meiotic chromatin. We monitored chromosome synapsis and SC formation by delineating the distribution of SYCP3, a component of axial/lateral SC elements, and SYCP1, a component of SC transverse filaments that marks synapsed chromosome regions [17,18]. SYCP1 and SYCP3 decorate the axes of all 19 synapsed pachytene stage autosomes in wild-type (wt) spermatocytes, except for the XY bivalent, in which only the pseudoautosomal region (PAR) contains SYCP1 (Figure 2A). While chromosomes in the most advanced Dmc1−/− nuclei were in a zygonema-like state and devoid of homologous synapsis (unpublished data; [11]), Dmc1Mei11/+ spermatocytes exhibited either a zygotene-like (37% versus 10% in wt controls) or aberrant pachytene-like morphology (62%), containing a mixture of synapsed, partially synapsed, and asynapsed chromosomes (Figure 2B). Consistent with testis histology, we observed rare (1%–2% versus 25% in wt) nuclei with a full complement of synapsed bivalents (Figure 2E). Similar results were obtained with the 125 B6- Dmc1Mei11/Dmc1− postleptonema nuclei we examined; 43% were zygotene stage and 57% displayed the aberrant pachytene-like chromosome morphology with regions of SYCP1 staining (Figure 2C) on some chromosome pairs. Thus, these compound heterozygotes progressed further than nulls.
We also monitored synapsis using an antibody against phosphorylated histone H2AX (γH2AX). While serving as a marker of leptotene DSBs [19], H2AX phosphorylation also occurs during pachynema of male meiosis as part of a meiotic signaling pathway involved in transcriptional silencing of unpaired chromatin [20,21]. In wt and (rare) mutant mid-pachytene spermatocytes with fully synapsed chromosomes spermatocytes, γH2AX was present only over the XY body (Figure 2D and 2E). However, in aberrant pachytene-like mutant nuclei, some autosomal meiotic cores stained positively, a pattern indicative of asynapsis (Figure 2F).
To assess the state of recombinant DSB repair in mutant spermatocytes, we characterized the distribution of the RAD51 and DMC1 recombinases by immunolabeling with an anti-RAD51 antibody that recognizes both proteins. These proteins colocalize along meiotic cores in “foci” that are coincident with early meiotic recombination nodules (RNs). We observed no obvious difference in numbers of RAD51/DMC1-positive foci between Dmc1Mei11/+, Dmc1Mei11/Dmc1−, and wt zygotene nuclei (Figure S2). Normally, RAD51/DMC1 foci disappear by mid-pachynema, except along the asynapsed meiotic cores of the XY bivalent (Figure 2G). The rare Dmc1Mei11/+ nuclei that progressed this far had a similar pattern (Figure 2H), but RAD51/DMC1 foci persisted on the asynapsed regions of meiotic cores in aberrant pachytene-like spermatocyte nuclei (Figure 2I, arrows), indicating the presence of unrepaired breaks. Mutant spermatocyte cores also displayed persistent RPA foci, indicating the presence of single-stranded DNA (ssDNA; Figure 2J). When we used a DMC1-specific antibody, we observed approximately half the number of foci in zygotene Dmc1Mei11/Dmc1− spermatocytes (average, 52; N = 4) compared with wt (average, 113; N = 6; Figure S3) and Dmc1+/− (average, 105; N = 5) spermatocytes, suggesting that DMC1Mei11 is deficient for RN incorporation (Figure S3).
To determine if crossing-over occurs in Dmc1Mei11/+ spermatocytes, we probed meiotic chromosomes with an antibody against the mismatch repair protein MLH3, a marker of chiasmata [22]. Whereas chromosome cores of wt mid-pachytene spermatocyte chromosomes display one to two MLH3 foci per synapsed homolog (Figure 2K), we saw none on synapsed regions of mid-pachytene–like B6− Dmc1Mei11/+ chromosomes (Figure 2L).
To examine effects of Dmc1Mei11 on female gametogenesis, we conducted breeding studies, histological analyses of ovaries, and immunocytological examination of meiotic chromosomes. Litter sizes of <6-mo-old Dmc1Mei11/+ females were normal, regardless of genetic background (Table S1), and the morphologies of young (6 wk) mutant ovaries appeared normal (Figure 3A and 3B). However, a few (4 of 26) C3H congenic females in extended matings (>4 mo) exhibited a premature decline in fertility. The ovaries of these older females were nearly devoid of developing follicles (Figure 3C and 3D). This observation prompted us to quantify the oocyte pool in young, sexually mature mice (6 wk old). As summarized in Figure S4, we noted a depletion of primordial and primary follicle pools in 6-wk-old B6− Dmc1Mei11/+ females compared with wt animals (by 32% and 31%, respectively). We observed approximately wt levels of more developed follicles, reflecting the ability of ovaries to recruit and mature a “normal” number of oocytes from a smaller resting pool.
Dmc1Mei11/Dmc1− females were sterile, regardless of genetic background. Interestingly, ovaries of such females fell into two phenotypic classes. Approximately 60% (8 of 13) exhibited a relatively normal ovarian morphology with developing and antral-stage follicles (Figure 3E), albeit fewer than in wt or Dmc1Mei11/+ (Figure S4). The remaining 40% of paired ovaries (5 of 13 females) were residual and devoid of follicles (Figure 3F), like Dmc1−/− [11].
To determine whether the reduced oocyte number in mutant adult females was related to recombination and synapsis defects, we examined prophase I chromosomes from embryonic ovaries. Whereas the majority of embryonic day 17.5 oocytes from wt embryos were at mid-pachynema (59% of 63 nuclei examined from 4 embryos, plus 11% in diplonema) with fully synapsed autosomes as indicated by colabeling with anti-SYCP1 and anti-SYCP3 (Figure 4A), only 18% of Dmc1Mei11/+ nuclei (of 131 examined from 3 embryos, plus only 4% in diplonema) had fully synapsed meiotic cores (Figure 4B), whereas the great majority (64%) had either a zygotene morphology or a pachytene-like mixture of completely asynapsed, partly synapsed, and completely synapsed homologs (Figure 4C). Dmc1−/− chromosomes were mostly asynapsed and zygotene-like (Figure 4D). Interestingly, we observed a unique type (~14% ) of aberrant Dmc1Mei11/+ oocyte nuclei that contained >40 asynapsed pachytene-like meiotic cores (Figure S5). These appear to be oocytes that failed to achieve synapsis and underwent some degree of meiotic core fragmentation as indicated by lack of SYCP1 staining, persistent unrepaired breaks (in the form of RAD51 foci), positive staining of cores for REC8 and STAG3 (not shown), and a lack of centromeric staining (as detected by immunolabeling with CREST antisera) on a subset of mutant oocyte cores. Consistent with the sterility of Dmc1Mei11/Dmc1− females, we found no oocytes with a full complement of synapsed bivalents; most were zygotene-like (unpublished data).
We assessed recombination in mutant females cytologically and genetically. Similar to spermatocytes, the fully synapsed chromosomes of normal Dmc1Mei11/+ mid-pachytene stage oocyte nuclei were devoid of RAD51/DMC1 foci (Figure 4H) and γH2AX staining (Figure 4F), but the unsynapsed cores of aberrant nuclei had persistent RAD51/DMC1 foci (Figure 4I, arrows) and γH2AX signals (Figure 4G), indicative of unrepaired DSBs. The numbers of MLH3 foci along fully synapsed chromosomes of wt and Dmc1Mei11/+ mid-pachytene oocytes were similar (an average of 25 and 24 foci per wt and mutant nucleus, respectively; Figure 4J and 4K), indicating that crossing-over was normal in this class of mutant oocytes. In aberrant pachytene-like oocyte nuclei, MLH3 foci were present on some meiotic cores but not others, presumably corresponding to synapsed versus unsynapsed bivalents (Figure 4L; arrows). To measure recombination genetically, we crossed C3H-Dmc1Mei11/+ N12 females to B6 males to create control and heterozygous F1 females, backcrossed them to B6, and genotyped the N2 progeny with polymorphic microsatellite markers along three chromosomes. The overall recombination frequencies were very similar (47.8% versus 46.4%), although the distribution of crossovers on different chromosomes varied (Table 1).
Because of the genetically dominant nature of Dmc1Mei11, we tested whether DMC1Mei11 retains the ability to interact with itself or DMC1 in the yeast two-hybrid system. No qualitative disruption in two-hybrid interactions was observed (Figure S6), suggesting that DMC1Mei11 could participate in the assembly of DMC1/DMC1Mei11 hybrid polymers.
In vitro, it has been shown that DMC1 nucleates on ssDNA to form extended filaments in an ATP-dependent manner [23–25], and that these nucleoprotein filaments promote a search for DNA homology and strand invasion on an intact duplex (D-loop). Recently it was shown that the HOP2-MND1 complex physically interacts with and stimulates strand invasion mediated by DMC1 [10,26]. The Dmc1Mei11 mutation causes an alanine to proline change at amino acid 272 of DMC1 (A272P), lying within the counterpart of the structurally disordered “Loop 2” (L2) domain of RecA [27] (Figure S1C) that was shown to be involved in DNA binding [28]. To assess the DNA-binding and recombinase activities of DMC1Mei11, we constructed an equivalent A272P mutation within the human protein (HuDMC1Mei11) and purified it in bacteria (Figure 5A). We analyzed the DNA-binding activity of this mutant by gel shift analysis and measured strand invasion activity using a system optimized for recombinant human DMC1 [10]. Both the wt and mutant protein preferentially bound ssDNA over double-stranded DNA (dsDNA), but the ssDNA-binding activity of the mutant protein was reduced approximately 3-fold (Figure 5B and 5C). Strikingly, mutant HuDMC1Mei11 was incapable of catalyzing strand invasion (assayed as D-loop formation), even after addition of the HOP2-MND1 complex (Figure 5D and Figure S7).
To assess possible dominant-negative effects, we measured D-loop activity in experiments where HuDMC1Mei11 was mixed with HuDMC1 in stoichemetric quantities. These experiments failed to reveal an inhibition of D-loop activity of the wt protein, either in the absence (unpublished data) or presence (Figure S7) of HOP2-MND1.
Sexual dimorphism in mammalian gametogenesis exists at several levels, including such basic differences as the timing of meiotic progression and the number of gametes produced, the frequency and distribution of crossovers, and differential checkpoints or checkpoint stringencies [3,29]. In general, sterile mutations in mice that affect chromosome synapsis but not DSB repair confer more severe defects during spermatogenesis than oogenesis. Examples of this are mutations in Spo11 and Mei1. Whereas mutant oocytes can proceed into pachynema and even undergo the first meiotic division, spermatocytes deficient for these genes and DSB repair genes such as Dmc1 arrest earlier, at the zygotene/pachytene transition [4,30,31]. A more dramatic example of sexual dimorphism is exemplified by animals deficient for the SC component SYCP3. While males arrest at epithelial stage IV (corresponding to mid-pachynema in wt spermatocytes) [5], mutant females are fertile; they produce aneuploid gametes linked to failed chiasmata, causing increased embryonic mortality [32]. These results suggest that meiosis in female mice is more error prone and/or that females are less capable of recognizing and eliminating gametes with meiotic errors, a phenomenon that is relevant to human reproductive biology since aneuplodies that arise in human fetuses have been largely attributed to meiotic errors of maternal origin [33]. The Dmc1Mei11 allele is unique in that oogenesis can be completed in a subset of carrier oocytes despite defects in DSB repair.
The reason for the common arrest point in males is not yet clear. It is possible that different checkpoint signals converge on the same apoptotic pathway, or that apoptosis is caused by a defect shared by the various meiotic mutants. Candidate defects include MSUC (meiotic silencing of unpaired chromatin) that occurs in response to asynapsis, and/or failure of sex chromosome inactivation in mid-pachynema [34], the latter of which may result from excessive autosomal MSUC (P. Burgoyne, personal communication).
The fact that Dmc1Mei11/+ females were fertile despite exhibiting a high percentage of abnormal meioses, and that most mutant oocytes were capable of normal crossing-over compared with the complete lack thereof in spermatocytes, underscores the stark sexual dimorphism with respect to checkpoint control and recombination mechanisms. A key underlying difference may have to do with the timing of checkpoint activation in meiotic prophase I. For example, an earlier checkpoint activation may occur in spermatocytes before expression of one or more essential components of late RNs, preventing them from being made or modified in a way that is required for late RN formation. Notably, MLH3 foci are apparent prior to pachynema in human oocytes [35], and components of early and late RNs appear earlier in female than male mice in terms of days lapsed from the beginning of meiotic prophase [36]. Later checkpoint activation in oocytes relative to the onset of meiosis might allow more time to resolve recombination/synapsis defects in a subset of mutant oocyte nuclei, allowing them to progress through meiosis and eventually ovulate [37]. Another possibility is that oocytes contain a sufficient or elevated pool of other proteins that can substitute for diminished DMC1 activity. For example, overexpression of RAD51 or RAD54 can partially rescue meiosis and sporulation of dmc1 yeast [38,39].
Whereas Dmc1−/− spermatocytes are essentially devoid of homologous synapsis, Dmc1Mei11/Dmc1− spermatocytes have a mixture of pachytene-like synapsed homologs and asynapsed chromosomes. Therefore, DCM1Mei11 clearly retains some activity that is sufficient to drive some degree of synapsis. Though it is completely unable to catalyze D-loop formation in vitro, the mutant protein might facilitate recombinant activities of other molecules so as to initiate interhomolog pairing. In mammals, DMC1 has been shown to interact with many proteins, including RAD51, RAD54B, BRCA2, MSH4, MND1, and HOP2 [40,41]. Therefore, the DMC1Mei11 protein may retain recombinogenic activity in vivo by virtue of protein–protein interactions that are not recapitulated in the in vitro system. Alternatively, other molecules such as RAD51 may be able to drive some amount of interhomolog pairing independently of DMC1, but DMC1Mei11 retains an additional, independent role in synapsis per se. Notably, S. cerevesiae is capable of low levels of crossover recombination in the absence of DMC1 [42].
Despite acting in a genetically dominant fashion in vivo, recombinant HuDMC1Mei11 did not exert a dominant negative effect on D-loop formation in vitro. One possibility for this discrepancy is that DMC1Mei11 forms dominant negative interactions with another key protein(s) in vivo, resulting in defective DSB processing. The mutant protein did not impair the ability of wt DMC1 to catalyze strand invasion in the presence of HOP2-MND1, ruling them out as the target of dominant-negative interactions. However, as stated above, mammalian DMC1 is known to interact with several other meiotic proteins involved in recombination and DNA repair. Since the Dmc1Mei11 mutation decreased DNA binding efficiency in vitro to approximately 30% of wt activity, consistent with a decreased number of DMC1 foci in mutants, it is possible that the dominant-negative interaction may result from binding to a key protein(s) that depends upon DMC1 for localizing to sites of DSB repair.
In Dmc1−/− mutant females, oocytes are depleted around birth prior to primordial follicle formation, resulting in residual ovaries that are depleted of ovarian follicles and oocytes [11,12]. The proximal cause of developmental arrest at this stage of oocyte meiosis appears to be the failure to repair DSBs, rather than asynapsis, since genetic elimination of DSBs via mutations in Mei1 or Spo11 in DMC1-deficient mice permits further meiotic progression through metaphase I in females [43,44]. Young Dmc1Mei11/+ females were normally fertile, but had a deficit of primordial and primary stage follicles. Depletion of early-stage follicles in response to unrepaired chromosome breaks and/or asynapsis has also been reported for mice with mutations in Spo11 and Sycp3 as well as mouse models of Turner's syndrome [37,43,45]. Overall, these findings indicate that a subset of oocytes was eliminated in mutant females prior to primordial follicle formation, probably leading to the age-related subfertility in a subset of mutant females. Despite the reduction in the oocyte pool, it appears that young Dmc1Mei11/+ females are able to compensate by recruiting a near-wt number of oocytes for ovulation. This type of compensation has also been reported for mouse models of Turner's syndrome, busulfan-treated rats, and ovariectomized mice [46–48]. The fact that some older females eventually ceased reproduction and displayed oocyte-depleted ovaries appears inconsistent with the notion that the oocyte pool is continually regenerated by circulating stem cells in adult females [49]. Notably, the literature contains a report of a human female with premature ovarian failure that was homozygous for a potential functional mutation in Dmc1 [50].
Cytological analysis of Dmc1Mei11/+ oocytes revealed that a significant proportion had unrepaired breaks and asynapsed chromosomes at the time when most oocytes normally reach pachynema and diplonema. It is likely that these meiotic defects (particularly unrepaired DSBs) are sufficient to trigger checkpoint-mediated apoptosis in a subset of oocytes, leading to the observed shortfall of follicles. Some or most of the oocytes that proceeded past this checkpoint were capable of fertilization and development. We do not know if this entire cadre of “successful” oocytes was completely normal, or whether some escaped checkpoint elimination with a subcritical level of aberrations. Future experiments may determine whether mutant oocytes that survive postnatally retain meiotic errors and are subsequently either eliminated prior to fertilization via the spindle checkpoint or alternatively, survive to produce some level of aneuploidy in fertilized oocytes.
Our genetic and functional studies of Dmc1Mei11 provide proof that dominant mutations can arise in a key meiotic recombination gene such as Dmc1 and cause infertility in mice. Because of the high conservation in the process of meiosis, it is highly likely this could occur in humans with similar consequences (i.e., male infertility, premature ovarian failure, and possible defects in fertilizable oocytes). What makes such a mutation particularly insidious is that it is transmissible through populations via females, who would not manifest any apparent problems until well into their reproductive lifespans.
Human wt and A272P mutant cDNAs were cloned into the pET-15b vector (Novagen, http://www.emdbiosciences.com) to generate a protein linked to a histidine tag on the amino terminus. The proteins were overexpressed in E. coli BL21 (DE3) and purified using the following series of chromatographic steps: Ni-NTA-agarose (Qiagen, http://www.qiagen.com), heparin (FF Hiprep16/10; Bio-Rad, http://www.bio-rad.com), MonoQ, and Superdex 200. The proteins were concentrated and stored in buffer (20 mM Tris-HCl [pH 7.4], 300 mM NaCl, 10 % glycerol) at −80 °C.
These oligonucleotides were synthesized by MWG Biotech (http://www.mwg-biotech.com): 1, 100-mer, TGTGGAATGCTACAGGCGTTGTAGTTTGTACTGGTGACGAAACTCAGTGTTACGGTACATGGGTTCCTATTGGGCTTGCTATCCCTGAAAATGAGGGTGG; 2, 100-mer, CCACCCTCATTTTCAGGGATAGCAAGCCCAATAGGAACCCATGTACCGTAACACTGAGTTTCGTCACCAGTACAAACTACAACGCCTGTAGCATTCCACA; 3, 60-mer AATGTTGAATACTCATACTCTTCCTTTTTCAATATTATTGAAGCATTTATCAGGGTTATT.
Oligonucleotide 1 was 5′ end-labeled with (γ-32P)-ATP using standard labeling procedures. The labeled strand was then annealed to the complementary strand (oligonucleotide 2) to create blunt-ended dsDNA and gel purified in 7% polyacrylamide, 1× TAE buffer. Finally, all the duplexes were purified using ion exchange columns (MERmaid Spin Kit; Qbiogene, http://www.qbiogene.com). The single-stranded and the double-stranded 100-mer oligonucleotides were then mixed with either wt or A272P HuDMC1 in reactions containing 20 mM Tris-HCl (pH 7.4), 100 mM NaCl, 10% glycerol, and 1 mM MgCl2. The reaction mixtures were incubated at 37 °C for 10 min, and products were analyzed by electrophoresis in 8% polyacrylamide gels in 1× TAE buffer at 5 V/cm for 3 h. The formation of nucleoprotein complexes was quantitated using a BAS 2500 Bio-imaging Analysis System (Fuji Medical System, http://www.fujimed.com).
Supercoiled pUC18 plasmid DNA was purified by CsCl banding. Oligonucleotide 3 (60-mer; 0.82 μM nt) with homology to the pUC18 plasmid was radiolabeled and mixed with supercoiled pUC18 (8 μM bp). Reaction mixtures contained 1 mM DTT, 20 mM Tris-HCl (pH 7.4), 2.5 mM MgCl2, and 2 mM ATP with an ATP-regenerating system (7.5 mM creatine phosphate and 30 U/ml creatine kinase). hDMC1 and/or A272P hDMC1 were preincubated with the 60-mer oligo for 5 min, followed by incubation with dsDNA for 15 min at 37 °C. When both HuDMC1 and A272P hDMC1 were incubated together, 100 μg/ml BSA was added to the reaction mixture. Reactions were stopped by the addition of 0.5% SDS and 1 mg/ml proteinase K and incubation for 15 min at 37 °C. Products were resolved on 1% agarose gels in 1× TAE containing 3 mM MgCl2. Gels were dried to DE81 paper and analyzed with the BAS 2500 Bio-imaging Analysis System.
The amount of D-loop relative to dsDNA was calculated using this formula: % dsDNA in D-loop = A × 100 × (Ppiprod,i / Ppitotal,i), where Ppiprod,i are pixels per inch of oligonucleotide ssDNA in supercoiled dsDNA (plasmid), Ppitotal,i are the sum of pixels per inch of free oligonucleotide ssDNA and oligonucleotide ssDNA in supercoiled dsDNA, and A is the molar ratio of oligonucleotide ssDNA with respect to dsDNA in the reaction mixture.
The stock of mice containing the Mei11 mutation was derived from a G1 daughter of chimera v6.4-C7, as indicated in Table 1 of Munroe et al. [14], who segregated the spermatogonial depletion (Sgdp) mutation. Sgdp, like Mei11, also causes male-specific infertility, but not due to meiotic arrest. At first, only the Sgdp phenotype was seen and characterized histologically. Afterwards, in the course of mapping Sgdp, males were phenotyped on the basis of testis size, an autosomal dominant pattern of inheritance was evident, and linkage to Chr 15 was obtained [6]. However, upon histological analysis of the affected animals in the mapping cross, we realized that the phenotype was not that of Sgdp, but of Mei11. We cannot determine whether Sgdp and Mei11 were both present in the original pedigree, or if Mei11 arose thereafter. Initial genetic mapping indicated that the Mei11 mutation was segregating with B6 microsatellite alleles on distal Chr 15. For high-resolution genetic mapping, C3H-Mei11/+ females were backcrossed to C3H, and 810 offspring were genotyped for the presence of crossovers in the 7.5 Mb Mei11 critical region, between markers D15Mit68 and D15Mit107. Recombinant males were scored for the mutant phenotype by checking for presence/absence of epididymal sperm or examining testis histopathology. Recombinant daughters were backcrossed to C3H for phenotyping of recombinant sons.
The wt allele was identified using the forward primer 5′ TTGTGACCAATCAAATGACAG 3′ and the common reverse primer 5′ CCTAGGATCATCCCCCAAGT 3′. The Mei11 mutant allele was identified using the forward primer 5′ TTGTGACCAATCAAATGACAC 3′ and the common reverse primer listed above. The Dmc1− allele was genotyped as described [11].
Total RNA was isolated from adult mouse testes using RNA midi-prep columns (Qiagen). cDNA was generated from 2 mg of randomly primed RNA using Superscript II (Invitrogen, http://www.invitrogen.com).
Testes or ovaries were fixed in Bouin's solution, dehydrated, and embedded in paraffin wax. The embedded samples were sectioned at 5 μM and stained with hematoxylin and eosin or periodic acid–Schiff.
The methodology used for surface spreading and immunolabeling of spermatocyte and oocyte and meiotic chromosomes was as described [7,51]. Primary antibody sources and dilutions used in our experiments are as follows: mouse monoclonal anti-SYCP3 (1:500, ab12452; Abcam, http://www.abcam.com); mouse DMC1 monoclonal (1:250; P. Moens [Department of Biology, York University, Toronto, Ontario, Canada] and M. Tarsounas [University of Oxford, Oxfordshire, United Kingdom]; this antibody was immunodepleted for cross-reactive activity against RAD51 [52]); rabbit anti-SYCP1 (1:1,000, C. Heyting, Wageningen University, Wageningen, The Netherlands); rabbit anti-RAD51 (1:250, PC130; Oncogene Research Products, http://www.emdbiosciences.com/html/CBC/home.html; also cross-reactive with DMC1); rabbit anti-phosphorylated H2AX (γH2AX), serine 139 (1:500, 07–164; Upstate Biotechnology, http://www.upstate.com); rabbit anti-STAG3 (1:1,000; R. Jessberger, Mount Sinai School of Medicine, New York, New York, United States and Dresden University of Technology, Dresden, Germany); rabbit anti-MLH3 (1:400; P. Cohen, Cornell University, Ithaca, New York, United States); rabbit anti-RPA (1:200; C. Ingles, University of Toronto, Toronto, Ontario, Canada), and human anti-CREST (1:500; B. R. Brinkley, Baylor College of Medicine, Houston, Texas, United States). All secondary antibodies conjugated with either Alexa Fluor 488 or Alexa Fluor 594 (Molecular Probes, http://probes.invitrogen.com) were used at a dilution of 1:1,000. All images were taken with a 100× objective lens under immersion oil.
Paraffin-embedded ovaries (see above) were entirely serial sectioned at 6 μm. Follicle counts were carried out under standard light microscopy, with every tenth section counted. Ovarian follicles were scored and categorized as primordial, primary, secondary, early antral, and antral according to previously described morphological criteria [53]. Only those follicles with clear oocyte nuclei were counted. For a numerical estimate of each follicle class in the entire ovary, counts were then multiplied by a “correction factor” of ten.
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession number for the mouse Dmc1 transcript is NM_010059.
|
10.1371/journal.ppat.1003275 | Monomeric Nucleoprotein of Influenza A Virus | Isolated influenza A virus nucleoprotein exists in an equilibrium between monomers and trimers. Samples containing only monomers or only trimers can be stabilized by respectively low and high salt. The trimers bind RNA with high affinity but remain trimmers, whereas the monomers polymerise onto RNA forming nucleoprotein-RNA complexes. When wild type (wt) nucleoprotein is crystallized, it forms trimers, whether one starts with monomers or trimers. We therefore crystallized the obligate monomeric R416A mutant nucleoprotein and observed how the domain exchange loop that leads over to a neighbouring protomer in the trimer structure interacts with equivalent sites on the mutant monomer surface, avoiding polymerisation. The C-terminus of the monomer is bound to the side of the RNA binding surface, lowering its positive charge. Biophysical characterization of the mutant and wild type monomeric proteins gives the same results, suggesting that the exchange domain is folded in the same way for the wild type protein. In a search for how monomeric wt nucleoprotein may be stabilized in the infected cell we determined the phosphorylation sites on nucleoprotein isolated from virus particles. We found that serine 165 was phosphorylated and conserved in all influenza A and B viruses. The S165D mutant that mimics phosphorylation is monomeric and displays a lowered affinity for RNA compared with wt monomeric NP. This suggests that phosphorylation may regulate the polymerisation state and RNA binding of nucleoprotein in the infected cell. The monomer structure could be used for finding new anti influenza drugs because compounds that stabilize the monomer may slow down viral infection.
| The RNAs of negative strand RNA viruses are encapsidated by their specific viral nucleoproteins, forming helical nucleoprotein-RNA structures that are the template for transcription and replication. All these nucleoproteins have two activities in common: RNA binding and self-polymerisation, and it is likely that these activities are coupled. All these viruses have to keep their nucleoprotein from binding to cellular RNA and from polymerisation before viral RNA binding. The non-segmented viruses solve this by coding for a phosphoprotein that binds to the nucleoprotein, blocking both activities. The segmented viruses, such as influenza and Bunyaviruses, do not code for a phosphoprotein and need to solve this problem differently. Here we present the atomic structure of monomeric influenza virus nucleoprotein. Although the structures of the influenza virus and the Rift Valley Fever Virus (Bunya virus) nucleoproteins are different, there are functional similarities when the monomer and polymer structures are compared. Both nucleoproteins have a core structure that is identical in the monomer and the polymer. They contain a flexible arm that moves over to a neighbouring protomer in the polymer structure but that folds onto the core in the monomer structure, hiding the RNA binding groove in the Rift valley Fever Virus nucleoprotein and modifying the electrostatic potential of the RNA binding platform of the influenza virus protein.
| Negative strand RNA viruses have an RNA genome in the opposite sense of that of messenger RNA. Therefore, the first viral activity after entering the host cell is transcription by the viral RNA-dependent RNA polymerase. The template for transcription is a complex between the viral RNA and the nucleoprotein (NP) that binds to the RNA sugar-phosphate backbone [1], [2]. NP is necessary for RNA elongation by the polymerase [3], [4]. However, its main function may be to separate the newly made mRNA from the template RNA because the infecting viral replication complexes do not contain helicases and purified influenza virus NP melts dsRNA [1].
Negative strand RNA viruses include non-segmented viruses like the Rhabdoviridae (ex. vesicular stomatitis virus (VSV) and rabies virus) and the Paramyxoviridae (ex. Sendai and measles virus) and segmented viruses like the Arenaviridae (Lassa fever virus), the Bunyaviridae (Rift Valley fever virus (RVFV)) and the Orthomyxoviridae (influenza viruses). When expressed in a transfected cell in the absence of other viral components, the nucleoproteins of most of these viruses bind to cellular RNA and form nucleoprotein-RNA complexes that are indistinguishable from the viral complexes [5]. The formation of such complexes results from two coupled activities of the nucleoproteins: RNA binding and self polymerisation. In infected cells, these nucleoproteins bind almost exclusively to their viral RNAs and, therefore, all these viruses have developed a mechanism to stop their NPs from binding to cellular RNA and from polymerizing. The non-segmented viruses code for another viral protein, the phosphoprotein (P), that binds with its N-terminal end to RNA-free nucleoprotein, indicated by N0 [6]–[8]. The structure of the N0P complex of VSV shows how the P binding site overlaps with the RNA binding groove on the nucleoprotein and with one of the sites involved in nucleoprotein polymerisation, thus blocking both activities [9].
The segmented viruses do not code for an equivalent of a phosphoprotein and solve the problem in different ways. The nucleoprotein of RVFV has been crystallised in two forms; as a monomer and as a hexameric ring [10], [11]. In the ring, two N-terminal α helices of NP swing out to the back of a neighbouring protomer for self polymerisation. Inside the ring there is a continuous positively charged surface that binds the RNA [12]. In the monomeric form the two N-terminal helices fold onto the positively charged surface of their own protomer. Thus, the monomeric, closed form avoids at the same time RNA binding and polymerisation. It is likely that, in the infected cell, a signal on the newly produced RNA or the polymerase itself changes the conformation of the nucleoprotein so that it binds to the viral RNA and polymerizes. The Lassa fever virus nucleoprotein was crystallised in its intact form that shows a C-terminal exonuclease domain [13], [14] and an N-terminal domain. The intact protein crystallised as a circular trimer with the C-terminal domain of one protomer binding to the N-terminal domain of the next one [13]. Neither of the domains had RNA bound. When the N-terminal domain was expressed alone it did bind RNA [15] and, compared with the closed structure of the intact protein, secondary structure elements had moved away to open the RNA binding cleft. The C-terminal domain may control the opening and closing of the RNA binding groove and a linear association of nucleoprotein protomers could permit the binding of RNA. Thus, although the structures of the RVFV and Lassa virus nucleoproteins are totally different, both proteins show inactive conformations with a closed RNA binding site and conformations in which secondary structure elements have moved away to open the RNA binding site [5]. For both proteins it was also suggested that polymerisation and RNA binding are coupled, like for the nucleoproteins of the non-segmented viruses.
The nucleoproteins of influenza A H1N1 and H5N1 crystallised as trimers [16], [17] and NP of influenza B virus as a tetramer [18]. In these structures, each protomer binds with a domain exchanged C-terminal tail loop (residues 402–428 for A/H1N1) in a groove on the core of a neighbouring protomer. On the opposite side of the loop, on the core of the protein, there is a large and shallow positively charged surface. Mutating residues on this surface lowers RNA binding [17]. NP purified from virus by CsCl gradient centrifugation exists in an equilibrium between monomers and oligomers going from trimers and tetramers to large structures resembling viral ribonucleoprotein complexes [19]. Although recombinant NP is generally considered to exist only as trimers [16], [17], it was recently shown that there exists an equilibrium between monomers and trimers/tetramers [20], [21]. This equilibrium is shifted to the oligomeric state at 300 mM salt whereas a stable population of monomers was found at 50 mM salt [20]. For some mutants like R416A and E339A the equilibrium is shifted to the monomeric form [16], [20], [22]–[24] whereas for the Y148A mutant the equilibrium is shifted to the trimeric form [20].
Here we show the structure of the monomeric R416A mutant and describe the RNA binding characteristics of monomeric wild type NP. In the infected cell NP may be kept monomeric by post translational modification. Mass spectroscopy analysis on NP isolated from viral RNP showed phosphorylation of serine 165. By mutagenesis we generated the S165D mutant to mimic this phosphorylation. S165D was monomeric with the same biophysical characteristics as the R416A mutant but showed high cooperativity for RNA binding at concentrations above the Kd.
We previously showed that influenza virus NP can be stabilised as monomers in 50 mM salt and as trimers and tetramers in 300 mM salt [20] and wanted to further study their RNA binding behaviour. Wild type NP monomers and trimers/tetramers were incubated with an RNA oligonucleotide of 51 nucleotides [20]. Monomeric NP bound rapidly to the RNA and formed circular structures within 1 hour (Figure 1 left). These structures resemble circular recombinant mini RNPs [25]. The trimer/tetramer binds rapidly to RNA with a three-fold higher affinity than monomeric NP [20] but upon binding remained as trimers and tetramers up to 18 hours incubation (Figure 1 middle). However, when oligomeric NP was added to the RNA and then diluted to 50 mM NaCl, higher order polymers and rings formed within 1 hour like for the monomeric NP (Figure 1 right), most likely through dissociation of the oligomers upon which the monomers polymerised onto the RNA. In agreement with the results published in Tarus et al. [20] our results suggest that influenza NP is in equilibrium between monomers and oligomers but only monomers can form circular NP-RNA complexes.
In order to study monomeric NP we tried to crystallise monomeric wt NP but we obtained the same crystals and structure as was obtained for the wt trimer [16]. Thus, the crystallisation conditions seem to push monomeric NP to form trimers in the crystal. We then crystallised the obligate monomeric R416A mutant of NP. Influenza A/WSN/33 R416A mutant NP was concentrated to 10 mg/ml and crystallised by vapour diffusion using the sitting drop method. Crystals with space group C2221 diffracted to 2.7 Å (Table 1). The structure was solved by molecular replacement starting from the structure of wt A/WSN/33 NP (PDB 2IQH) in which residues 400 to 498 had been removed. Even though the crystals of wt and mutant NP have the same space group, wt NP crystallises as a trimer whereas mutant NP crystallised as a monomer but with 3 molecules in the asymmetric unit; the cell dimensions are different and the positions of the monomers are different from those of the protomers in the trimer. The structure (Figures 2B, C and Figure S1 in Text S1) shows residues 21–391 and 408–498. Most of the mutant protein structure is identical to that of the wt, from residue 22 to 385, the rmsd is 1.09 Å for 364 aligned Cα's. From residue 386 onward the structure is different because the exchange domain folds into the groove of its own protomer rather than in the groove of a neighbouring protomer. Figure 2 shows the comparison of the structures of trimeric wt and monomeric R416A NP. The core of the protein is in grey because this part of the structure is the same for the wt trimer and the mutant monomer, residues 386 to 401 are in green, the trimer exchange domain (residues 402–428) is in yellow, and residues 429–498 are in red (Figure S1 in Text S1). Figure 2A shows the trimer structure and only the residues 402–489 of the dark grey protomer are shown. In the orientation shown residues 386–401, the green part, are behind its own protomer and, thus, not visible. In the wt trimer (Figures 2A, 3D, E and Figure S1 in Text S1), the “green” part of the chain forms a random coil to residue 402 of the exchange domain. Then, residues 402–421 (yellow, figure 2A) of the dark grey protomer form a hairpin that binds in a groove on the surface of the light grey protomer [16], [17]. This domain exchange is terminated by α-helix 19 that is also bound to the light grey protomer. Then the chain (in red) loops back to its own protomer starting with α-helix 20. The remainder of the C-terminal domain forms a random coil bound to the surface of the core of NP.
In the monomeric mutant protein (Figures 2B, C, 3A, B and Figure S1 in Text S1), residues 386–390 form a β-strand (β7) and residues 392–407 form a flexible loop that points towards the RNA binding surface. Then, from the mutated R416 up to α-helix 19 the chain binds in the groove on its own surface making very similar interactions as the domain exchange hairpin of the wt trimer. α-Helix 19 binds at exactly the same place as in the trimer structure but on its own protomer rather than on its neighbour. α-Helix 20 binds in exactly the same position to its own protomer in the monomer structure and in the trimer structure (rmsd of 0.5 Å between the two structures). The remaining “red” strand goes towards β-strand 7 to fold into β-strand 9 forming a two stranded β-sheet.
Figure 3 compares the H1N1 R416A monomer structure with the trimer structures of H1N1 and H5N1. The major difference between the H1N1 and H5N1 trimer structures is that the exchange domains don't bind to the same neighbours and, thus, point into different directions (Figure 3D and G). For trimeric H1N1 the last visible C-terminal residue is 489 and for H5N1 this is residue 496 and in both structures the C-termini point away from the RNA binding surface. However, in the structure of R416A, the chain is visible until the penultimate residue and points into the RNA binding surface, reducing the space for RNA binding and changing the electrostatic characteristics of this surface (compare figures 3C, F and I, blue is positively charged). At the other side of the monomer the 392–407 loop also points into the RNA binding surface. Although we could not model this loop because its density is missing, the RNA binding surface may be reduced from both sides.
Mutations in the domain exchange loop and in the surface groove in which the loop binds do not only influence the stability of the trimer but also the stability of the monomer. In the trimer structures, R416 makes an ionic bond with E339 and it was assumed that R416A and E339A formed monomers because this bond in the trimer was disrupted [16], [23], [26]. In the R416A structure, E339 makes hydrogen bonds with R461 and the mutated R416A points towards R461; if the wt monomer structure were the same as the R416A structure there could be a clash between arginines 416 and 461. Therefore, the R416A mutation may both stabilise the monomer and destabilise the trimer.
Because we did not succeed in crystallising monomeric wt NP, we used a variety of biophysical methods to compare the wt and mutant monomers. Monomeric R416A and monomeric wt NP stabilised at low salt have an identical sedimentation behaviour with an S20,w of 4.3 S [20]. This S-value corresponds to a hydrodynamic radius of 3.3 nm. This means that the exchange domain of the wild type monomer does not extend out in solution in the same conformation as in the oligomeric form but must be close to the core of the protein like in the R416A structure. The circular dichroism spectrum of the trimer form of the wt protein is identical to that of the obligate Y148A trimer, both at 50 and 300 mM NaCl (Figure 4A). The spectra could not be measured below 200 nm because of the high NaCl concentration and the secondary structure content could not reliably be determined. The spectra had a clear α-helical signature with a strong minimum at 222 nm. All spectra of monomeric NP (wt NP at 50 mM salt and R416A mutant NP at both 50 and 300 mM salt) were identical. They showed the same value at 222 nm as oligomeric NP but had a more pronounced minimum at 207 nm (Figure 4B and C). Each of these experiments was repeated with at least three independent protein preparations and always gave the same results. This indicates that all monomeric proteins have the same secondary structure content, slightly different from the oligomeric proteins. We also used CD to determine the melting temperature by heating the proteins while measuring the helical content at 208 and 222 nm. Again, the wt and mutant monomers showed an identical behaviour with an apparent denaturation midpoint value of 43.5±0.5°C (average of 6 independent measurements) whereas the wt trimer denatured at 47±0.5°C. All results for the monomeric wt and mutant proteins are identical and different from those for the oligomeric form thus, it is likely that monomeric wt NP has a similar structure as the R416A mutant NP.
Because only monomeric NP forms RNP-like NP-RNA structures, NP newly produced in the infected cell should remain monomeric and free of nucleic acid until binding to viral RNA. However, accumulation of NP in the nucleus at physiological temperatures would likely result in the formation of trimers/tetramers and the affinity of monomeric NP for RNA is very high; Kd = 41 nM [20]. NP was reported to be phosphorylated on several serine residues [27], [28]. Phosphorylation was found to be a highly dynamic process and phosphorylated NP was also detected in RNPs [29]. We analyzed the NP isolated from influenza virus A/PR/8/34 RNPs by Mass Spectrometry. Several phosphorylated serines were detected, but only one residue is conserved in all A and B viruses: Serine 165. In the monomeric structure, this serine is situated at the interface between the two lobes of the core of NP, between α-helix 19 and residue Phe488, close to the C-terminus (Figure 5A). There is space for a phosphate group that could be stabilised by residue R267 nearby. In the trimer structure there does not seem to be enough space to locate a phosphate group because it would clash with main chain Ser407 and the negatively charged side chain of Glu405. We produced the S165D mutant mimicking the phosphorylation of Ser165. The mutant protein was monomeric at low and high salt when investigated by EM (Figure 5C), had the CD signature of a monomeric protein (Figure 4D), had a hydrodynamic radius of 3.3 nm and a thermal denaturation midpoint as determined by CD of 41±0.5°C and, thus, resembled the monomeric form of NP. The Kd for RNA of S165D was determined by surface plasmon resonance (SPR) using a 24 nt oligoribonucleotide [20] and by a filter binding assay using a radioactive panhandle RNA [1]. The SPR analysis gave a Kd of 730 nM (Figure S2 in Text S1); much higher than the Kd for wt monomeric NP of 41 nM [20]. The Kd derived from the filter binding experiment was in the same order of magnitude; 250 nM compared with 30 nM for wt recombinant NP and for NP isolated from virus (Figure S3 in Text S1). Thus, the affinity of the mutant was 10–20 times lower than that of wt protein. However, when studied by DLS at higher concentrations, the S165D mutant showed a much higher cooperativity upon RNA binding than wt NP. S165D NP polymerisation onto a 24-mer RNA oligonucleotide reached a plateau in 30 minutes compared to 2 hours for wt NP (Figure 5B). The same observation was made for the formation of NP-RNA rings as followed by negative staining EM. NP-RNA complexes formed immediately upon mixing of S165D NP with a 51 nt RNA oligo [20] (Figure 5C, T = 0) whereas complex formation with wt NP was slower (Figure 1). Thus, although the Kd for RNA is lower for the S165D mutant than for wt NP, the kinetics of assembly on RNA is more rapid.
Like for other negative strand RNA viruses, here we argue that influenza virus NP also exists in a monomeric form that is free of RNA and that only monomeric NP can form virus-like NP-RNA complexes. The structure of the monomer can perhaps best be described as self-inhibited in which the exchange domain that is involved in trimer formation takes up equivalent positions on its own core rather than on the core of a neighbouring protomer. A flexible loop formed by residues 392–407 and the C-terminus point towards the RNA binding surface which may be the reason for the lowered affinity for RNA, in particular for the obligate monomeric mutant R416A [20], [22], [24]. Wild type monomeric NP can bind oligonucleotides of 8 residues while remaining monomeric. However, when the RNA binding site is saturated with 24+ nucleotides [30], [31], the monomer oligomerises [20], possibly because the RNA pushes out secondary structure elements that stabilise the monomer. The stability of the wt monomer is enhanced by post-translational modification of serine 165. The biochemical behaviour of the monomeric S165D mutant is different with a lowered affinity for RNA but with an enhanced polymerisation at RNA concentrations above the Kd compared to wt monomeric NP. Recently, the lab of Ervin Fodor published the phosphoproteome of influenza A and B viruses [32]. Among various sites they also observed phosphorylation of S165. Recombinant virus with a S165A mutation could not be recovered indicating the importance of this phosphorylation site for the activity of the protein. Using RNP reconstitution experiments in 293 T cells they could measure transcription and replication for the S165A but not for a S165E mutant. As mentioned above the Ser to Asp mutation can be accommodated into the monomer structure but a mutation to Glu would lead to steric hindrance. While this paper was under review, two papers were published describing the structure of the intact viral RNP [33], [34]. Both papers describe that the distance between the NP protomers in the RNP is larger than the distance between the protomers in the trimers that are free from RNA [16] (Figure S4 in Text S1). This could imply that the insertion of the exchange domain into a neighbouring NP protomer in the RNP is not equivalent to that in the RNA-free trimer and this may be related to our findings that the S165D mutant does not form trimers but can polymerize onto RNA and that trimers can bind RNA but do not form NP-RNA rings.
Influenza NP binds to RNA without sequence specificity [1]. The only RNA sequence specificity for influenza virus proteins seems to reside in the polymerase that binds 5′ viral RNA and 5′-3′ panhandle structures [35]–[39]. In a recent model for influenza virus replication it was suggested that soluble, newly produced polymerase binds to the newly replicated 5′ end after which NP polymerises onto the elongated replicate [40], [41]. NP binds to the polymerase with a loop containing residues R204, W207 and R208 that is disordered in both the trimer and the monomer structures [42]. The mobility of this loop was recently suggested to influence RNA binding affinity [43]. It is possible that binding of NP to the polymerase leads to opening up of the RNA binding site after which the NP binds cooperatively to the primed NP-RNA site. Phosphorylation and de-phosphorylation of NP probably plays a regulatory role in vRNA encapsidation.
Because of its importance during the viral life cycle, NP is widely recognised as an antiviral target. Through high-throughput testing Kao et al. and Su et al. identified “nucleozin” that has antiviral activity and appears to aggregate NP in the cell leading to interference with nuclear import [44], [45]. A later crystal structure and mutational analysis showed that two NP trimers stick together into a hexamer by six nucleozin-derived molecules, with two complementary sites per NP protomer [46]. The crystal structure of the trimer has also been used for structure-aided drug design [26]. Peptides derived from the exchange domain (residues 402–428) interfere with polymerisation of NP and viral replication and molecules designed to interfere with the E339-R416 salt bridge have antiviral effect. The structure of the NP monomer presented here may serve for the design of new antiviral molecules. In particular, drugs that stabilise the monomer should have an antiviral effect.
The full-length NP gene of the H1N1 strain A/WSN/33 with a 6-His-tag at its C-terminal end was cloned in the pET22 vector (Novagen) under the control of a T7 promotor. The R416A, Y148A and S165D mutations were introduced by using PfuUltra DNA polymerase with the QuikChange II site-directed mutagenesis kit (Stratagene). Escherichia coli BL21 (DE3) cells carrying the plasmids were induced 4 hours by adding 1 mM isopropyl-β-D-thiogalactopyranoside (IPTG) at 37°C or 12 h at 28°C (mutant proteins) and collected by centrifugation. The pellet was resuspended and sonicated in a lysis buffer composed of 50 mM Tris at pH 7.4 with 300 mM NaCl, 15 to 30 mM imidazole, 1 M NDSB201 (Sigma), 5 mM β-ME and 5 mM MgCl2. The protein was purified by Ni2+ affinity chromatography (Ni-NTA, Qiagen) followed by a Heparin column (GE Healthcare). The protein was then dialyzed against 20 mM Tris pH 7.4 and 50 mM or 300 mM NaCl. The last purification step was size-exclusion chromatography using a Superdex S200 column. The protein was eluted in high or low salt according to the required polymerisation state. The protein concentration was determined by using the extinction coefficient ε = 55500 M−1.cm−1 at 280 nm.
The R416A mutant protein was crystallized by vapor diffusion using the sitting drop method. The crystals were obtained in 0.1 M Hepes pH 7.5, 1.2 M potassium sodium tartrate with a protein concentration of about 10 mg/ml. Data were collected at the ESRF (beamline ID14-4) and processed with the XDS package [47], [48]. The structure was solved by molecular replacement using the wild-type H1N1 nucleoprotein structure (PDB ID code 2IQH) as a model. Model building and refinement were performed using CCP4i suite program for crystallography (MOLREP, REFMAC, COOT) [49]. The coordinates have been deposited in the Protein Data Bank under PDB ID code 3zdp. The protein structure figures were made using PyMOL [50].
Samples were applied between a carbon and a mica layer and negatively stained with 2% (w/v) sodium silicotungstate (pH 7.0). The carbon film was covered by a copper grid and air dried. Micrographs were recorded with a JEOL 1200 EX II microscope at 100 kV with a nominal magnification of 40,000×. Micrographs were taken on a 2K×2K CCD camera (Gatan Inc.). The RNA binding kinetics was performed with a protein solution at 100 µM and an RNA of 51 nucleotides [20] with a final ratio NP/RNA of 3/1. For the time points for EM analysis a fraction of the mix was diluted to have a final protein concentration between 10 and 20 µg/ml. The dilutions were made with a buffer with the same salt concentration except for the test of salt dilution on the monomer-oligomer equilibrium where the dilution was done with 20 mM Tris-HCl pH 7.4 in order to reduce the salt concentration from 300 to 50 mM NaCl.
The measurements were performed on a Malvern nanosizer instrument thermostated at 20°C, in 20 mM Tris pH7.5, 50 or 300 mM NaCl according to the sample. The protein concentrations were in a range of 5 to 50 µM and for each sample, 12 to 18 scans were averaged. All experiments were repeated at least 3 times. The scattering intensity data were processed using the instrumental software to obtain the hydrodynamic radius and the size distribution of scatterers in each sample.
Protein samples were centrifuged at 16000× g and the protein concentration in the supernatant was adjusted to 0.5 mg/ml. CD experiments were performed on a JASCO-810 spectrometer at 20°C with a 1 mm path-length quartz cell using a bandwidth of 1 nm, an integration time of 1 second and a scan rate of 50 nm/min. Each spectrum is the average of 10 scans. All spectra were corrected by subtracting the buffer spectrum acquired under the same conditions. All data were normalized to mean residue ellipticity. The melting temperature was obtained for each protein sample by measuring the CD signal at 208 nm and 222 nm from 20°C to 80°C every 2°C. At 80°C all proteins precipitated. The value of the denaturation midpoint (°C) is the average of at least 3 measurements on independent protein preparations.
NP from egg grown influenza A/PR/8/34 virus was cut out from a Coomassie-stained 1D SDS-PAGE gel and digested with Trypsin (Roche Applied Science). The digest was analyzed by LC-ESI-MS/MS with an Ion trap MS HCT ultra PTM Discovery System (Bruker Daltonics, Bremen, Germany) coupled with a Nano-LC 2D HPLC system (Eksigent, Dublin, CA, USA). A CapRod monolithic C18 column (100 µm×15 cm, Merck, Darmstadt, Germany) was used to separate the peptides. The gradient was 10–70% ACN within 20 min at 300 nl/min flow rate. The top two precursor ions were selected over m/z range from 400 to 1200 for fragmentation. Fragmentation was performed subsequently for 60 ms over 300 to 2000 m/z. The raw data were processed with DataAnalysis (Version 3.1, Bruker, Bremen Germany). The extracted MS/MS data were submitted to MASCOT (version 2.103, Matrix Science, London, UK) in-house server via Biotools (version3.0, Brucker, Bremen, Germany). Proteins were identified by searching the peptide lists against SwissProt. The following parameters were used: Taxonomy: Viruses; Enzyme: trypsin; Max Missed Cleavages: 2; Variable modifications: oxidation (M); carbamidomethyl (C); phospho (ST), phospho(Y); Peptide Mass Tolerance: ±0.5 Da: Fragment Mass Tolerance: ±0.5 Da.
|
10.1371/journal.pcbi.1002126 | Towards a System Level Understanding of Non-Model Organisms Sampled from the Environment: A Network Biology Approach | The acquisition and analysis of datasets including multi-level omics and physiology from non-model species, sampled from field populations, is a formidable challenge, which so far has prevented the application of systems biology approaches. If successful, these could contribute enormously to improving our understanding of how populations of living organisms adapt to environmental stressors relating to, for example, pollution and climate. Here we describe the first application of a network inference approach integrating transcriptional, metabolic and phenotypic information representative of wild populations of the European flounder fish, sampled at seven estuarine locations in northern Europe with different degrees and profiles of chemical contaminants. We identified network modules, whose activity was predictive of environmental exposure and represented a link between molecular and morphometric indices. These sub-networks represented both known and candidate novel adverse outcome pathways representative of several aspects of human liver pathophysiology such as liver hyperplasia, fibrosis, and hepatocellular carcinoma. At the molecular level these pathways were linked to TNF alpha, TGF beta, PDGF, AGT and VEGF signalling. More generally, this pioneering study has important implications as it can be applied to model molecular mechanisms of compensatory adaptation to a wide range of scenarios in wild populations.
| Understanding how living organisms adapt to changes in their natural habitats is of paramount importance particularly in respect to environmental stressors, such as pollution or climate. Computational models integrating the multi-level molecular responses with organism physiology are likely to be indispensable tools to address this challenge. However, because of the difficulties in acquiring and integrating data from non-model species and because of the intrinsic complexity of field studies, such an approach has not yet been attempted. Here we describe the first example of a global network reconstruction linking transcriptional and metabolic responses to physiology in the flatfish, European flounder, a species currently used to monitor coastal waters around Northern Europe. The model we developed has revealed a remarkable similarity between network modules predictive of chemical exposure in the environment and pathways involved in relevant aspects of human pathophysiology. Generally, the approach we have pioneered has important implications as it can be applied to model molecular mechanisms of compensatory adaptation to a wide range of scenarios in wild populations.
| Modelling the responses and compensatory adaptations of living organisms to a changing environment is extremely important both in terms of scientific understanding and for its potential impact on global health. Although computational modelling of ecological systems has been utilised in ecotoxicology, the application of systems biology approaches to non-model organisms in general presents formidable difficulties, partly due to limited sequence information for environmentally relevant sentinel species. Moreover, the number of samples and the depth of information available are often limited and there may be a lack of truly relevant physiological endpoints. Thus, omics have proven effective in finding responses of aquatic organisms to model toxicants in laboratory-based experiments [1] but the environment poses a greater challenge as anthropogenic contaminants are present as complex mixtures and responses will additionally be dependent upon natural life history traits and other environmental factors.
Relatively few omics studies have focussed upon the ecotoxicology of environmentally sampled fish [2]–[7]. Although we have previously shown [8], [9] that expression of stress response genes could be used to distinguish fish from environmental sampling sites with different underlying contaminant burdens, this gave little insight to the health outcomes of these molecular differences. In this context, identifying molecular mechanisms of compensatory and toxic responses from observational data (reverse engineering), an approach that has been so successful in clinical studies and in laboratory model organisms, is highly challenging in field studies. We addressed this challenge by developing a novel network inference strategy based on the integration of multi-level measurements of populations of fish exposed to a diverse spectrum of environmental pollutants. This provides a useful model for a network biology approach generally applicable to non-model species and represents a breakthrough in the way we study the mechanisms whereby organisms respond to chemical exposure in the environment. We directed our efforts towards modelling molecular networks representative of populations of the flatfish European flounder (Platichthys flesus) sampled from marine environments of North Western Europe, including locations significantly impacted by anthropogenic chemical contaminants.
The study integrated measurements representing a broad spectrum of samples characterized using transcriptomics, metabolomics, conventional biomarkers and analysis of chemicals in sediments from the sampling sites. Previous studies have shown both anthropogenic contamination and higher prevalence of pre-neoplastic and neoplastic lesions in flounder from the Elbe estuary [10] and from the Mersey and Tyne [11], together with elevated levels of hepatic DNA adducts at these sites [12]. Data integration was achieved by implementing a systems biology framework for network reconstruction, starting from cross-species mapping of sequence information to the integration of multi-level datasets within a framework for network inference [13] and culminating in the identification of network modules predictive of physiological responses to chemical exposure, valuable for marine monitoring [14].
The networks we identified demonstrate a remarkable parallel between human liver carcinogenesis and environmental effects on fish liver as well as revealing potentially novel adaptation mechanisms. The broader application of network biology approaches to other non-model species sampled from the environment is therefore likely to profoundly change our understanding of how living systems are likely to adapt to complex environments.
An important assumption in many eco-toxicology studies is that the molecular states of organisms reflect their biological responses to complex chemical mixtures present within that environment. Indirect evidence suggests that this hypothesis may be correct. For example, consistent with previous studies [9], we have identified genes and metabolites that were differentially expressed between environmental sites (the results obtained are shown in detail in Table 1 and Text S1, Tables S1 and S2). Many of these were either known to be associated with stress responses or were previously shown to respond to anthropogenic chemical contaminants in fish. Although these results were encouraging they did not provide a direct link between molecular status and response to specific chemicals. Since sediment chemistry data was available, we assessed whether chemical contaminant profiles could be inferred from gene expression data and whether these would at least partially match the known sediment composition. Our analysis was performed by linking genes differentially expressed between each sampling site and the reference site, with chemical-gene relationships within the Comparative Toxicology Database (CTD) [15]. The Alde estuary was chosen as the reference site due to its low concentrations of major anthropogenic chemical contaminants (Table 1), both in sediment and in flounder livers. These significant associations may be regarded as predictive of the most important classes of chemicals exerting their biological effects upon flounder gene expression amongst the highly complex chemical mixtures within the sediments at these sites. Results were consistent with the initial hypothesis, as where contaminants were highlighted both by chemistry and CTD analysis; sediment concentrations all exceeded the lower OSPAR ecotoxicological assessment criteria, except for PAHs at the Morecambe site. At Brunsbuttel, elevated chromium and polychlorinated biphenyls (PCBs); at Cuxhaven, chromium, nickel, lead, zinc, polycyclic aromatic hydrocarbons (PAHs) and PCBs; at Helgoland nickel, zinc, manganese and PCBs, at Mersey PCBs; at Morecambe Bay arsenic, nickel and PAHs; and at Tyne arsenic and PCBs were all predicted by the CTD analyses and confirmed by chemistry data (Table 2 and Table S3). PAHs were predicted at Morecambe and Cuxhaven, with the AhR-inducer beta-naphthoflavone predicted at Brunsbuttel, Helgoland and Tyne, consistent with our finding of CYP1A transcriptional induction at all sites in comparison with the Alde. Additionally, Ingenuity Pathway Analysis (IPA) of all genes identified as significantly differentially expressed between sites showed significant associations with a number of toxicologically important processes and outcomes (Table 3).
As there were clear relationships between geographical location, chemical exposure and molecular profiles of flounder livers, we proceeded to reconstruct a network model representing the relationships between transcriptomic and metabolomic data, morphological measurements, protein biomarkers and microsatellite markers (Figure 1). This network was constructed from all data, not limited to molecules that differed between sites.
Inspection of the resulting network (Figure 2) showed that transcriptional and metabolic networks separated into two different areas of the network layout. Interestingly, modules whose hubs were fish morphometric measurements occurred exactly at the interface between these two areas and these modules contained metabolite (46%) and transcript (50%) measurements as well as fish morphometric measurements (4%).
Different areas of the inferred network (Figure 2) were characterised with different functional profiles. The modules close to the interface with metabolism (A) showed enrichment (FDR<0.05) for the annotation terms mitochondrion (GO:0005739), oxidoreductase activity (GO:0016491), endoplasmic reticulum (GO:0005783), protein folding (GO:0006457) and antioxidant activity (GO:0016209), and the two KEGG pathways hsa03050:proteasome and hsa00480:glutathione metabolism. The second sub-network (B) was enriched for immune response (GO:0006955) and response to stress (GO:0006950). The third sub-network (C) was enriched for proteolysis (GO:0006508) and digestion (GO:0007586).
Each individual network module was tested for its ability to predict geographic sampling sites (Figure 3), the presence of parasites (Figure 4A) and the presence of any of the liver histo-pathological abnormalities shown in Table 1H (Figure S2). Modules that were predictive of environmental sampling site were concentrated in two sub-networks. The larger (Figure 3, area A) was centred on the interface between metabolic and transcriptional networks and consequently included 14 modules consisting of morphometric indices as well as metabolites and transcripts (1% morphometric indices, 36% metabolite bins, 63% transcripts).
Modules that were predictive of parasitic copepod infection by Acanthochondria sp. and Lepeophtheirus sp. were similarly distributed in the network, but with additional modules localized in the sub-network B that were enriched in annotation to the immune response. Modules that were predictive of infection by Anisakid nematodes (Figure 4A) displayed a different profile, being more concentrated within group B. Hierarchical clustering of the profile of modules that were predictive of parasite infections showed that responses to copepod infections by Lepeophtheirus and Acanthochondria clustered together and were distinguished from responses to infection by the Anisakid nematodes.
We have previously shown that there is a strong link between laboratory exposure to individual chemicals and flounder hepatic gene expression [8]. It was therefore reasonable to hypothesize that genes differentially expressed in laboratory exposures may map onto modules predictive of sampling location. Fisher's Exact Test was therefore used to identify modules where genes differentially regulated as a result of single chemical laboratory exposures were over-represented (these were determined by ANOVA FDR<0.05 over 16 day time-courses post-intraperitoneal injection). Responses to lindane are shown as an example in Figure 4B. This highlighted the temporal change in responses to toxicants, with the majority of overlapping modules occurring in both sub-networks A and B at early timepoints, followed by a shift towards sub-networks B and C at later timepoints. We have previously shown [8] that this temporal change is associated with an early induction of transcripts for chaperones, phase I and II metabolic enzymes, oxidative stress and protein synthesis that diminishes by the later timepoints and is replaced by induction of protein degradation, immune-function and inflammation-related transcripts.
The results for all treatments are illustrated graphically in Figure S2 E to L. All treatments showed overlap with modules in group A, at the metabolite/transcript interface, and this was clearest for cadmium, that only affected this area, apart from one module in group C. All other treatments showed overlap between responsive genes and group B modules to varying extents and all except estradiol and cadmium overlapped with at least two modules within group C. These results are supported by our previous study [9] in which we found that employing transcripts altered during laboratory exposures to a range of individual toxicants improved predictivity of environmental sampling sites.
Having defined network modules predictive of geographical location, Ingenuity Pathway Analysis was used to elucidate the detailed structure of molecular pathways and their potential association with specific signatures of liver pathology. We performed these analyses under the hypothesis that the underlying response to chemical exposure would be consistent with what is known of human liver molecular pathophysiology. It was therefore expected that significant associations between the modules defined by our analysis and networks stored in the Ingenuity database would be informative of the underlying molecular mechanisms. We indeed observed a remarkable overlap between modules predictive of geographical location and modules containing genes whose transcriptional profile has been previously associated with liver fibrosis, cirrhosis and hepatocellular carcinoma in mammals.
Modules whose component genes related to hepatotoxicity are shown in Figure 5. The major group of site-predictive modules shows significant overlap with modules relating to liver cholestasis and hepatocellular carcinoma, whereas the secondary group overlaps with liver fibrosis. The annotation gained from Ingenuity, with key regulators inferred from networks based on interaction information, was combined and clustered in the TMEV software package using 5 different algorithms. These show (Table 4) that genes and metabolites a) involved in bile acid synthesis, transport and amino acid metabolism b) predictive of parasite infection c) linked to hepatocellular carcinoma, reproductive disorders and liver cirrhosis d) responding to oxidative stressors tert-butylhydroperoxide (tBHP) and cadmium, the hormone estradiol and rodent peroxisome proliferator perfluoro-octanoic acid (PFOA) are closely linked to differences between environmental sites. Additional relationships with inflammation, immune response, energy, fatty acids and nucleic acid metabolism, response to other toxicants and regulation by insulin, huntingtin, MYC and hepatocyte nuclear factor HNF4A were also highlighted. Functional analysis of the modules that were both site-predictive and associated with hepatocellular carcinoma showed significant overlap with mitochondrion, proteasome, tricarboxylic acid cycle, melanosome, protein dimerization activity, membrane-enclosed lumen, glutathione metabolism, coenzyme binding, microsome, translation, protein transport and carbohydrate catabolism (enrichment score >2, FDR<0.05).
The models we have developed are a high level representation of the molecular network's underlying response to environmental exposure. In order to generate specific hypotheses on the molecular pathways modulated during compensatory adaptation and toxicity further in-depth analyses of the specific interactions between genes and metabolites were performed. In this context, we combined the genes and metabolites represented in each group of predictive modules (Groups A and B in Figure 3) and input these to IPA software. The most statistically significant networks derived from each group are shown in Figure 6 and Figure S3, coloured by expression represented as a ratio between a highly polluted site and the reference site (Brunsbuttel versus Alde). The component genes and metabolites were clustered and the resulting expression profiles are shown in Figure S1. The Ingenuity networks are further described in Text S1 and are discussed below.
This is the first network level analysis of an environmental study integrating multilevel omic datasets. We discovered that the overall molecular state of the flounder liver (transcriptomics and metabolomics) is representative of the chemical contaminant burden of the sediments. Network reconstruction showed that the interface between transcriptional and metabolic network domains is linked to fish morphometric indices and is predictive of environmental exposure. In-depth analyses of predictive networks have identified putative novel pathways representative of responses to exposure. This approach provides a framework both for prediction of chemical pollutants in complex mixtures and for prediction of the health outcomes for exposed animals.
The chemical exposures predicted from CTD interactions were partly confirmed by chemical data (Table 2) despite the complexity of the environment, potentially including mixture effects, bioaccumulation and non-chemical stressors. Additional stressors were indicated that had not been chemically measured. Taking the Brunsbuttel site as an example, ethinyl-estradiol was a predicted contaminant and serum VTG protein, a canonical marker of endocrine disruption, was induced relative to the Alde (Table 1). Perfluorooctane sulfonic acid [16] and other persistent organic pollutants including PCBs, dieldrin and endosulfan [17] have been detected at elevated concentrations in the Elbe estuary and floodplain, and were all identified by our approach. Additional chemicals highlighted included systhane and vinclozolin fungicides, the halogenated aromatic hydrocarbon pesticide lindane, chlorine and tetradecanoylphorboyl acetate (TPA). It is uncertain whether these compounds are in fact present at this site as, for example, the presence of TPA appears unlikely. However TPA is a well-known tumour promoter [18], so detection of its associated gene expression changes might be viewed as a biomarker of effect, not necessarily of a specific exposure. At a number of sites flavonoids and flavonols, such as epicatechin gallate, were predicted, potentially indicating plant-derived exposures not of anthropogenic origin. At Morecambe Bay and the Tyne the prediction of paraquat perhaps reflected an oxidative stress response rather than the presence of this particular compound.
These results support the use of a knowledge-based approach to infer chemical exposure profiles from molecular responses and validate the underlying assumptions in the study. Predictions from interrogation of the CTD database (Table 2; Table S3) differed between sites suggesting that the approach can be sufficiently sensitive to specific differences in the exposure profiles. However, we do not propose that these associations necessarily indicate the presence of each specific contaminant at each site, for example ‘tobacco smoke pollution’ in the Mersey, we instead hypothesise that these represent the effects of related stressors, for example, AhR inducers at the Mersey site.
The development of a modular network, representing the integration between molecular and physiological readouts, provided us with an interpretive framework to analyse the complex molecular signatures linked to exposure. One of the most interesting findings is that the modules that predict environmental exposure with greatest accuracy represent the interface between metabolite and transcriptional networks and link to higher level indicators of fish health, such as condition factor and hepatosomatic index (Figure 2).
Consistent with this observation, network modules at the interface between metabolite and transcriptional networks were also differentially regulated in response to single chemical laboratory exposures. It should be borne in mind that the environmentally sampled fish have been chronically exposed to pollutants, and that chronic exposure can result in different responses than acute exposure [19], [20]. In addition bioavailability, mixture effects, metabolism and bioaccumulation affect compound-specific responses within the livers of these fish. This is illustrated by the modules containing genes that responded to 16-day treatments of flounder with individual toxicants (Figure 4 B, Figure S2). While all toxicants induced changes in the metabolic-interface genes, they also affected the secondary area of the network that related more to acute stress and immune response (Figure 2, area B), in contrast to the differences between environmental sites, where only one module (40) in this area was affected.
The characterisation of transcripts and metabolites that differed between sites was undertaken to provide insights into the molecular mechanisms that they describe, and to inform on the potential health outcomes for the fish. Canonical pathways that contributed to these differences included those relevant to metabolism of toxicants; AhR signalling, metabolism of xenobiotics by cytochromes P450, the NRF2-mediated oxidative stress response, glutathione metabolism and bile acid bioysnthesis (Table 3). Together these describe phase I and II metabolism of xenobiotics, such as aromatic hydrocarbons, and their excretion via the bile. Additional endobiotic metabolic pathways were affected. Changes in glycolysis, pyruvate metabolism, the citric acid cycle and oxidative phosphorylation implied disturbances to the energy pathways of the liver that could reflect the energetic requirements of xenobiotic metabolism and lead to further metabolic disruption. Changes in amino acid synthesis and proteasomal protein degradation also indicated reorganisation of metabolism.
This change in metabolic state and gene expression could be viewed as a successful compensatory response to toxicants and thus of little concern for the health of individual fish and these fish populations. Further examination of the annotation of transcripts and metabolites differing between sites implied that this hypothesis was false. As illustrated in Figure 5, and shown in Tables 3 and 4, there is a remarkable overlap between site-predictive modules and modules associated with hepatocellular carcinoma (HCC). Additionally, liver cholestasis -annotated modules overlapped with HCC and site predictive modules and this area of the network was highly associated with bile acid biosynthesis. Apart from this metabolic interface group only one other module (module 40) was predictive of site. This was also associated with hepatocellular carcinoma, and additionally with liver fibrosis, indicative of chronic liver damage, and occurred in an area of the network associated with inflammation. Therefore flounders inhabiting differentially contaminated sites show transcript and metabolite changes that have been associated with liver carcinogenesis in mammals. A question remains as to whether this simply represents the detection of HCC in the liver samples, as histopathology data were unavailable for the fish sampled off Germany. By comparison with studies of tumours from the closely related flatfish dab (Limanda limanda) this does not appear to be the case. In dab tumours the metabolites choline, phospocholine and glycine were reduced in concentration and lactate increased, an indication of the switch to anaerobic metabolism in the bulk tumours [21]. In, for example, the Brunsbuttel samples compared with non-tumour bearing Alde fish, choline, phosphocholine and glycine increased, and lactate decreased. Additionally, transcripts for ribosomal proteins showed co-ordinated induction in bulk tumours from dab, indicative of proliferation [22], but no such induction was apparent from the present samples. The changes in gene expression and metabolites detected in this study do not recapitulate those found in bulk tumours, and may be viewed as indicating either an earlier stage of tumourigenesis or a permissive micro-environment in which hyperplastic tissue may form and lead to tumour formation.
Ingenuity networks, based on mammalian interaction data, permitted more detailed biological characterisation of the site-associated modules. Complete pathways were not recapitulated by these analyses, as only a minority of the transcripts and metabolites from flounder liver were examined. Nevertheless, the analyses highlighted important processes and inferred key regulators. Here the most significant network derived from site-predictive modules is discussed in detail and additional networks are discussed in terms of their key inferred regulators.
The most striking finding from the Ingenuity analyses was the co-ordinated repression of proteasomal subunit genes at the Brunsbuttel site (Figure 6A; Figure S3A1). This was not so marked at other sites (Figure S1), indeed at Morecambe Bay these genes were induced in comparison with the Alde fish. Proteasome maturation protein (POMP) has been found to be a critical regulator of proteasomal activity [23] and has been shown to be repressed by the halogenated aromatic hydrocarbon 2,3,7,8-tetrachlorodibenzodioxin (TCDD) in an AhR-independent manner [24]. Although TCDD concentration was not measured, the mean expression of proteasomal genes was inversely correlated (r = −0.79) with fish liver PCB concentrations but did not correlate well with sediment PAH or PCB concentrations. Tyne fish, for example, displayed relatively high proteasomal gene expression and had low liver PCB but high PAH concentrations (Table 1, Figure S1). Therefore the repression of proteasomal genes may represent a halogenated aromatic hydrocarbon-related response (Figure 7). In trout Oncorhynchus mykiss, a proteasome inhibitor reduced PAH-dependent CYP1A induction [25], in contrast to mammalian studies [26]. This difference may contribute to the lower inducibility of CYP1A in flounder in comparison with many mammals. Ingenuity analysis also predicted an interaction between the proteasome and NF kappa B, a key regulator of mammalian hepatocarcinogensis [27]. The proteasome represses NK kappa B activation, and potentially disruption of proteasomal activity could have extensive additional effects on intracellular protein levels due to its role in the degradation of numerous proteins. We found no significant changes in NF kappa B gene expression between sites, and the consequences of putative activation at the Brunsbuttel site, and repression at the Morecambe site, due to changes in the proteasome, are difficult to predict, as in the early stages of carcinogenesis NF kappa B can have a protective effect, whereas in later stages it can promote tumourigenesis [27].
From the Ingenuity networks a number of key regulatory molecules were inferred. These included insulin (Figure 6B, Figure S3A2, S3A4, S3B1), estrone, luteinizing hormone (LH) and follicle stimulating hormone (FSH) (Figure S3A3 and S3A6), platelet derived growth factor beta (PDGFBB) (Figure 6B, Figure S3A2, S3B1), transforming growth factor beta (TGF-beta) (Figure S3A8), vascular endothelial growth factor (VEGF) (Figure 6B, Figure S3A7, S3B1), tumour necrosis factor (TNF) (Figure S3A5), and angiotensinogen (Figure 6B, Figure S3B1).
Insulin, in fish as in mammals, is a key hormonal regulator of energy, glucose and lipid metabolism, all pathways that were identified as affected by sampling site. By the Ingenuity networks it was linked to protein kinases, metabolites (including glucose and lactate) and the glucose transporter SLC2A4. The most obvious explanation for changes in insulin and related parameters would be differences in diet between fish from different sites. Amino-acid levels are more important regulators of insulin in carnivorous fish such as the flounder than sugars [28]. Dietary parameters would be expected to be highly variable depending upon recent feeding history of the fish, which was unknown for these individuals. However, insulin can also be modulated by exposure to toxicants including organophosphates [29] that was suggested to lead to an increase in lipogenesis, in agreement with our observations of phospholipidosis in fish from polluted sites (Table 1). Mild estrogenic endocrine disruption was suggested by VTG induction in Brunsbuttel fish, and networks shown in Figures S3A6 and S3A3 inferred that estrogen receptor alpha (ESR1), FSH and LH target genes were modulators of the different responses between sampling sites. ESR1 and HNF alpha were linked in Figure S3A6 and both are involved in hepatic cholestasis, indeed EE2-induced hepatotoxicity has been linked to alterations in bile acid biosynthesis in mice [30].
PDGFBB is the dimeric form of platelet derived growth factor beta (PDGF-B). Notably, PDGF-B over-expressing mice spontaneously developed liver fibrosis [31], and PDGF-BB was inferred as part of the network deriving from the liver fibrosis-annotated module 40 in our analysis. Additionally PDGF-B over-expressing mice developed hepatocellular carcinoma in response to phenobarbital and diethylnitrosamine treatment and induced TGF-beta and VEGF expression. TGF-beta was inferred to be an important regulator in site-specific responses (Figure S3A8) and is a well-known mediator of cancer initiation, progression and metastasis, via interaction with the inflammatory response [32]. Furthermore, the pro-inflammatory cytokine TNF-alpha, an initiating signal for the innate immune response in fish as well as mammals [33], was also identified by Ingenuity analysis (Figure S3A5). Release of TNF alpha from Kupffer cells leads to hepatocyte cell death, regeneration and fibrosis that can lead to hepatocellular carcinoma [34]. VEGF, best known as a stimulator of angiogenesis, was also highlighted in both the fibrosis-related and carcinoma-related sections of the network, and was linked with cell cycle, oncogenes and tumour suppressor genes (CDKN1A, TP53, MYC). Angiogenesis is a key requirement for the transition from fibrosis to hepatocellular carcinoma [35].
Angiotensinogen (AGT) is the precursor of angiotensin and was found to be repressed at all sites in comparison to the Alde reference site (Table S1). Angiotensin is a signal for vasoconstriction in mammals and in fish its expression is related to osmoregulation [7] with repression in liver in response to higher salinity. As the sampling sites differed in salinity, alteration of AGT transcription was not a surprising finding. As shown in Figure 6B, AGT was a member of the fibrosis-related module 40 and was predicted to form part of a complex network with VEGF, PDGF and intracellular kinases. Angiotensin has indeed been linked to stimulation of inflammatory liver fibrosis [36], via fibroblast proliferation and production of inflammatory cytokines and growth factors, including TGF-beta. Inhibition of the angiotensin system by antagonism of its receptor [37] or inhibition of angiotensin-converting enzyme [38] has been shown to reduce hepatic fibrosis.
VEGF, TGF beta, TNF alpha, PDGF and AGT are all intimately related to the progression of fibrosis to cirrhosis and hepatocellular carcinoma in mammals. These molecules were all highlighted as important regulators of the differences between molecular profiles of flounder livers from different sampling sites using an unbiased approach combining network inference and predictive algorithms.
A combination of omics, multiple biomarkers and bioinformatics were used to identify and characterise hepatic molecular changes between fish sampled from several environmental sites. Based on these data, parasite infection, fish morphology and genetics do contribute to the differences between sites, but do not explain the majority of changes seen. For example, within-site tests showed that morphometric parameters and parasite infections could be significantly associated only with a small proportion (<3%) of the gene expression differences between sites (Table S1, Table S4). Taken as a whole with our previous studies [8], [9], we find that anthropogenic chemical contamination of the marine environment is a major factor in explaining the molecular differences between fish sampled from these sites.
The different methodologies employed displayed different strengths and weaknesses. Histopathology was a good guide to broad levels of pollution effect, but provided little information upon the nature of the contaminant profile. Protein biomarkers and enzyme activities were useful for categorising sites by major classes of toxicant, but gave little information on the potential health outcomes. 1H NMR metabolomics showed low technical variability, and metabolite profiles alone were more predictive of sampling site than gene expression profiles alone, however the annotation of metabolites is not yet well advanced, limiting the functional information currently available. Transcriptomics exhibited higher variability than metabolomics, but was more informative due to better annotation. Overall the methodologies were highly complementary, allowing analyses that would be impossible if one were limited to a single technique.
The gene expression signatures associated with fish from each sampling site were used to predict the presence of chemical contaminants using the CTD gene expression-chemical interaction database. Mixture effects, other environmental influences and the similarity of certain stressors, such as the metals, might be expected to confound this approach. Additionally the incomplete nature of the flounder microarray and the CTD database and the limited numbers of samples for certain sites, which is a common issue in field studies, reduce the potential of this analysis. Therefore we did not expect to predict all environmental contaminants by this method. While this approach was useful with the current dataset, it may be expected to improve in future as both the CTD database and transcriptomic data become more comprehensive.
Data integration and network analyses were essential; both to predicting health outcomes and to identifying and examining affected biological pathways. They allowed visualisation of the highly complex dataset and facilitated comparison of the effects of different stimuli upon the model system. Modules associated with specific parameters could then be examined in detail, utilising interaction databases (Ingenuity) for further characterisation. Detailed examination of these networks illustrated the changes detected by broader classification of modules by annotation terms. In addition to potential interactions with diet and salinity, the majority of networks contained key regulators of inflammation, hepatic fibrosis and hepatocellular carcinoma. Therefore we propose that network biology approaches can lead to the identification of health impacts of environmental pollutants upon non-model organisms.
The molecular differences between reference and contaminated sampling sites were associated with carcinogenesis, and this outcome is supported by previous histopathology [10], [39]. Flatfish hepatic histopathology has long been associated with chemical contamination [39] and our results demonstrate the linkages between toxicants and histopathology via alterations in molecular signalling pathways and metabolism.
The sampling sites employed in this study were: In UK waters; on the Irish Sea, the Mersey estuary, at Eastham Sands, Liverpool (lat 53°19N, long 2°55W) and Morecambe Bay (lat 54°10N, long 2°58W); on the North Sea, the Alde estuary, Suffolk (lat 52°95N, long 01°33E) and the Tyne estuary at Howdon, Tyne and Wear (lat 54°57N, long 1°38W): In North Sea waters off Schleswig-Holstein, Germany; the Elbe estuary at Cuxhaven (lat 53°53N, long 08°15–19E) and Brunsbuttel (lat53°52N, long 09°09–10E) and off Helgoland (lat 54°06N, long07°15–08°00E). Adult European flounders (Platichthys flesus) were caught during statutory monitoring programs carried out by the Centre for Environment, Fisheries and Aquaculture Science (Cefas) at UK sites in April 2006 and by AWI Bremerhaven at FRG sites in October 2004 and April 2005. Fish were caught using beam trawls and held in tanks of flowing sea water onboard ship and were dissected either onboard ship or on return to shore. Livers were immediately removed and 100 mg samples for microarrays and metabolomics and 200 mg samples for biomarker assays were flash frozen in liquid nitrogen, with liver slices taken for histopathology. Blood was extracted and stored at 4°C overnight before plasma preparation for vitellogenin (VTG) analysis. Fin clips (1 cm2) were preserved in 70% ethanol at 4°C for genotyping. After sexing, livers from males that were 12 to 34 cm long were used for further analyses, this sample set included n = 20 for Alde, n = 16 for Tyne, n = 22 for Mersey, n = 23 for Morecambe Bay, n = 22 for Helgoland, n = 24 for Cuxhaven and n = 48 for Brunsbuttel. Fish lengths and weight, condition factor (K; body wt/length3×100), liver weight and hepatosomatic index (HSI; liver wt/body wt×100) were determined for all samples, gonad weight and gonadosomatic index (GSI; gonad wt/body wt×100) for FRG fish only.
Chemical determinations were carried out on sediment samples and independent sets of flounder liver samples from the same samplings by Cefas and Deutsches Ozeanographiches Datenzentrum, Germany and submitted to the International Council for the Exploration of the Sea (ICES), Copenhagen, Denmark as part of the national marine monitoring programmes. UK data was analysed from that collected as part of the Clean Safe Seas Environmental Monitoring Programme (CSEMP) and archived in the UK's Marine Environment Monitoring and Assessment National database (MERMAN). For sediment; metal concentrations (Al, As, Cd, Cr, Cu, Fe, Hg, Li, Mn, Ni, Pb and Zn); polycyclic aromatic hydrocarbons (PAHs) (anthracene, benzo[a]anthracene, benzo[a]pyrene, benzo[ghi]perylene, chrysene/triphenylene, fluoroanthrene, indo[123-c]pyrene, naphthalene, phenanthrene and pyrene); total organic carbon and polychlorinated biphenyls (PCBs) (congeners CB28, 52, 101, 118, 138, 153 and 180) were determined and the sum of PAHs and the sum of ICES 7 priority PCBs calculated for all sites. For flounder livers; metals (As, Cd, Cr, Cu, Fe, Hg, Ni, Pb, Se and Zn) and PAHs (acenaphthylene, acenaphthene, benzo[a]anthracene, C1-, C2- and C3- naphthalene, C1-phenanthrene/anthracene, chrysene, fluoroanthrene, fluorene, naphthalene, phenanthrene and sum of PAHs) were determined for Alde, Tyne and Mersey fish, with partial metal concentration data for Morecambe Bay, Helgoland, Cuxhaven and Brunsbuttel samples. Polychlorinated biphenyls (PCBs) (congeners CB28, 52,101, 118, 138, 153, 180 and sum of ICES 7 PCBs) were determined for liver samples from all sites. Data are available from the Merman database (http://www.bodc.ac.uk/projects/uk/merman/).
UK flounders were examined for external lesions, liver gross appearance and parasite infection. Liver pathology was assessed according to the criteria of Feist et al. [40]. Sections of liver tissue were removed, placed into individual histological cassettes, transferred to 10% neutral buffered formalin and processed for histopathology as described previously [11]. The presence of toxicopathic lesions, foci of cellular alteration, benign neoplasia, malignant neoplasia and non-specific inflammatory lesions was determined.
Plasma vitellogenin (VTG) concentrations (mg/ml) were determined by the method described by Kirby et al [41]. Hepatic metallothionein (MT) concentration (µg per mg) and glutathione reductase (GR) (nmol/mg), glutathione-S-transferase (GST) (µmol/mg) and ethoxyresorufin-o-deethylase (EROD) (pmol/mg) activities were determined by the methods of George and Young [42]. These assays were carried out for all except Cuxhaven and Helgoland fish.
Flounder fin-clip samples (n = 50) from all sites were surveyed for six neutral microsatellite markers (all polymorphic) and 13 detoxification gene-associated size variants within introns of flounder cytochrome P450 1A (CYP1A) [43], GST-A [44] and peroxisome proliferator activated receptors (PPAR) alpha, beta and gamma [45]. Following targeted PCR spanning each polymorphic site; DNA fragments were detected and sized by fluorescent capillary electrophoresis (Beckman CEQ8800 sequencer). Chromatogram files were individually inspected, and alleles were identified/scored manually. Four standard flounder DNA samples were analysed in each genotyping run (96 sample plate) to maintain scoring consistency. Standard genetic analyses for both single and multi-locus conformance to Hardy-Weinberg expectations within samples and examination of potential allelic differentiation among sites were undertaken using GENEPOP [46]. PHYLIPv3.5 software [47] was then employed to compute and compare four different measures of genetic distance (Nei's standard and Da distances; Cavalli-Sforza chord distance; Reynolds distance) and to construct unrooted neighbour-joined dendrograms (branch points being bootstrap-supported).
The GENIPOL flounder cDNA microarray has been described previously [48], [49]. The methods and design were similar to those employed in earlier experiments, with minor modifications [8], [9]. Briefly, liver tissue from individual flounders was homogenised in a methanol/water mixture [50] and aliquots were taken for both metabolomics and transcriptomics. Liver homogenates were used to prepare total RNA (Qiagen, Crawley, UK), reverse-transcribed to cDNA and labeled with Cy5-dCTP fluorophore (GE Healthcare, Amersham, UK). Labeled cDNAs were individually statically hybridised overnight to the microarray versus a common Cy3-labeled synthetic reference, before stringent washing and scanning (Axon 4000B; Molecular Devices, Wokingham, UK). Data were captured using Genepix software (Molecular Devices), and each slide was checked in detail, with spots showing poor morphology or arrays showing gross experimental artefacts being discarded. The data consisted of local background-subtracted median 635 nm intensities. MIAME-compliant gene expression data are available from ArrayExpress under accession E-MTAB-396. As the microarray is redundant, CAP3 clustering [51] had been used to identify contiguous sequences [48].
For metabolomics, liver homogenate aliquots were further extracted individually using methanol/chloroform/water (2∶2∶1.8 final volumes) [50], [52]. One-dimensional 1H NMR spectroscopy was performed upon the hydrophilic fraction as previously described [53]. Briefly, NMR spectra were measured at 500.11 MHz using an Avance DRX-500 spectrometer and cryogenic probe (Bruker, Coventry, UK), with 200 transients collected into 32k data points. NMR data sets were zero-filled to 64k points, exponential line-broadenings of 0.5 Hz were applied before Fourier transformation, and spectra were phase and baseline corrected, then calibrated (TMSP, 0.0 ppm) using TopSpin software (version 1.3; Bruker). The subsequent processing and statistical analyses of the NMR data have been described in detail in a previous study [53]. Briefly, taurocholic acid, an abundant bile acid with highly variable concentration in the liver extracts was subtracted from each spectrum using Chenomx NMR metabolomics software (version 4.6; Chenomx, Edmonton, Canada). Next, residual water was removed, each spectrum was segmented into 0.005 ppm bins, and the total area of each binned spectrum was normalized to unity so as to facilitate comparison between the samples. Subsequently to statistical analyses, significantly changing metabolite ‘bins’ were identified as particular metabolites by comparison with spectral libraries of reference compounds and were annotated with PubChem CID accessions (NCBI).
Microarray data were filtered to remove spots where 20% or more of the data were undetectable over all samples and background-subtracted intensity values of 0 or below were set to 0.5. Data were log2 transformed, quantile normalised and de-noised by a) removing data where SD/mean was more than 0.9 and b) removing data where maximum–minimum was less than 1.5. Missing data were estimated using MetaGeneAlyse probabilistic principal components analysis (PCA) algorithm [54]. Array slide batch effects were resolved using an empirical Bayes correction [55]. A representative clone with greatest average expression across all samples was chosen for each contiguous sequence cluster where the Pearson correlation score was greater than 0.6 to other members of the cluster. Where the correlation failed to pass this cut-off, data were discarded. The noise level for each metabolomics NMR spectrum was estimated by dividing the spectrum into 32 regions and calculating the smallest bin SD for each region and multiplying this by 3. These results were used to de-noise the data [56]. Data from metabolomics, transcriptomics and fish measurements (K, length, weight, liver weight, HSI) were then combined where all were available. The final data set therefore consisted of n = 15 for Alde, n = 9 for Tyne, n = 9 for Mersey, n = 13 for Morecambe Bay, n = 21 for Helgoland, n = 23 for Cuxhaven and n = 36 for Brunsbuttel. An additional dataset was generated for the omics samples that also possessed genetic data.
Normalised combined microarray and metabolomic data were input to Genespring GX 7.3.1 (Agilent Technologies, Santa Clara, CA, USA). Statistically significantly changing genes were found by 1-way ANOVA with a multiple testing correction [57] for a false discovery rate (FDR)<0.05, and with Welch T-tests employing the same FDR. Fold change cutoffs of 1.5-fold were additionally applied. A classification algorithm was used to compare previous data [9] with the current data; this employed the Support Vector Machines algorithm within Genespring with the Kernel Function Polynomial Dot Product (Order 3), a Diagonal Scaling Factor of 0, for all genes passing QC cut-offs in both experiments. Gene ontology (GO) analyses were carried out within Blast2GO [58], [59] employing the GOSSIP package [60]. As flounder is a non-model species, genes were annotated with gene symbols of their putative human orthologs, found by employing a Conditional Stepped Reciprocal Best Hit approach between flounder and zebrafish (Danio rerio) and human transcriptome databases, similar to Herbert et al. [61], with additional manual curation.
Chemical-gene expression interactions were downloaded from the Comparative Toxicology Database (CTD) [15], for all annotated genes. These represent a database of the previous literature on chemical-gene expression interactions. The chemical-gene pairs from this list were segregated into inducers and repressors, duplicates were removed, and the two lists uploaded into TMEV [62], thereby annotating each gene with its ‘chemical inducers’ and ‘chemical repressors’. Lists of genes (ANOVA, FDR<0.05, fold change versus Alde>1.5; illustrated in Table S1) were interrogated for enrichment of chemical associations using EASE (Expression Analysis Systematic Explorer) within TMEV, and FDR calculated. Where associations were found between an inducing chemical and induced genes and also between the same chemical acting as a repressor and repressed genes, FDRs were multiplied to produce a final FDR value. Lists of genes and metabolites were additionally interrogated by Ingenuity Pathway Analysis (Ingenuity IPA 8.5; Ingenuity Systems, Redwood City, CA, USA), employing Human Gene Organisation (HUGO) gene identifiers and PubChem CID compound identifiers, with statistical tests using Benjamini and Hochberg multiple testing corrections.
The overall approach taken for network construction and analysis is shown in Figure 1. It is conceptually sub-divided into: 1) Selection of network hubs: 2) Construction of a fully connected network: 3) Identification of network modules representing the neighbourhood of the hubs: 4) Assembly of the final modules and graphical representation: 5) In-depth analysis of gene interactions using Ingenuity Pathway Analysis (IPA) software.
The network was constructed from all measured variables, including transcript, metabolite, morphometric, protein biomarker and genetic data. Within the network each individual variable is described as a node. We selected 99 ‘hub’ nodes representing transcripts with known toxicological and regulatory relevance in order to identify the molecular network representing the interactions between these hubs and all the other nodes in the multi-level dataset. In addition, morphometric indices and metabolite peaks were also included in the list of hubs to represent the complexity of the metabolic networks, which, we reasoned are likely to closely influence liver physiology. The network inference methodology ARACNE [13] was used to create the network. Statistically significant interactions were selected on the basis of mutual information between the nodes at cut-off of P<1e-6. We defined 99 modules derived from each selected hub and its neighbourhood. Many nodes were present in multiple modules. The overlap index was calculated between each pair of modules by dividing the number of overlapping nodes by the total number of nodes in the smaller module. The final network was then constructed as the union of all network modules and visualized using a force driven layout available in the software application Cytoscape [63]. In the final network, the edge distances between the modules are relative to the overlap index and the node sizes are relative to the size of the module. We also compared this strategy to develop mutual information-based modules from hub variables with a more complex method [64], shown in Text S1, and discovered that they both gave similar results.
Subsequently the multivariate selection algorithm GALGO [65] was applied to each module to determine its predictivity for parameters including fish sampling site, parasite infection, and the presence or absence of liver pathologies. The cut-off employed for identification of predictive modules was >70% specificity and >70% sensitivity. Genes were annotated with HUGO identifiers for their putative human orthologs. DAVID v 2008 and v6.7 [66], [67] was used to classify module genes and groups of modules inferred from the topology of the module graph, by their shared Gene Ontology (GO) and other annotation terms. Flounder laboratory treatment data was employed to relate gene expression changes seen in the environmental samples to those elicited by model toxicant treatments. These treatments consisted of a single intraperitoneal injection with cadmium chloride (Cd, 50 µg/kg), 3-methylcholanthrene (3-MC, 25 mg/kg), aroclor 1254 (50 mg/kg), tert-butyl-hydroperoxide (tBHP, 5 mg/kg), lindane (25 mg/kg), perfluoro-octanoic acid (PFOA, 100 mg/kg), estradiol (l0 mg/kg) and furunculosis vaccine (killed Aeromonas salmonicia, Aquavac Furovac 5; 1 ml/kg) with gene expression monitored over a 16-day timecourse versus appropriate controls. Full details are shown in Williams et al. [48], Williams et al. [8] and Diab et al. [49], with data available from ArrayExpress under accessions E-MAXD-32 and E-MAXD-38. Overlap between module genes and genes differentially expressed by toxicant treatments over the 16 day timecourse, ANOVA, FDR<0.05 was determined by Fisher's Exact Test with a cut-off of P<0.01. Modules and groups of modules were interrogated by Ingenuity Pathway Analysis. Key regulatory molecules were inferred from networks generated within Ingenuity, that also output functional enrichment within the lists of nodes (P<0.05). Modules were annotated with associated diseases, functions, canonical pathways, toxicity and hepatotoxicity terms within Ingenuity and with inferred key regulators, as well as with environmental and parasitological predictivity and overlap with laboratory treatment data to produce a binary matrix. This was clustered within TMEV [62] using hierarchical clustering, SOTA self organising tree, K-means, QT and SOM self organising map algorithms. Grouped modules that were predictive of sampling site were subjected to Ingenuity analyses and were overlaid with Brunsbuttel expression data relative to Alde, the reference site. These genes and metabolites were subjected to K-means clustering within TMEV and the clusters functionally annotated using DAVID.
|
10.1371/journal.ppat.1002708 | Induction of GADD34 Is Necessary for dsRNA-Dependent Interferon-β Production and Participates in the Control of Chikungunya Virus Infection | Nucleic acid sensing by cells is a key feature of antiviral responses, which generally result in type-I Interferon production and tissue protection. However, detection of double-stranded RNAs in virus-infected cells promotes two concomitant and apparently conflicting events. The dsRNA-dependent protein kinase (PKR) phosphorylates translation initiation factor 2-alpha (eIF2α) and inhibits protein synthesis, whereas cytosolic DExD/H box RNA helicases induce expression of type I-IFN and other cytokines. We demonstrate that the phosphatase-1 cofactor, growth arrest and DNA damage-inducible protein 34 (GADD34/Ppp1r15a), an important component of the unfolded protein response (UPR), is absolutely required for type I-IFN and IL-6 production by mouse embryonic fibroblasts (MEFs) in response to dsRNA. GADD34 expression in MEFs is dependent on PKR activation, linking cytosolic microbial sensing with the ATF4 branch of the UPR. The importance of this link for anti-viral immunity is underlined by the extreme susceptibility of GADD34-deficient fibroblasts and neonate mice to Chikungunya virus infection.
| Nucleic acids detection by multiple molecular sensors results in type-I interferon production, which protects cells and tissues from viral infections. At the intracellular level, the detection of double-stranded RNA by one of these sensors, the dsRNA-dependent protein kinase also leads to the profound inhibition of protein synthesis. We describe here that the inducible phosphatase 1 co-factor Ppp1r15a/GADD34, a well known player in the endoplasmic reticulum unfolded protein response (UPR), is activated during double-stranded RNA detection and is absolutely necessary to allow cytokine production in cells exposed to poly I:C or Chikungunya virus. Our data shows that the cellular response to nucleic acids can reveal unanticipated connections between innate immunity and fundamental stress pathways, such as the ATF4 branch of the UPR.
| During their replication in host cells, RNA and DNA viruses generate RNA intermediates, which elicit antiviral responses mostly through type-I interferon (IFN) production [1], [2]. Several families of proteins are known to sense double-stranded RNA (dsRNA), including endocytic Toll-like receptor 3 (TLR3) [3], the dsRNA-dependent protein kinase (PKR) [4] and the interferon-inducible 2′-5′-oligoadenylates and endoribonuclease L system (OAS/2-5A/RNase L) [5]. Viral dsRNA and the synthetic dsRNA analog polyriboinosinic:polyribocytidylic acid (poly I:C) are also detected by different cytosolic DExD/H box RNA helicases such as the melanoma differentiation-associated gene 5 (MDA5), DDX1, DDX21, and DHX36, which, once activated, trigger indirectly the phosphorylation and the nuclear translocation of transcription factors such as IRF-3 and IRF-7, resulting predominantly in abundant type-I IFN and pro-inflammatory cytokines production by the infected cells [1], [6], [7].
Alphaviruses such as Chikungunya virus (CHIKV) are small enveloped viruses with a message-sense RNA genome, which are known to be strong inducers of type-I IFN in vivo [8], [9], a key response for the host to control the infection [10], [11], [12]. In vitro, however, response to RNA viruses is heterogeneous, since Sindbis virus (SINV), do not elicit detectable IFN-α/β production in infected of murine embryonic fibroblasts (MEFs) [13]. The specific points of blockage of type-I IFN production during infection are still not well delineated, but SINV and other alphaviruses could antagonize IFN production by shut-off of host macromolecular synthesis in infected cells [14], [15], [16]. Recently, human fibroblasts infection by CHIKV was shown to trigger abundant IFN-α/β mRNA transcription, while preventing mRNA translation and secretion of these antiviral cytokines [13], [15]. Contrasting with these reports, other groups using different CHIKV strains have observed abundant type-I IFNs release in the culture supernatants of CHIKV-infected human monocytes [17], human lung cells (MRC-5), human foreskin fibroblasts and MEFs [10]. Type-I IFN stimulation of non-hematopoietic cells has also been shown to be essential to clear infection upon CHIKV inoculation in mouse, but CHIKV was found to be a poor inducer of IFN secretion by human plasmacytoïd dendritic cells [10]. Thus, great disparities regarding alphavirus-triggered IFN responses exist between viral strains and the nature of host cells or animal models.
Once bound to their receptor on the cell surface (IFNAR), type-I IFNs activate the Janus tyrosine kinase pathway, which induces the expression of a wide spectrum of cellular genes including Pkr [18]. These different genes participate in the cellular defense against the viral infection. PKR is a serine–threonine kinase that binds dsRNA in its N-terminal regulatory region and induces phosphorylation of translation initiation factor 2-alpha (eIF2α) on serine 51 [19], [20], leading to protein synthesis shut-off and apoptosis. PKR has been also been shown to participate in several important signaling cascades, including the p38 and JNK pathways [21], as well as type-I IFN production [22], [23]. Inhibition of translation, IFN responses and triggering of apoptosis combine to make PKR a powerful antiviral molecule, and many viruses have evolved strategies to antagonize it [24], [25]. Interestingly, several positive RNA-strand viruses (eg. Togaviridae or Picornaviridae) have been shown to activate PKR, resulting in phosphorylation of eIF2α and host translation arrest [26], while viral mRNA could initiate translation in an eIF2-independent manner by means of a dedicated RNA structure, that stalls the scanning 40S ribosome on the initiation codon [25].
Despite the existence of these viral PKR-evading strategies, the importance of PKR for type-I IFN production has been strongly debated over the years and even considered dispensable since the discovery of the innate immunity function of the DExD/H box RNA helicases [27], [28]. However, several PKR-deficient cell types have reduced type-I IFN production in response to poly I:C [23], [29], [30], while PKR was demonstrated to be required for IFN-α/β production in response to a subset of RNA viruses including Theiler's murine encephalomyelitis, West Nile (WNV) and Semliki Forest virus (SFV), but not influenza, Newcastle disease, nor Sendai virus [31], [32], [33], [34]. These studies raise therefore the possibility that some but not all viruses induce IFN-α/β in a PKR-dependent and cell specific manner. Infection of PKR or RNAse L deficient mice demonstrated that these enzymes were not absolutely necessary for type I IFN-mediated protection from alphaviruses such as SFV or WNV, but still contributed to levels of serum IFN and clearance of infectious virus from the central nervous system [25], [35]. Similarly, deficient mice for both PKR and RNAse L showed no increase in morbidity following SINV infection, although, like during WNV infection, increased viral loads in draining lymph nodes were observed [35], [36]. These observations support a non-redundant and cell specific role for these enzymes in the amplification of type-I IFN responses to viral infection more than in their initiation [31], [32], [35]. Nevertheless, the exacerbated phenotypes observed upon alphavirus infection of mice deficient for type-I IFN receptor (IFNAR), underlines the limits of the individual contributions of PKR and RNAse L to the anti-viral resistance of adult animals [10], [35], [36].
In addition to dsRNA detection, different stress signals trigger eIF2α phosphorylation, thus attenuating mRNA translation and activating gene expression programs known globally as the integrated stress response (ISR) [37]. To date, four kinases have been identified to mediate eIF2α phosphorylation: PKR, PERK (protein kinase RNA (PKR)-like ER kinase) [38], GCN2 (general control non-derepressible-2) [39], [40] and HRI (heme-regulated inhibitor) [41], [42]. ER stress–mediated eIF2α phosphorylation is carried out by PERK, which is activated by an excess of unfolded proteins accumulating in the ER lumen [38]. Activated PERK phosphorylates eIF2α, attenuating protein synthesis and triggering the translation of specific molecules such as the transcription factor ATF4, which is necessary to mount part of a particular ISR, known as the unfolded protein response (UPR) [43], [44]. Interestingly DNA viruses, such as HSV, that use the ER as a part of its replication cycle, have been reported to interfere with the ER stress response through different mechanisms, such as the dephosphorylation of eIF2α by the viral phosphatase 1 activator, ICP34.5 [45], [46].
We show here, using SUnSET, a non-radioactive method to monitor protein synthesis [47], that independently of any active viral replication, cytosolic poly I:C detection in mouse embryonic fibroblasts (MEFs) promotes a PKR-dependent mRNA translation arrest and an ISR-like response. During the course of this response, ATF4 and its downstream target, the phosphatase-1 (PP1) cofactor, growth arrest and DNA damage-inducible protein 34 (GADD34, also known as MyD116 and Ppp1r15a) [48], are strongly up-regulated. Importantly, although the translation of most mRNAs is strongly inhibited by poly I:C, that of IFN-ß and Interleukin-6 (IL-6) is considerably increased under these conditions. We further demonstrate that PKR-dependent expression of GADD34 is critically required for the normal translation of IFN-ß and IL-6 mRNAs. We prove the relevance of these observations for antiviral responses using CHIKV as a model: we show that GADD34-deficient MEFs are unable to produce IFN-ß during infection and become permissive to CHIKV. We further show that CHIKV induces 100% lethality in 12-day-old GADD34-deficient mice, whereas WT controls do not succumb to infection. Our observations demonstrate that induction of GADD34 is part of the anti-viral response and imply the existence of distinct and segregated groups of mRNA, which require GADD34 for their efficient translation upon dsRNA-induced eIF2α phosphorylation.
We monitored protein synthesis in MEFs and NIH-3T3 cells after poly I:C stimulation, using puromycin labeling followed by immunodetection with the anti-puromycin mAb 12D10 [47]. Poly I:C delivery to MEFs and NIH-3T3, rapidly and durably inhibited protein synthesis, concomitant with increased eIF2α phosphorylation (P-eIF2α) (Fig. 1A and Fig. S1A). In MEFs, a strong eIF2α phosphorylation was observed after 4 h of poly I:C treatment, followed by a steady dephosphorylation at later times (Fig. 1A). Protein synthesis arrest was confirmed in individual cells by concomitant imaging of poly I:C delivery, mRNA translation and P-eIF2α (Fig. 1B and Fig. S1B), and with a wide range of dsRNA concentrations (Fig. S1C). Poly I:C-induced eIF2α phosphorylation and subsequent translation arrest were not observed in PKR-deficient MEFs (Fig. 1C and 1D), while eIF2α phosphorylation induced by the UPR-inducing drug thapsigargin (th) (an inhibitor of SERCA ATPases) or arsenite (as) was unchanged in PKR−/− cells (Fig. 1C). PKR is therefore necessary to induce protein synthesis inhibition in response to cytosolic poly I:C.
When levels of IFN-ß were quantified in culture supernatants and compared to total protein synthesis intensity, we found that most of the cytokine production occurred after 4 to 8 h of pIC delivery (Fig. 1E, WT, and S1D), a time at which mRNA translation was already considerably decreased (Fig. 1A and S1E). We measured the amount of cytokine produced in NIH-3T3 cells at a time (7 h) at which translation was already strongly inhibited (Fig. 1G and 1F). To prove that IFN-β production truly occurred during this poly I:C-induced translation arrest, cells exposed for 7 h to poly I:C were washed and old culture supernatants replaced with fresh media for 1 h (with or without CHX), prior translation monitoring (Fig. 1F, right) and IFN-ß dosage (Fig. 1G, right). We observed that close to 30% of the total IFN-ß produced over 8 h of poly I:C stimulation is achieved during this 1 h period, despite a close to undetectable protein synthesis in the dsRNA-treated cells (Fig. 1F). The neo synthetic nature of this IFN was further demonstrated by the absence of the cytokine in CHX-treated cell supernatants. IFN-β production in response to poly I:C is therefore likely to be specifically regulated and occurs to a large extent independently of the globally repressed translational context. As previously observed in MEFs, IFN-ß production in response to poly I:C was independent of PKR (Fig. 1E) [31]. This suggests that although its production occurs during cap-mediated translation inhibition, it does not directly depend on a specialized open reading frame organization, as described for the translation of the mRNAs coding for the UPR transcription factor ATF4 or the SV 26S mRNA upon eIF2α phosphorylation [26], [49]. This hypothesis is also supported by the ability of MEFs expressing the non-phosphorylatable eIF2α Ser51 to Ala mutant (eIF2α A/A), to produce normal levels of IFN-ß in response to poly I:C (Fig. 1E), while global translation was not inhibited by poly I:C in these cells (Fig. S2).
We went on to investigate the molecular mechanisms promoting this paradoxical IFN-ß synthesis in an otherwise translationally repressed environment. Induction of eIF2α phosphorylation by PERK during ER stress promotes rapid ATF4 synthesis and nuclear translocation, followed by the transcription of many downstream target genes important for the UPR [50]. Similarly, in presence of PKR, nuclear ATF4 levels were found to be up-regulated in MEFs responding to cytosolic poly I:C, albeit less importantly than upon a bona fide UPR induced by thapsigargin (Fig. 2A).
One of the key molecules involved in the control of eIF2α phosphorylation is the protein phosphatase 1 co-factor GADD34, which relieves translation repression during ER stress by promoting eIF2α dephosphorylation [50], [51],[52]. GADD34 is a direct downstream transcription target of ATF4 [53]. Expression of GADD34 was quantified by qPCR and immunoblot in WT and PKR−/− MEFs (Fig. 2B). In WT cells GADD34 mRNA expression was clearly up-regulated (20 fold) in response to poly I:C, while GADD34 protein induction was equivalent in poly I:C- and thapsigargin-treated cells. GADD34 mRNA transcription and translation were not observed in PKR−/− cells responding to poly I:C, but occurred normally upon thapsigargin treatment, paragoning eIF2α phosphorylation (Fig. 2B, right).
We next investigated the importance of ATF4 for GADD34 transcription by monitoring the levels of GADD34 mRNA in ATF4-deficient cells. ATF4−/− MEFs displayed higher basal levels of GADD34 mRNA than WT cells. However, in absence of ATF4, MEFs were unable to efficiently induce GADD34 mRNA transcription in response to any of the stimuli tested (Fig. S3). GADD34 mRNA expression was induced only 2 fold in ATF4−/− MEFs exposed to poly I:C, suggesting that its transcription is mostly dependent on ATF4 in this context. We further investigated P-eIF2α requirement for GADD34 expression and found that eIF2α A/A expressing MEFs were incapable of up-regulating GADD34 in response to poly I:C (Fig. 2C). Phosphorylation of eIF2α by PKR in response to cytosolic poly I:C induces therefore a specific integrated stress response (ISR), that allows ATF4 translation, its nuclear translocation and subsequent GADD34 mRNA transcription.
We next evaluated the relevance of GADD34 induction, by treating WT and GADD34ΔC/ΔC fibroblasts with poly I:C or with drugs known to induce ER stress, such as thapsigargin and the N-glycosylation inhibitor tunicamycin [52]. As expected, in WT cells eIF2α phosphorylation was rapidly increased in response to all ISR-inducing stimuli and decreased concomitantly with the expression of GADD34 over time (Fig. 3A and S4) [52]. Consequently eIF2α phosphorylation was greatly increased in GADD34ΔC/ΔC MEFs in all the conditions tested (Fig. 3A and S4A).
In thapsigargin-treated cells, protein synthesis was reduced in the first hour of treatment and rapidly recovered (Fig. 3B) [54]. Poly I:C, however, nearly completely inhibited translation despite active eIF2α dephosphorylation. This was particularly obvious when poly I:C was co-administrated together with thapsigargin. Indeed, poly I:C dominated the response by preventing the translation recovery normally observed after few hours of drug treatment (Fig. 3B). Surprisingly, in absence of functional GADD34, although eIF2α phosphorylation induction by poly I:C was augmented dramatically, no further decrease in protein synthesis was observed upon treatment of GADD34ΔC/ΔC cells with the dsRNA mimic (Fig. 3A and 3C). The functionality of GADD34 in translation restoration was, however, fully demonstrated, when the same cells were treated with thapsigargin, and protein synthesis was completely inhibited by this treatment [52] (Fig. 3C). Thus, cytosolic dsRNA delivery induces a type of protein synthesis inhibition, which requires eIF2α phosphorylation for its initiation, but conversely cannot be reverted by GADD34 induction and subsequent GADD34-dependent eIF2α dephosphorylation.
The potential contribution of the OAS/2-5A/RNAse L system to this P-eIF2α-independent inhibitory process was evaluated by investigating RNA integrity in MEFs exposed to poly I:C. We used capillary electrophoresis to establish precise RNA integrity numbers (RIN) computed from different electrophoretic traces (pre-, 5S-, fast-, inter-, precursor-, post-region, 18S, 28S, marker) and quantify the degradation level of mRNA and rRNA potentially resulting from the activation of this well characterized anti-viral pathway. No major RNA degradation could be observed upon poly I:C delivery (Fig. S5), suggesting that global RNA degradation does not contribute extensively to the long term translation inhibition observed upon poly I:C delivery in our experimental system.
We have observed that GADD34 expression counterbalances PKR activation by promoting eIF2α dephosphorylation, however it has little impact on reversing the global translation inhibition initiated by poly I:C. We next monitored the production of specific proteins and cytokines in WT and GADD34ΔC/ΔC MEFs (Fig. 4). Cystatin C, a cysteine protease inhibitor was chosen as a model protein, since its secretion ensures a relative short intracellular residency time so that its intracellular levels directly reflect its synthesis rate [55]. This is confirmed by the N-glycosylated- and total Cystatin C accumulation in cells treated with brefeldin A (Fig. 4A, left panel). Cystatin C levels were found to follow a similar trend to that observed with total translation, being strongly reduced upon poly I:C exposure and not profoundly influenced by GADD34 inactivation (Fig. 4A, right panel). Thapsigargin treatment induced a brief drop in cystatin C levels, prior to some levels of GADD34-dependent recovery. 6 hours of tunicamycin treatment affected more cystatin C accumulation than anticipated (Fig. 4A, right panel), probably due to interference with the N-glycosylation and associated folding of this di-sulfide bridge containing protein [55], thereby promoting its degradation by endoplasmic reticulum-associated protein degradation (ERAD) [56].
We next turned towards PKR, which displayed a pattern of expression completely different from cystatin C (Fig. 4B). As expected from its IFN-inducible transcription, levels of PKR were increased in poly I:C-treated MEFs (Fig. 4B), despite the strong global translation inhibition observed in these cells (Fig. 3). GADD34 inactivation appeared to influence the accumulation of PKR, since the cytoplasmic dsRNA sensor levels were not up-regulated and even decreased in poly I:C-treated GADD34ΔC/ΔC MEFs (Fig. 4B). Control treatment with tunicamycin and thapsigargin did not alter significantly PKR levels (Fig. 4B), suggesting that ER stress did not influence the kinase expression. The absence of PKR up-regulation in the poly I:C-treated GADD34ΔC/ΔC MEFs led us to investigate the capacity of these cells to produce anti-viral and inflammatory cytokines, which normally drive PKR expression through an autocrine loop. We ruled out any interference from the UPR in triggering IFN-ß production in our experimental system, since, as anticipated from PKR expression, tunicamycin and thapsigargin treatments were not sufficient to promote cytokine production in MEFs (Fig. S6) [43], [44].
We therefore investigated IFN-ß and IL-6 production in response to dsRNA in WT, GADD34ΔC/ΔC and CReP−/− MEFs. CReP−/− MEFs were used as a control, since CReP (Ppp1r15b) is a non-inducible co-factor of PP1 and displays some functional redundancy with GADD34 [57]. Although basal levels of eIF2α phosphorylation were higher in CReP−/−, PKR expression and translation inhibition upon poly I:C delivery were equivalent in WT and CReP−/− MEFs (Fig. S7A and S7B). Quantification of IFN-ß and IL-6 levels in culture supernatants indicated that, although abundant and comparable amounts of these cytokines were secreted by WT and CReP−/− cells, they were both absent in poly I:C-treated GADD34ΔC/ΔC MEFS (Fig. 4C and S7C).
Quantitative PCR analysis revealed that, IFN-ß, IL-6 and PKR transcripts were potently induced in poly I:C treated GADD34ΔC/ΔC MEFs (Fig. 4D), thus excluding any major transcriptional alterations in these cells, as confirmed by the normal levels of cystatin C mRNA, which remained constant in all conditions studied. Moreover, using confocal immunofluorescence microscopy, we could not detect intracellular IFN-β in poly I:C-stimulated GADD34ΔC/ΔC MEFs, in contrast to WT cells, which abundantly expressed the cytokine, despite the global translation arrest (Fig. S8). Thus, we could attribute the deficit in cytokine secretion of the GADD34ΔC/ΔC MEFs to a profound inability of these cells to synthesize cytokines, rather than to a defect in transcription or general protein secretion.
GADD34 induction by poly I:C is therefore absolutely necessary to maintain the synthesis of specific cytokines and probably several other proteins in an otherwise translationally repressed context. Importantly, GADD34 exerts its rescuing activity only on a selected group of mRNAs including those coding for IFN-ß and IL-6, but not on all ER-translocated proteins, since cystatin C synthesis was strongly inhibited by poly I:C in all conditions tested.
Interestingly, in GADD34ΔC/ΔC MEFs, PKR mRNA strongly accumulated in response to poly I:C (Fig. 4D), despite the absence of detectable IFN-ß production and PKR protein increase (Fig. 4B). This continuous accumulation of PKR mRNA in response to poly I:C suggests the existence of alternative molecular mechanisms, capable of promoting PKR mRNA transcription and stabilization independently of autocrine IFN-β detection. Nevertheless in these conditions PKR expression, like IFN-β, was found to be dependent on the presence of GADD34 for its synthesis (Fig. 4B).
Recent results indicate that PKR participates to the production of IFN-α/ß proteins in response to a subset of RNA viruses including encephalomyocarditis, Theiler's murine encephalomyelitis, and Semliki Forest virus [31]. Even though IFN-α/ß mRNA induction is normal in PKR-deficient cells, a high proportion of mRNA transcripts lack their poly(A) tail [31]. As GADD34 induction by poly I:C was completely PKR-dependent, we wondered whether the phenotypes observed in PKR−/−cells and GADD34ΔC/ΔC MEFs could be related. Oligo-dT purified mRNA extracted from cells exposed to poly I:C were therefore analyzed by qPCR. PolyA+ mRNAs coding for IFN-ß and IL-6 were equivalently purified and amplified from WT and GADD34ΔC/ΔC MEFs (Fig. S9). This confirms that albeit the phenotypes of PKR−/− and GADD34ΔC/ΔC cells might be linked, mRNA instability is not the primary cause of the cytokine production defect observed in GADD34ΔC/ΔC. Taken together these observations suggest the existence of a specific mRNAs pool, encompassing cardinal immune effectors such as IFN-ß, IL-6, and PKR, which are specifically translated in response to dsRNA sensing and increased levels of P-eIF2α. This mRNAs pool requires GADD34 for their translation during the global protein synthesis shut-down triggered by dsRNA detection.
We verified that GADD34 inactivation, and no other deficiency, was truly responsible for the loss of cytokine production by complementing GADD34ΔC/ΔC MEFs with GADD34 cDNA prior poly I:C delivery. IFN-ß secretion was partially restored in transfected GADD34ΔC/ΔC cells while eIF2α was efficiently dephosphorylated in both WT and GADD34ΔC/ΔC transfected MEFs (Fig. 4E). To further demonstrate that the phosphatase activity of GADD34 controls cytokine production upon dsRNA detection, we treated WT MEFs with guanabenz, a small molecule, which selectively impairs GADD34-dependent eIF2α dephosphorylation [58]. Upon treatment with this compound, a dose dependent inhibition of IFN-ß secretion was observed in poly I:C-treated MEFs, confirming the importance of GADD34 in this process (Fig. S10).
Fibroblasts of both human and mouse origin constitute a major target cell of Chikungunya virus (CHIKV) during the acute phase of infection [59]. In adult mice with a totally abrogated type-I IFN signaling, CHIKV-associated disease is particularly severe and correlates with higher viral loads. Importantly, mice with one copy of the IFN-α/ß receptor (IFNAR) gene develop a mild disease, strengthening the implication of type-I IFN signaling in the control of CHIKV replication [59]. Recently, human fibroblasts infection by CHIKV was shown to induce IFN-α/ß mRNA transcription, while preventing mRNA translation and secretion of these antiviral cytokines. CHIKV was found to trigger eIF2α phosphorylation through PKR activation, however this response is not required for the block of host protein synthesis [15].
We tested the importance of PKR during CHIKV infection by infecting WT and PKR−/− MEFs with CHIKV-GFP, at a multiplicity of infection (MOI) of 10 and 50. Productive infection was estimated by GFP expression (Fig. 5A, left panel), while culture supernatants were monitored for the presence of IFN-β (5A, right panel). PKR was found to be necessary to control CHIKV infection in vitro, since at least 60% of PKR–inactivated cells were infected after 24 of viral exposure, compared to only 15% in the control fibroblasts population. WT MEFs produced efficiently IFN-β, while the hypersensitivity to infection of the PKR−/− MEFs was correlated to a reduced type-I IFN production capacity after infection. Thus, during CHIKV infection, PKR is required for normal IFN production by MEFs. We also monitored protein synthesis in infected WT and PKR−/− fibroblasts using puromycin labeling followed by immunofluorescence confocal microscopy (Fig. 5B). CHIKV-GFP positive PKR−/− MEFs were found to incorporate efficiently puromycin, while in their infected WT counterpart protein synthesis was efficiently inhibited. Thus CHIKV, in this experimental model, induces a PKR-dependent protein synthesis inhibition and is therefore particularly relevant to further confirm our observations on the role of GADD34 in controlling type-I IFN production during response to viral RNAs.
GADD34ΔC/ΔC MEFs were exposed to CHIKV-GFP (MOI of 10 or 50) for 24 and 48 h. Productive infection was estimated by GFP expression and virus titration (Fig. 6A), and culture supernatants monitored for the presence of type-I IFN (Fig. 6B, left). Only minimal CHIKV infection (15%) could be observed at maximum MOI in WT MEFs (Fig. 6A, left), while robust IFN- β amounts were already produced at the lowest MOI (Fig. 6B). Contrasting with WT cells and regardless of the MOI used, a higher level of viral replication was observed in GADD34ΔC/ΔC MEFs (Fig. 6A). The GADD34-inactivated cells were clearly more sensitive to CHIKV, displaying a 50% infection rate after 24 h of infection (MOI 50) and a log more of virus titer in culture supernatants (Fig. 6A, right). Correlated with their susceptibility to CHIKV infection, IFN-β production was nearly undetectable in GADD34ΔC/ΔC MEFs (Fig. 6B). Such observation confirms the incapacity of GADD34-deficient cells to produce cytokines in response to cytosolic dsRNA, a deficiency likely to facilitate viral replication. This interpretation is further supported by the abrogation of viral replication in both WT and GADD34ΔC/ΔC MEFs briefly treated with IFN-β (Fig. 6C). Thus, GADD34 inactivation does not favor viral replication per se, but is critical for type-I IFN production. Interestingly infection levels were found to be higher in PKR−/− than in GADD34 ΔC/ΔC MEFs, although this difference could be attributed to clonal MEFs variation, it more likely suggests that PKR-dependent translation arrest could be key in preventing early viral replication in this system. In addition, the relatively lower permissivity of GADD34ΔC/ΔC MEFs to infection at high MOI could indicate the existence of GADD34-dependent defense mechanisms, which could be independent from IFN production and eIF2-α dephosphorylation. To strengthen and generalize these observations, we treated a different strain of WT MEFs with guanabenz and examined the consequences for CHIKV infection. Biochemically, GADD34 expression was induced upon CHIKV infection, and guanabenz treatment resulted in a clear increase in eIF2α phosphorylation, demonstrating the importance of GADD34 in limiting this process during infection (Fig. 6D, right). As observed with GADD34ΔC/ΔC cells, pharmacological and RNAi inhibition of GADD34 was found to increase significantly the sensitivity of MEFs to infection, while reducing their IFN-β production (Fig. 6D and S10). Thus, induction of GADD34 and its phosphatase activity during CHIKV infection, in vitro, participates to normal type-I IFN production and control of viral dissemination.
Several components of the innate immune response have been shown to impact on the resistance of adult mice and to restrict efficiently CHIKV infection and its consequences in vivo [10]. We decided to investigate the importance of GADD34 upon intradermal injections of CHIKV to WT (FVB) and GADD34ΔC/ΔC mice. Neither strain of adult mice was affected by intradermal injections of CHIKV, with little statistically significant differences in the virus titers found in the different organs. Thus, GADD34 deficiency does not annihilate all the sources of type-I IFN in the infected adult animals, a situation exemplified by the capacity of GADD34ΔC/ΔC bone-marrow derived dendritic cells to produce reduced, but measurable IFN-β in response to poly I:C [60]. This also infers that the light impact of GADD34 inactivation on mouse development [61] does not render these animals more sensitive to CHIKV infection.
As in Humans, CHIKV pathogenicity is strongly age-dependent in mice, and in less than 12 day-old mouse neonates, CHIKV induces a severe disease accompanied with a high mortality rate [59]. GADD34 function was therefore evaluated in this more sensitive context by injecting intradermally CHIKV to FVB (WT) and GADD34ΔC/ΔC neonatal mice. As previously observed for C57/BL6 mice [59], when CHIKV was inoculated to FVB neonates, a rate of 50% of mortality was observed 3 days after the infection of 9-day-old mice, while 12-day-old pups were found essentially resistant to the virus lethal effect (Fig. 7A). Strongly contrasting with these results, all CHIKV infected GADD34ΔC/ΔC neonates died within 3–5 days post inoculation whatever their age (Fig. 7A). When infection was monitored 5 days post-inoculation of 12-day-old mice at, GADD34ΔC/ΔC pups displayed considerably more elevated CHIKV titers (10–100 folds) in most organs tested, including liver, muscle, spleen and joints, the later being primarily targeted by the virus (Fig. 7B, left). As expected, and in full agreement with the in vitro data, infected GADD34ΔC/ΔC tissues showed a considerably reduced IFN-ß production (40–50%) compared to control tissues (Figure 7B, right), while serum levels were reduced by 20% (not shown). Although Infectious virus was poorly detected in the heart of WT animals, elevated titers of virus were observed in the heart of GADD34-deficient pups, matching the limited production of IFN in this organ. We further investigated the possible pathological consequences of cardiac tissue infection by carrying-out comparative histopathology. Hearts of infected GADD34-deficient animals displayed severe cardiomyocytes necrosis with inflammatory infiltrates by monocytes/macrophages and very important calcium deposition (Fig. 8), all being indicative signs of grave necrotic myocarditis. As a consequence, the left ventricles were strongly dilated, being probably the cause of acute cardiac failures and of the important death rate observed in GADD34ΔC/ΔC infected pups. Histology of infected FVB mice hearts was, however, normal with only few inflammatory cells (mainly lymphocytes) observed in the close vicinity of capillaries.
GADD34 expression is therefore necessary to allow normal type-I interferon production during viral infection and to promote the survival of young infected animals. We could circumvent the age-related acquisition of viral resistance in GADD34ΔC/ΔC mice to 17 days, since mice inoculated at that age survived CHIKV inoculation. In these animals, 3 days post-infection, enhanced viral replication was observed in the spleen and muscles, matching the relatively low level of type-IFN production in these tissues (Figure 7C). Functional GADD34 is therefore required to mount a normal innate response against the virus, but in older mice type-I IFN production by non-infected innate cells is probably capable to gradually overcome GADD34-deficiency and limit viral proliferation in vital organs, such as the heart.
Translation inhibition occurs in response to stress, when other cellular activities have to be reassigned or suspended momentarily. We demonstrate here that the activation of PKR by cytosolic dsRNA results in a stress response, leading to ATF4 and GADD34 induction. GADD34 expression has been observed during the infection of cells by different types of viruses [62] or intracellular bacteria such as Listeria monocytogenes [63]. Our observations demonstrate that GADD34 expression is a direct consequence of PKR activation and dsRNA sensing. Interestingly, although GADD34 induction by poly I:C promotes eIF2α dephosphorylation, this is not sufficient to prevent global protein synthesis arrest. The uncoupling of efficient eIF2α dephosphorylation from global translation recovery in response to cytosolic poly I:C implies therefore the existence of additional mechanisms inhibiting global translation. The 2-5A/RNAse L pathway does not seem to be sufficiently active in our experimental setting to explain this prolonged protein synthesis inhibition. The cleavage or the inactivation of other translation factors could work in concert with eIF2α to block or affect the efficiency of other individual steps of mRNA translation [64]. For instance, the phosphorylation of translation elongation factor 2 (eEF-2) is also controlled by eIF2α phosphorylation. Thus, Thr56 phosphorylation of eEF-2, which is known to inhibit its translational function by reducing its affinity for ribosomes, could contribute directly to the protein synthesis inhibition induced by PKR activation [65]. Independently of general protein synthesis inhibition, eIF2α dephosphorylation is necessary for the production of specific proteins upon dsRNA-induced translation inhibition. As demonstrated for ATF4, translation of a given mRNA during stress could rely on the structure and organization of its coding sequence, as well as the presence of multiple alternative initiation codons [49]. Surprisingly, functional GADD34 expression was found necessary for the translation of IL-6, IFN-β, and PKR. This observation points to the existence of a distinct group of mRNAs efficiently translated upon dsRNA detection and dependent on GADD34 activity.
GADD34 is extremely short lived and has been shown to accumulate on the ER, when over-expressed [51]. GADD34 could mediate its activity at the ER level and influence differently eIF2α sub-cellular distribution according to the type, localization, and level of activity displayed by the different eIF2α kinases. The strong eIF2α phosphorylation mediated by PKR in response to poly I:C or viral infection and leading to the initiation of translation inhibition, could be circumvented through GADD34 activity solely at the ER level, thereby allowing local cytokine production in absence of other functional protein synthesis. This selectivity for translation of several specific mRNAs among other ER-secreted molecules suggests further that GADD34 dependent mRNAs might display specific features allowing their efficient identification by GADD34 and associated molecules, as well as allowing their translation in presence of minimal levels of active guanine nucleotide exchange factor eIF2B.
GADD34 and PKR are necessary to produce anti-viral cytokines during CHIKV infection, and probably other types of infection. PKR, ATF4 and GADD34 should therefore be considered as an essential module of the innate anti-viral response machinery. The importance of PKR in anti-viral type-I IFN responses has been the object of contradictory reports [30], [31], [66], [67]. Our observations, however, suggest that PKR function should be re-evaluated by integrating the impact of viral detection on cellular translation. In eIF2A/A and PKR−/− cells, cytokine transcription is induced normally following poly I:C detection by DExD/H box RNA helicases, while as expected in these cells, no eIF2α phosphorylation and subsequent host translation inhibition are observed. This lack of translation arrest in the absence of potent eIF2α phosphorylation allows for normal cytokine production during dsRNA detection, with no requirement for an operational GADD34 feedback loop. The importance of PKR and GADD34 for IFN-β and other cytokines production could therefore be directly linked to the efficiency of the cellular translation inhibition induced by RNA viruses, as exemplified here with CHIKV, which in MEFs strongly activates PKR and subsequent protein synthesis inhibition.
GADD34ΔC/ΔC neonates are extremely sensitive to CHIKV infection and display signs of acute myocarditis and ventricles dilatation probably causing recurrent cardiac failures. CHIKV cardiac tropism is not normally observed in WT mouse and inability of heart tissues to produce sufficient type-I IFN in GADD34ΔC/ΔC could allow abnormally high viral replication, myocarditis and dilated cardiomyopathy. Interestingly many cases of myopericarditis induced by CHIKV and leading to dilated cardiomyopathies in infected patients have been reported since the 1970s after the different western Indian Ocean islands and Indian subcontinent disease outbreaks [68], [69]. These particular symptoms and complications might therefore be the consequences of great variation in the tissue-specific type-I IFN levels induced in CHIKV-infected patients, who might display particular polymorphisms in their innate viral sensing pathways increasing their peculiar susceptibility to viral dissemination in the heart.
Importantly, our data reveal a link between pathogen-associated molecular patterns (PAMPs) and the UPR through the activation of the eIF2-α/ATF4 branch [70]. Similarly, several laboratories have reported that TLR stimulation activates the XBP-1 branch of the UPR and that XBP-1 production was needed to promote a sustained production of inflammatory mediators, including IL-6 [71], [72]. Here, we identify GADD34 as a novel functional link between ISR and PAMPs detection in MEFs, required for the production of cytokines including type-I IFN. It will now be important to explore the therapeutic potential of targeting GADD34 to reduce cytokines overproduction during inflammatory conditions.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals the French Ministry of Agriculture and of the European Union. The protocol was approved by the Committee on the Ethics of Animal Experiments of the Institut Pasteur and Région PACA (Autorisation # 13.116 issued by DDSV/Préfecture des Bouches du Rhône, Marseille, France) and were performed in compliance with the NIH Animal Welfare Insurance #A5476-01 issued on 02/07/2007. All experiments were performed under isoflurane anesthesia (Forene, Abbott Laboratories Ltd, United-Kingdom), and all efforts were made to minimize suffering. Animals were housed in the Institut Pasteur and CIML animal facilities accredited by the French Ministry of Agriculture to perform experiments on live mice.
Matched wild-type (129 SvEv), and PKR−/− MEFs (Yang et al., 1995) were a gift from Caetano Reis e Sousa (Cancer Research UK, London); primary eIF2α S/S and eIF2α A/A MEFs were a gift from Randal J. Kaufman (Department of Biological Chemistry, University of Michigan Medical Center, USA); Matched wild-type (129 SvEv), ATF4−/−, GADD34ΔC/ΔC and CReP−/− MEFs were a gift from David Ron (Skirball Institute of Biomolecular Medicine, New York). All MEFs were cultured in DMEM, 10% FCS (HyClone, Perbio), 100 units/ml penicillin, 100 µg/ml streptomycin, 2 mM glutamine, 1× MEM non-essential amino acids and 50 µM 2-mercaptoethanol. NIH3T3 cells were cultured in RPMI 1640 (Gibco) supplemented with 10% FCS (HyClone, PERBIO), 100 units/ml penicillin and 100 µg/ml streptomycin. All cells were cultured at 37°C and 5% CO2. MEFs and NIH3T3 were treated for the indicated time with 10 µg/ml poly I:C (InvivoGen) in combination with lipofectamine 2000 (Invitrogen). Thapsigargin, tunicamycin, sodium arsenite, and guanabenz (all from SIGMA) were used at 200 nM, 2 µg/ml, 0.5 mM, and 10 µM respectively. The plasmid GADD34 (FLAG epitope tagged at N-terminus, CMV2-based mammalian expression) was a kind gift from David Ron (Institute of Metabolic Sciences, University of Cambridge, UK).
Puromycin labelling for measuring the intensity of translation was performed as previously described [47]. For immunoblots, 10 µg/ml puromycin (Sigma, min 98% TLC, cell culture tested, P8833, diluted in PBS) was added in the culture medium and the cells were incubated for 10 min at 37°C and 5% CO2. Where indicated, 25 µM cycloheximide (Sigma) was added 5 min before puromycin. Cells were then harvested, centrifuged at 4°C and washed with cold PBS prior to cell lysis and immunoblotting with the 12D10 antibody.
Cells were lysed in 1% Triton X-100, 50 mM Hepes, 10 mM NaCl, 2.5 mM MgCl2, 2 mM EDTA, 10% glycerol, supplemented with Complete Mini Protease Inhibitor Cocktail Tablets (Roche). Protein quantification was performed using the BCA Protein Assay (Pierce). 25–50 µg of Triton X-100-soluble material was loaded on 2%–12% gradient or 8% SDS-PAGE before immunoblotting and chemiluminescence detection (SuperSignal West Pico Chemiluminescent Substrate, Pierce). Nuclear extraction was performed using the Nuclear Complex Co-IP kit (Active Motif). Rabbit polyclonal antibodies recognizing ATF4 (CREB-2, C-20), GADD34 (C-19), Lamin A (H-102) and eIF2-α (FL-315) were from Santa Cruz Biotechnology, as well as mouse monoclonal anti-PKR (B-10). GADD34/PPP1R15A (Catalog No. 10449-1-AP) rabbit polyclonal antibody was purchased from PROTEINTECH.
Rabbit polyclonal anti-eIF2α[pS52] and Cystatin C were from Invitrogen and Upstate Biotechnology, respectively. Mouse monoclonal antibodies for β-actin and HDAC1 (10E2) were purchased from Sigma and Cell Signaling Technologies. Secondary antibodies were from Jackson ImmunoResearch Laboratories.
MEFs and NIH3T3 were grown on coverslips overnight and stimulated for the indicated time with poly I:C complexed with Lipofectamine 2000. Cells were fixed with 3% paraformaldehyde in PBS for 10 min at room temperature, permeabilized with 0,5% saponin in 5% FCS PBS with 100 mM glycine, for 15 min at room temperature and stained for 1 h with indicated primary antibodies. Anti-P-eIF2α was from BioSource; anti-dsRNA (clone K1) from English & Scientific Consulting Bt.; anti-IFN-β-FITC-conjugated from PBL Interferon Source; anti-puromycin (clone 2G11, mouse IgG1) has been previously described [47]. Alexa-conjugated secondary antibodies (30 min staining) were from Molecular Probes (Invitrogen). Coverslips were mounted on a slide and images taken with a laser-scanning confocal microscope (LSM 510; Carl Zeiss MicroImaging) using a 63× objective and accompanying imaging software. When PKR WT and PKR−/− were infected with CHIKV, protocol was performed as follows: cells were fixed with 4% paraformaldehyde in PBS for 20 min, then permeabilized for 30 min in 0.1% Triton 100X (Sigma) and blocked in 10% of normal goat serum (Vector Laboratories). Cells were stained with a mouse monoclonal antibody directed against CHIKV capsid coupled to Alexa-488 and a mouse antibody against puromycin coupled to Alexa-555 and a rabbit antibody anti-eIF2α[pS52] (Invitrogen) and a Cyanin-3 secondary antibody, and finally counterstained with Hoechst (Vector Lab). Cells were observed with an AxioObserver microscope (Zeiss). Pictures and Z-stacks were obtained using the AxioVision 4.5 software.
IFN-β and IL-6 quantification in culture supernatant was performed using the Mouse Interferon Beta ELISA kit (PBL InterferonSource) and Mouse Interleukin-6 ELISA kit (eBioscience) respectively, according to manufacturer instructions.
Total RNA was isolated from cells using the RNeasy miniprep kit (QIAGEN) combined with a DNA digestion step (RNase-free DNase set, QIAGEN). cDNA was synthesized using the Superscript II reverse transcriptase (Invitrogen) and random hexamer primers. Quantitative PCR amplification was carried out using complete SYBR Green PCR master mix (Applied Biosystems) and 200 nM of each specific primer. 5 µl of cDNA template was added to 20 µl of PCR mix, and the amplification was tracked via SYBR Green incorporation by an Applied Biosystems thermal cycler. cDNA concentration in each sample were normalized by using HPRT. A nontemplate control was also routinely performed. The primers used for gene amplification (designed with Primer3 software) were the following: GADD34 (S 5′-GACCCCTCCAACTCTCCTTC-3′, AS 5′-CTTCCTCAGCCTCAGCATTC-3′); HPRT (S 5′-AGGCCAGACTTTGTTGGATTT -3′, AS 5′-GGCTTTGTATTTGGCTTTTCC -3′); IFN-β (S 5′-CCCTATGGAGATGACGGAGA-3′, AS 5′-ACCCAGTGCTGGAGAAATTG-3′); IL-6 (S 5′-CATGTTCTCTGGGAAATCGTG-3′, AS 5′-TCCAGTTTGGTAGCATCCATC-3′); PKR (S 5′-CCGGTGCCTCTTTATTCAAA -3′, AS 5′-ACTCCGGTCACGATTTGTTC-3′); Cystatin C (S 5′-GAGTACAACAAGGGCAGCAAC-3′, AS 5′-TCAAATTTGTCTGGGACTTGG-3′). ATF4 (5′-GGACAGATTGGATGTTGGAGA-3′, AS 5′-AGAGGGGCAAAAAGATCACAT3-′).
mRNA isolation from total RNA was performed with oligodT columns (Genelute mRNA miniprep kit (Sigma). Data were analyzed using the 7500 Fast System Appled Biosystems software.
RNA integrity upon poly I:C stimulation was measured by capillary electrophoresis using the the Agilent RNA 6000 Pico Chip kit (Agilent Technologies) in an Agilent 2100 Bioanalyser, according to manufacturer instructions.
GADD34ΔC/ΔC and the corresponding WT control MEFs were infected at a multiplicity of infection (MOI) of 10 or 50 with CHIKV-GFP generated using a full-length infectious cDNA clone provided by S. Higgs [71]. By 24 h and 48 h post infection, 30 000 cells were analyzed in triplicate by FACS for expression of GFP. At the same time-points, culture supernatants were collected and IFN-β protein assessed by ELISA. In experiments with exogenous IFN-β, cells were treated with mouse IFN-β (PBL InterferonSource) for 3 h before infection with CHIKV-GFP. When guanabenz was used to specifically inhibit GADD34, MEFs cells were treated for 2 h with 10 µM of Guanabenz or DMSO and then infected in the same medium. Three hours post infection the inoculum was removed and fresh medium with Guanabenz or DMSO was added and maintained all along the experiment. RNAi for GADD34 was performed as described in [60].
FVB WT mice were obtained from Charles River Laboratories (France). GADD34ΔC/ΔC FVB mice were obtained from L. Wrabetz (Milan). Mice were anesthetized and inoculated via the intradermal route with 106 PFU of CHIKV-21 isolate [72]. Viral titers in tissues and serum were determined as described before [59], and expressed as tissue cytopathic infectious dose 50 (TCID50)/g or TCID50/ml, respectively. Organs including heart, liver, skeletal muscles and spleen were collected for histopathological procedures. organs were then fixed in 4% paraformaldehyde solution, paraffin-embedded, sectioned coronally in 5–10 µm thickness and stained with hematoxylin-eosin.
|
10.1371/journal.pgen.1000034 | Genetic Networks of Liver Metabolism Revealed by Integration of Metabolic and Transcriptional Profiling | Although numerous quantitative trait loci (QTL) influencing disease-related phenotypes have been detected through gene mapping and positional cloning, identification of the individual gene(s) and molecular pathways leading to those phenotypes is often elusive. One way to improve understanding of genetic architecture is to classify phenotypes in greater depth by including transcriptional and metabolic profiling. In the current study, we have generated and analyzed mRNA expression and metabolic profiles in liver samples obtained in an F2 intercross between the diabetes-resistant C57BL/6 leptinob/ob and the diabetes-susceptible BTBR leptinob/ob mouse strains. This cross, which segregates for genotype and physiological traits, was previously used to identify several diabetes-related QTL. Our current investigation includes microarray analysis of over 40,000 probe sets, plus quantitative mass spectrometry-based measurements of sixty-seven intermediary metabolites in three different classes (amino acids, organic acids, and acyl-carnitines). We show that liver metabolites map to distinct genetic regions, thereby indicating that tissue metabolites are heritable. We also demonstrate that genomic analysis can be integrated with liver mRNA expression and metabolite profiling data to construct causal networks for control of specific metabolic processes in liver. As a proof of principle of the practical significance of this integrative approach, we illustrate the construction of a specific causal network that links gene expression and metabolic changes in the context of glutamate metabolism, and demonstrate its validity by showing that genes in the network respond to changes in glutamine and glutamate availability. Thus, the methods described here have the potential to reveal regulatory networks that contribute to chronic, complex, and highly prevalent diseases and conditions such as obesity and diabetes.
| Although numerous quantitative trait loci (QTL) influencing disease-related phenotypes have been detected through gene mapping and positional cloning, identifying individual genes and their potential roles in molecular pathways leading to disease remains a challenge. In this study, we include transcriptional and metabolic profiling in genomic analyses to address this limitation. We investigated an F2 intercross between the diabetes-resistant C57BL/6 leptinob/ob and the diabetes-susceptible BTBR leptinob/ob mouse strains that segregates for genotype and diabetes-related physiological traits; blood glucose, plasma insulin and body weight. Our study shows that liver metabolites (comprised of amino acids, organic acids, and acyl-carnitines) map to distinct genetic regions, thereby indicating that tissue metabolites are heritable. We also demonstrate that genomic analysis can be integrated with liver mRNA expression and metabolite profiling data to construct causal, testable networks for control of specific metabolic processes in liver. We apply an in vitro study to confirm the validity of this integrative method, and thus provide a novel approach to reveal regulatory networks that contribute to chronic, complex, and highly prevalent diseases and conditions such as obesity and diabetes.
| Genetic linkage and association studies have the power to establish a causal link between gene loci and physiological traits. These studies can make novel connections between biological processes that would not otherwise be predictable based on current knowledge. The pace of gene discovery has greatly accelerated in recent years, and numerous quantitative trait loci (QTL) influencing disease-related phenotypes have been identified through gene mapping and positional cloning. While it has become relatively straightforward to map a phenotype to a broad genomic region, identification of the individual gene(s) responsible for the phenotype remains difficult. Consequently, only a few percent of the many QTL that have been mapped have had their underlying gene(s) identified [1]–[7]. Another limitation of traditional QTL mapping is that it is based on association with a physiological phenotype, but often does not reveal the molecular pathways leading to that phenotype.
One way to uncover molecular mechanisms of disease states is to broadly expand the types of phenotypes analyzed in genetic screens. For example, with microarray technology, one can measure the abundance of virtually all mRNAs in a segregating sample. Importantly, mRNA abundance shows sufficient heritability in outbred populations and experimental crosses to allow mapping of gene loci that control gene expression, termed expression QTL (eQTL) [8],[9]. When eQTL co-localize with a physiological QTL, one can hypothesize a shared regulator and offer a potential pathway leading to the physiological trait [9],[10].
The pathway between a QTL and a physiological trait often involves changes in the steady-state levels of metabolic intermediates, in addition to changes in mRNA abundance. These metabolites can correlate with the genetic, transcriptional, translational, post-translational, and environmental influences on phenotype [7],[11]. Moreover, metabolites are intermediates in signaling pathways that can regulate gene expression. For example, fatty acids act as ligands for several of the PPAR nuclear hormone receptors, bile acids activate FXR in liver, and diacylglycerol regulates protein kinase C [12]–[14]. Metabolite abundance reflects a biological response to exogenous and endogenous inputs, and when investigating pathways from genotype to phenotype, metabolites can provide a powerful complement to gene expression data and give novel insights into disease pathogenesis mechanisms [7], [11], [15]–[25].
Our laboratories have begun to apply targeted metabolic profiling to study mechanisms underlying obesity-induced diabetes [15]–[20], but have not yet attempted to integrate these methods with genotyping and transcriptional profiling. This has included the application of gas chromatography/mass spectrometry (GC/MS) and tandem mass spectrometry (MS/MS) for measurements of acyl-carnitine, organic acid, amino acid, free fatty acid, and long and medium-chain acyl-CoA metabolites in tissue extracts and bodily fluids. Herein, we have applied these methods to measure various metabolites in liver samples from mouse strains that differ in susceptibility to obesity-induced diabetes.
C57BL/6 (B6) leptinob/ob mice are obese but essentially resistant to diabetes, whereas BTBR leptinob/ob mice are severely diabetic [22]. In an F2 cohort derived from these parental strains, we have shown that the range of blood glucose, insulin levels, and body weight exceeds that of either the C57BL/6 (B6) leptinob/ob or BTBR leptinob/ob parental strains. We went on to identify several diabetes-related QTL in this F2 sample [21],[22]. In the current study, we focused on a subset of 60 F2 mice that have previously been evaluated in detail with regard to liver gene expression profiles [24] to ask if the abundances of hepatic metabolic intermediates would show sufficient heritability to enable us to map metabolic QTL (mQTL). Because we previously performed mRNA expression profiling on liver samples from this F2 sample, we were also able to investigate the potential for integrative analysis of the expression profiling and metabolite data sets.
We show that liver metabolites do map to distinct genetic regions, thereby demonstrating that tissue metabolite profiles are heritable. In addition, we show that mQTL co-localize with eQTL, suggesting common genetic regulators. Finally, as a proof of principle of the practical significance of this multi-disciplinary approach, we illustrate the construction of a specific causal network that links gene expression and metabolic changes, and demonstrate its validity by targeted gene expression analysis.
We determined the concentration of 67 liver metabolites, comprised of 15 amino acids and urea cycle intermediates, 45 acyl-carnitines, and 7 organic acids (TCA cycle intermediates and related metabolites) in the F2 sample. The specific analytes are summarized in Table S1.
We created a correlation matrix of all pairwise comparisons among individual metabolites. Unsupervised hierarchical clustering revealed several “hot spots” of highly correlated metabolites (Figure 1). It is striking that several hot spots correspond to the biochemical pathway to which the metabolites belong. For example, 12 of the 15 amino acids cluster in this matrix. Moreover, when we consider pairwise correlations between all amino acids, 75% had absolute correlation coefficients greater than 0.5 (p<0.01) (Table S2). Permutation analysis of these pairwise correlations confirm that the 15 amino acids correlate as a functional group (p<0.001). Several specific acyl-carnitine derivatives are also clustered, such as hexadecadienoyl carnitine (C16∶2), 3-hydroxy-tetradecanoyl carnitine or dodecenedioyl carnitine (C14∶1-OH/C12∶1-DC), and 3-hydroxy-palmitoleoyl carnitine or cis-5-tetradecenedioyl carnitine (C16∶1-OH/C14∶1-DC). The fact that metabolites of a common functional group are highly correlated suggests that there are potential regulators of these biochemical pathways segregating in this F2 sample.
In another cluster, pyruvate correlates most highly with alanine (r = 0.53, p<0.01), and also with lactate and tiglyl carnitine (C5∶1) (p<0.01). Alanine and short-chain acyl-carnitines are products of peripheral protein and fatty acid catabolism, respectively, and are delivered to the liver. The liver uses alanine, along with pyruvate and lactate, as gluconeogenic substrates and rapidly interconverts these metabolites through transamination and oxidation/reduction. The clustering of these metabolites based on their relative concentration in F2 animals suggests that static metabolic profiling can be used as a marker for changes in flux through certain metabolic pathways. All metabolite-metabolite correlation coefficients are listed in Table S2.
It has been demonstrated that mRNA abundance, as determined with microarray technology, is sufficiently heritable to map QTL [7], [8], [10], [23]–[27]. Lan et. al. showed that using expression mapping, specifically in this F2 intercross, can uncover mechanisms that explain correlations between specific transcripts [8]. We therefore sought to determine if metabolite abundance, as measured in F2 liver samples by mass spectrometry, was similarly heritable. If so, resulting metabolic QTL (mQTL) could be integrated with expression QTL (eQTL) to form network models of gene expression that might ultimately help to explain diabetes susceptibility and resistance in the BTBR leptinob/ob and B6 leptinob/ob strains, respectively [28],[29].
We found that individual metabolites mapped to specific regions of the genome. By permutation analysis, 21% of the metabolites map significantly to genomic regions (LOD>5.0, p<0.05), indicating those genomic regions could potentially influence (either directly or indirectly) the abundance of these metabolites. We used LOD threshold of 3.0 to investigate both major and minor putative mQTL where groups of metabolites map. Figure 2 displays a heat map, with metabolites organized by hierarchical clustering as in Figure 1. The twelve amino acids that clustered based on correlation (citrulline, tyrosine, and alanine are the exceptions) map to common mQTL, e.g., an overlapping region of chromosome 9. Amino acids that act together in specific pathways show additional common mQTL. For example glx (glutamine+glutamate) and urea cycle intermediates arginine, asx (asparagine+aspartate), and ornithine, map to a common region of chromosome 7. The gluconeogenic substrates alanine and pyruvate have a mapping profile distinct from the majority of amino acids in that they lack the prominent mQTL on chromosome 9 (Figure 2). This unique alanine/pyruvate mQTL may explain why alanine clusters with pyruvate rather than the amino acids in the correlation matrix (Figure 1).
The foregoing results demonstrate that metabolites of a functional class often are correlated with one another and have common mQTL. To better understand how gene expression and metabolites are related, we adopted the approach used by Carrari [30] and created a correlation matrix between liver metabolites and selected liver transcripts of our 60 F2 mice. Three categories of transcripts were chosen, based on gene ontology terms relating to the biological process in which they play a role: 1) carbohydrate metabolism (glucose metabolism, gluconeogenesis, glycolysis, carbohydrate biosynthesis, TCA cycle, glucose transport, and glycogen metabolism); 2) lipid metabolism (fatty acid biosynthesis, fatty acid oxidation, steroid metabolism, cholesterol metabolism and biosynthesis, and lipid biosynthesis); and 3) protein metabolism (urea cycle, amino acid biosynthesis, protein catabolism, and amino acid transport). We organized the metabolites into functional classes to reveal whether biochemical groups of metabolites correlated in a specific pattern with transcripts of a particular pathway (Figure 3).
We found evidence for correlations among functionally similar metabolites and transcripts when organized by biological process. For example, several long-chain acyl-carnitine species show a positive correlation with groups of transcripts involved in glycolysis, fatty acid biosynthesis, steroid metabolism, cholesterol metabolism, and lipid biosynthesis. In contrast, a subset of medium-chain acyl-carnitines and short chain acyl-carnitines exhibit a negative correlation to these same individual transcripts. These findings are consistent with recent studies from our laboratories showing that long-chain acyl-carnitines accumulate in muscle of animals with diet-induced obesity at the expense of short-chain acyl-carnitines, and that this abnormality is resolved when obese animals are exercised [17].
The 15 amino acids displayed a common correlation pattern with mRNA transcripts in pathways of protein metabolism, as well as glycolysis, the TCA cycle, and several lipid metabolism transcripts. These amino acids are very tightly correlated with one another, leading us to investigate the role played by individual transcripts in control of amino acid abundance. Our data show that two very highly correlated metabolites often correlate with the same set of individual transcripts. However, we also see that within this metabolite group, subsets of amino acids will have a unique transcript correlation pattern (Table S3, Table S4). For example, thirteen of fifteen amino acids correlate (r>0.35, p<0.01) with Slc38a3, a sodium-dependent transporter that mediates entry of a select group of amino acids across the plasma membrane. There are pathways by which the few known Slc38a3 amino acid substrates (alanine, asparagine, histidine, and glutamine) could serve as precursors for biosynthesis of non-substrate amino acids that also correlate with this transporter [31],[32]. In contrast, only valine and leucine+isoleucine correlate as highly (r>0.35, p<0.01) with Ppargc1a mRNA, and could represent a unique metabolic pathway involving the branched-chain amino acids.
One hypothesis that follows from our results is that unique genetic regulators could affect the abundance of clusters of metabolites. Unlike mRNA transcripts, metabolites can be interconverted with other metabolites, generating a cluster to which the precursor metabolite will be highly correlated [33]. The downstream product metabolites will also be correlated with the regulatory transcript and co-map with the eQTL of the regulatory transcript [7],[34].
Glutamate is a substrate and product in amino acid catabolic and biosynthetic pathways. Glutamate can act either as an ammonium donor or acceptor in transamination reactions (via α-ketoglutarate) and the glutamate dehydrogenase reaction, and can also be rapidly synthesized from glutamine via glutaminase, thus providing precursor metabolites for the generation of other organic acids and amino acids. Glutamine can also act as a signaling molecule to alter expression of urea cycle and gluconeogenic enzymes [35]–[39]. Given that glutamine and glutamate (glx) can generate a network of related metabolites and can also change gene expression, we focused on glx as the start-point for building a proof-of-principle causal network from the F2 liver expression and metabolite profiling data sets. We generated a network featuring glx and a limited number of transcripts that passed multiple, stringent selection filters (see materials and methods). This provided a testable network that would enable us to gain insights into metabolite-transcript relationships.
Transcript nodes of the network are highly correlated to glx (p<0.05 by 10,000 permutations) as well as other amino acids (Table 1, Table S4). Table 1 depicts the overlap of the glx mQTL interval and the physical location of the transcripts or their eQTL encompassing a 1.5 LOD support interval around LOD peaks that are at least 3.0 [40],[41]. We note that glx is correlated with mRNA of two transporters: sodium-dependent amino acid transporter Slc38a3 and glutamate transporter Slc1a2, whose genes are located on chromosomes 9 (102.5 Mb) and 2 (107.5 Mb), respectively. Additionally, the glx mQTL on chromosome 9 spans a region containing Slc38a3 and the mQTL on chromosome 2 and 9 overlaps with the eQTL of Slc1a2 (Table 1). We hypothesize that both Slc1a2 and Slc38a3 could mediate the entry of glx into liver cells, but that Slc1a2 may also have expression regulated by glx abundance.
Table 1 also shows that glx is significantly correlated to argininosuccinate synthetase 1 (Ass1), arginase 1 (Arg1), phosphoenolpyruvate carboxykinase 1 (Pck1), isovaleryl coenzyme A dehydrogenase (Ivd) and alanine∶glyoxylate aminotransferase (Agxt) mRNAs. The physical location and/or mapping location of these transcripts with respect to the glx mQTL indicates that the metabolite-transcript relationship may go beyond correlation. For example, on chromosome 2, we see that the glx mQTL co-maps with the eQTL for Agxt, Arg1, Ass1, and Ivd [41]. This is consistent with network models in which the QTL regulates glx, which then regulates gene expression or conversely, the QTL regulates mRNA abunance of the four transcripts, which then regulate glx [9].
Using the method described by Chaibub et al. (in review), we generated a causal network consisting of glx and these highly correlated transcripts (FDR = 0.014), incorporating mQTL and eQTL to determine directionality between the nodes (Figure 4). This network model predicts that modulation of glutamine and/or glutamate levels should lead to a change in the expression of Agxt, Arg1, and Pck1. To test this prediction, we isolated hepatocytes from lean B6 and BTBR parental strains and measured changes in gene expression as a result of addition of 10 mM glutamine to the cultured cells. Glutamine exposure changed transcript abundance, and no transcript-specific strain differences in glutamine effect on gene expression were found (p = 0.53) (Figure 5). Glutamine significantly increased expression of Agxt, Arg1, Pck1, and Ass1 in both strains (p<0.05 for both strains); the increases in Pck1 and Ass1 confirm prior studies [36]–[39].
Given its role as a glutamate transporter, it is not surprising that Slc1a2 is upstream of glx in the best proposed causal network (BF = 163) (Figure 4, solid lines). However, glutamine exposure in vitro reduced Slc1a2 expression in isolated hepatocytes from either mouse strain, supporting the second-best causal network solution (Figure 4, dotted lines). Glutamine also reduced Ivd expression in the B6 strain but showed no effect in the BTBR strain, despite Ivd being upstream of glx in our best causal network. Our causal network predicts Slc38a3 should be unchanged by glutamine treatment. Our hepatocyte experiments confirm this prediction (Figure 5). Argininosuccinate lyase (Asl), which is neither correlated nor co-maps with glx, served as a negative control and indeed was not altered by glutamine treatment.
Genomics, transcriptomics, proteomics, and metabolomics have delivered large arrays of data, allowing one to correlate physiological states with patterns of gene expression, protein levels, and metabolite abundance. A major challenge in the analysis and interpretation of this data is delivering models of causation from correlations [9],[42]. Mouse models of diabetes provide a unique method for exploring correlation structure since metabolic dysregulation creates a window for simultaneous application of multiple “omic” technologies.
We have previously shown that diabetes traits show strong heritability in an F2 intercross between the diabetes-resistant C57BL/6 leptinob/ob and the diabetes-susceptible BTBR leptinob/ob mouse strains. We assume that the disease phenotype is brought about by a complex pattern of gene expression changes in key tissues [21],[22]. However, we also recognize the complexity inherent in discriminating the gene expression changes that cause diabetes from those that occur as a consequence of the disease. For example, many genes are known to be responsive to elevated blood glucose levels [43]. Through correlation alone, it is difficult to distinguish these “reactive” genes from ones that are “causal” for the disease.
We have taken advantage of the high heritability of mRNA abundance phenotypes, and via microarray technology, have mapped gene loci controlling gene expression at the genome-wide level [8]. This establishes at least one node in a network simply because genetic variation leads to changes in gene expression and not vice versa. However, it does not establish whether the link between a locus and a phenotype is direct or via multiple steps and pathways [27],[44].
The purpose of the current study was to explore the possibility that the levels of metabolites in tissues are sufficiently heritable in an F2 intercross to provide significant linkage signals, leading to metabolic QTL. Given that many pathways converge upon common metabolites and that these pathways have multiple controllers, any one genetic locus may not alter metabolite levels significantly, and therefore may not be identified as a metabolite QTL. Nonetheless, in our F2 sample, we found significant linkage signals, including some that are quite strong (e.g. tyrosine: LOD>7, p<0.005; chromosome 2).
Our results reveal that metabolites can be mapped to distinct genetic regions, much like mRNA transcripts. Although QTL mapping in an F2 sample does not provide sufficient resolution to identify individual genes with high certainty, it can yield novel information about regulatory networks. Phenotypes mapping to the same locus can be hypothesized to be co-regulated by that locus. With our definition of “phenotype” now including transcripts, metabolites, and physiological traits, we can begin to devise relationships between these phenotypes and genetic regions.
This F2 study provides evidence of co-regulation of biologically related pathways. An example is the correlations we found between amino acids and short-chain acyl-carnitine derivatives. These findings are consistent with our understanding of metabolic physiology. In a catabolic, “glucose starved” state, muscle degrades proteins and delivers amino acids to the liver for glucose production. The liver transaminates amino acids to corresponding α-keto acid gluconeogenic substrates. Alpha-ketoglutarate is often the α-keto acid acceptor for these transaminase reactions, generating glutamate as a product. Glutamate, which can also be generated from glutamine in the glutaminase reaction, is then deaminated to produce ammonia by glutamate dehydrogenase, to be fixed through the urea cycle. Additionally, hepatic fatty acid oxidation and amino acid catabolism yield even and odd-numbered short-chain acyl CoAs, which can be used for fuel and for production of ketone bodies. These short-chain acyl-CoA species are readily converted to the cognate carnitine esters, which we have profiled by MS/MS in this study.
The amino acid metabolites provide the most striking evidence of functional clustering. We see in both the correlation matrix (Figure 1) and the genetic linkage data (Figure 2) that the majority of amino acids group together. However, a subset of the amino acids, asx, glx, arginine, and ornithine uniquely map to chromosome 7. Our data predict that these metabolites are driven by different genetic regulators, leading to a unique mapping signature, even within a group of highly correlated metabolites. The C/EBP transcription factors have been shown to alter expression of enzymes acting in the urea cycle and gluconeogenic pathway [45]–[51], and the C/EBPα isoform is encoded on chromosome 7. Although we cannot determine that metabolites are mapping to the same individual genes, we can identify genetic regions that coordinate groups of metabolites and transcripts and contain plausible candidate genes.
The relationship between mRNA transcripts and metabolites, however, can be bi-directional. Our network identifies a specific metabolite, glx that regulates gene expression. This is consistent with previous studies where glutamine alone increases hepatic expression of argininosuccinate synthetase and phosphoenolpyruvate carboxykinase, but when combined with other essential amino acids, alters additional transcripts of urea cycle and gluconeogenic pathways [36]–[38],[52]. Our work extends these prior observations by showing that glutamine also changes expression of Agxt, Arg1, Ivd, and Slc1a2, but does not alter Slc38a3, despite the positive correlation with this transcript. The combination of pathway construction based on transcriptional and metabolic profiling and direct model testing in living cells provides evidence for a new pathway by which glx can regulate a key gluconeogenic enzyme. Future studies will be needed to investigate if this pathway is perturbed in development of diabetes.
The glutamine induced reduction in Slc1a2 expression was unexpected given that this glutamate transporter is upstream of glx in the best-proposed causal network (Figure 4, solid lines). Slc1a2 mRNA abundance, however, maps in trans (to a locus distinct from the physical location of the gene) to chromosome 9, its eQTL overlapping with the glx mQTL. It is therefore possible that glutamine could regulate Slc1a2, as indicated by the second causal network (Figure 4, dotted lines). Several studies have shown that Slc1a2 expression in astrocytes is reduced by increased ammonia [45]–[47], [51], [53]–[55]. Despite the positive correlation between Slc1a2 and glx in vivo, the glutamine-treated hepatocytes produce ammonia via glutaminase, and could decrease expression of hepatic Slc1a2 in vitro. We also did not predict altered expression of Ivd, an enzyme of leucine oxidation. It is interesting to note that Ivd is a case where a gene maps both in cis (to the locus containing the Ivd gene) and in trans, here overlapping with the glx mQTL on chromosomes 2 and 13. Studies have shown that glutamine has an inverse relationship with leucine oxidation, and this could be mediated by glutamine-induced decreased Ivd expression [48],[50].
We show that the combined use of eQTL and mQTL, with correlations allows one to derive a network and establish data-driven hypotheses about metabolite and gene expression relationships. For example, glycine and serine are the two amino acids most highly correlated with glx, and the transcript most highly correlated with glx is Agxt (Table 1, Table S2). Indeed, in our experiments, Agxt was upregulated by glutamine. We hypothesize that the upregulation by glx of Agxt is one mechanism by which glx is correlated with glycine and serine since Agxt catalyzes the transamination of glyoxalate to form glycine, which can then be converted to serine. In further support of this hypothesis, in the F2 sample, serine and glycine correlate (r>0.5, p<0.01) to Agxt.
The concurrent use of transcriptomics and metabolomics is not limited to one biochemical pathway. For example, the correlation between amino acids and transcripts of carbohydrate and lipid metabolism might reflect a broader signaling function of amino acids beyond pathways of protein metabolism. Furthermore, this correlation, co-mapping, and causal network analysis can uncover roles for transcripts of unknown function. We note Riken clones and ESTs are among the transcripts highly correlated to individual metabolites (Table S3). By incorporating these transcripts of unknown function as nodes into causal networks, along with transcripts from known pathways, we may infer the functions of these previously unidentified mRNA species.
In conclusion, this study shows that metabolites, in addition to transcripts and physiological traits, can be mapped to genetic regions, providing a powerful tool to establish connections between genetic loci and physiological traits. The groups of metabolites and transcripts that are correlated or co-map to physiological traits in our F2 sample may offer insight into metabolic pathways that are causal or reactive to diabetes pathology.
BTBR, B6, and B6-ob/+ mice were purchased from The Jackson Laboratory (Bar Harbor, ME) and bred at the University of Wisconsin. The lineage and characteristics of the BTBR strain have been reviewed by Ranheim et al. Mice were housed in an environmentally controlled facility (12-hour light and dark cycles) and were weaned at 3 weeks of age onto a 6% fat diet (Purina; #5008). Mice had ad libitum access to food and water, except for 4 hour fasting periods before blood draws and killing (by CO2 asphyxiation). Plasma glucose levels were measured using a commercially available kit (994-90902; Wako Chemicals). Plasma insulin levels were measured by radioimmunoassay (RI-13K; Linco Research).
The facilities and research protocols were approved by the University of Wisconsin Institutional Animal Care and Use Committee.
Sixty F2 leptinob/ob mice ranging in age from 13 to 26 weeks were genotyped as previously described [22]. Mapmaker/EXP was used to compile genotype data into framework map.
Liver RNA was arrayed as described in Lan et. al [8]. Ten to 12 week old male and female F2 leptinob/ob mice were killed by CO2 asphyxiation after a 4-h fast. Total RNA from sixty F2 mice using RNAzol reagent (Tel-Test) and was further purified using an RNeasy kit (Qiagen). The sample labeling, microarray hybridization, washing, and scanning were performed according to the manufacturer's protocols (Affymetrix). Labeled cRNA was prepared and hybridization assay procedures including preparation of solutions were carried out as described in the Affymetrix GeneChip Expression Analysis Technical Manual. A total of 60 MOE430A and MOE430B arrays were used to monitor the expression levels of approximately 45,000 genes or ESTs. The distribution of fluorescent material on the array was obtained using G2500A GeneArray Scanner (Affymetrix). Microarray Suite (MAS) version 5.0 and GeneChip Operating Software (GCOS) supplied by Affymetrix was used to perform gene expression analysis. Expression levels of all the transcripts were estimated using the RMA algorithm [49].
Amino acids, acyl-carnitines and organic acids were measured using stable isotope dilution techniques [15],[18],[56]. Amino acids and acyl-carnitine species were measured using flow injection tandem mass spectrometry and sample preparation methods described previously [15],[56]. Briefly, samples were equilibrated with a cocktail of internal standards, de-proteinated by precipitation with methanol, aliquoted supernatants were dried, and then esterified with hot, acidic methanol (acyl-carnitines) or n-butanol (amino acids). The data were acquired using a Micromass Quattro micro TM system equipped with a model 2777 autosampler, a model 1525 µ HPLC solvent delivery system and a data system controlled by MassLynx 4.0 operating system (Waters, Milford, MA) [15],[56]. Organic acids were quantified using a previously described method that utilizes Trace GC Ultra coupled to a Trace DSQ MS operating under Excalibur 1.4 (Thermo Fisher Scientific, Austin, TX) [18].
Sixty-seven liver metabolites were measured, comprised of 15 amino acids and urea cycle intermediates, 45 acyl-carnitine derivatives, and 7 organic acids (TCA cycle intermediates and related analytes). The specific metabolites are listed in Table S1. All MS analyses employed stable-isotope-dilution. The standards serve both to help identify each of the analyte peaks and provide the reference for quantifying their levels. Quantification was facilitated by addition of mixtures of known quantities of stable-isotope internal standards from Isotec (St. Louis, MO), Cambridge Isotope Laboratories (Andover, MA), and CDN Isotopes (Pointe-Claire, Quebec, CN) to samples, as follows: Acyl-carnitine assays–D3-acetyl, D3-propionyl, D3-butyryl, D9-isovaleryl, D3-octanoyl, and D3-palmitoyl carnitines; Amino acid assays–15N1,13C1-glycine, D4-alanine, D8-valine, D7-proline, D3-serine, D3-leucine, D3-methionine, D5-phenylalanine, D4-tyrosine, D3-aspartate, D3-glutamate, D2-ornithine, D2-citrulline, and D5-arginine; Organic acid assays–D3-lactate, D3-pyruvate, 13C4-succinate, D2-fumarate, D4-glutarate, 13C1-malate, D6-alpha-ketoglutarate, and D3-citrate. In addition to mass, analytes are identified on the basis of the particular MS/MS transitions that we monitor for each class of metabolites. For example, all acyl-carnitine methyl esters produce a fragment m/z 99. We make the assumption that all even mass precursors ions of m/z 99 are acyl-carnitines to which we assign plausible molecular structures. We differentiate isobaric structures e.g., dicarboxylic and hydroxylated acyl-carnitines, by comparing of MS/MS spectra for precursors of m/z 85 butylated acyl-carnitine species. We can infer whether the original compound had one or two carboxyl groups on the basis of the mass change from methyl to butyl esters.
Given our sample size, we initially analyzed metabolite abundance by hierarchical clustering using the distance function 1-correlation [40], [57]–[60]. Pairwise Spearman correlation coefficients of r>0.254 and r>0.330 reflected p-values p<0.05 and p<0.01, respectively. To test whether the 15 amino acids are significantly correlated as a group, groups of 15 metabolites were permuted 1,000 times and the percentage of pairwise correlations exceeding 0.5 was recorded for each group. The fifteen amino acids cluster significantly as a group based on 1,000 permutations (p<0.001).
Detection and mapping of QTL was performed as previously described [8],[22]. Briefly, genotypes of 512 F2 mice at 293 markers were assembled using MAPMAKER/EXP [61]. A previously established subset of 60 mice with transcript data was used for expression QTL analysis [24]. Interval mapping methods adjusted for sex as implemented in R/qtl [62] were used to compute linkage to the traits of interest and to investigate mode of inheritance. The traits included the 45,265 probe sets surveyed by microarray analysis, and the 67 liver metabolites assayed by MS methods.
We used standard interval mapping implemented in R/qtl to map each of the transcripts and liver metabolites at 1-cM resolution with age as additive covariates and sex as both additive and interactive covariates [62]. A LOD threshold of 5.0 is required to reach a level of p<0.05 in this data set with sample size 60 based on 10,000 permutations. We used threshold of 3.0 in order to highlight genetic regions to which groups of metabolites map. To visualize regions of mQTL co-localization in highly correlated metabolites (Figure 2), we constructed heat maps where metabolites are ordered as in hierarchical clustering using 1-correlation, as in Figure 1. When mice with the B6 allele at a marker have greater levels of metabolites on average than mice with the BTBR allele at that marker, the LOD score at that marker is multiplied by −1. This adjustment allows us to visualize whether the B6 or BTBR allele results in elevated metabolite abundance.
Hepatocytes from 10-week lean male and female BTBR and B6 parental strain mice (n = 5 for each genotype) were isolated by liver perfusion [63]. Hepatocytes were seeded at subconfluency (3.5 × 106 cells/6 well plate) in low glucose DMEM (GIBCO) supplemented with FBS (10% vol/vol; GIBCO), pen/strep antibiotic (1%, GIBCO), glutamine (2 mM; GIBCO), and pyruvate (1 mM; GIBCO). Cells were left to attach for 3 hours in an incubator at 37°C, 5% CO2. After a wash with PBS, the cells were treated with unsupplemented DMEM (Sigma) with 1 g/L glucose, pen/strep (1%), and +/− 10 mM glutamine. Cells were treated for 24 hours.
RNA was extracted from hepatocytes using RNeasy kits (Qiagen) after treatment described above. Hepatocytes in 6-well plates were homogenized in 0.35 ml of RLT buffer and stored at −80 C. RNA was purified using RNeasy-mini columns (Qiagen) according to the manufacturer's directions. The ratio of the optical densities from RNA samples measured at 260 and 280 nm was used to evaluate nucleic acid purity and total RNA concentrations were determined by the absorbance at 260 nm. The quality of total RNA was estimated based on the integrity of 28S and 18S rRNA separated using 1% agarose gel electrophoresis.
Gene expression was measured using a 7500 fast real-time PCR system (Applied Biosystems). cDNA was synthesized from 1 ug of total RNA using the SuperScriptIII first-strand cDNA synthesis kit (Invitrogen) primed with a mixture of oligo-dT and random hexamers. Primers were obtained from Integrated DNA Technologies and MWG Biotechnology. The SYBR Green PCR core reagent kit (Applied Biosystems) was used to determine relative expression. The housekeeping gene Actb was used as a normalization control. Primer sequences and gene accession codes for transcripts of the glx network are provided in Table S5.
Causal networks were constructed using the methods of Chaibub, et al. (in review). Although the network has the ability to accommodate 100 or more transcripts, we chose a limited number of transcripts passing several selection filters. The transcripts for the glx network were derived from the top 250 most correlated transcripts (p<0.002) according to the WebQTL software (www.genenetwork.org). A hypergeometric test was performed and identified the GO term category “metabolism” as one of the two processes significantly enriched by these correlates (p<0.004). Transcripts were chosen from this category, with an additional requirement being that they have at least one eQTL overlapping with the glx mQTL (Table 1). QTL in the genetic region encompassing a 1.5 LOD support interval around LOD peaks that are at least 3.0 are also included [41],[42]. Based on 10,000 permutations for each of the transcripts, the LOD threshold is significantly higher to reach significance (LOD>5.0 is required for p<0.05), but the 3.0 threshold was used include major and minor putative QTL [8],[24].
If more than one probe set was used to identify a transcript of interest, only probe sets with a grade A annotation on Affymetrix were considered. For these probe sets, only those with all eleven oligonucleotides aligning (via BLAST) to their appropriate target sequence provided by the National Center for Biotechnology Information (www.ncbi.nlm.nih.gov) were considered acceptable. If more than one primer set still identified the transcript, an average of the probe sets in the network.
We built an undirected dependency graph (UDG) of order 6 with glx and these transcripts as nodes with a two-tailed significance level of 0.05 [64]. We remove edges that are based on spurious or partial correlations, and then orient causal edges between all pairs of connected phenotypes using associated multiple QTLs to break likelihood equivalence. Quantitative trait loci for glx and the selected transcripts were identified with R/qtl [62] using a 3.0 LOD cutoff; the marker closest to each peak provided key information for inferring causal direction. We oriented phenotype edges using our QTL-directed dependency graph (QDG) approach. For any two phenotypes connected by an edge, the direction LOD score was computed by regressing these phenotypes on each other and on their respective multiple QTLs, adjusting for age and for QTL-sex interactions, and by other phenotypes that might be directly connected to either phenotype by an UDG edge. For each edge, we evaluate a LOD score comparing the two possible orientations and we orient the edge in favor of the direction with the higher likelihood in the ratio. P-values for the direction of the edges were computed using 10,000 permutations. Our QDG algorithm used random starts to converge to possible solutions. The best solution was determined by an approximate Bayes factor (BF) [65],[66]. A detailed materials and methods section describing the construction of causal networks is provided in Supporting Protocols (Protocol S1).
We estimated network parameters from the true data and simulated synthetic data according to the causal network in Figure 4. We simulated 1,000 realizations from the causal network and for each edge, we recorded the percentage of undirected edges recovered by the UDG algorithm and the percentage correctly inferred direction by the orientation steps of the QDG algorithm. Overall, the average percentage of true recovered edges was 75% and the average percentage of correctly inferred direction was 83%. False edges were detected at a rate below 2%. To calculate the false discovery rate for the network, we simulated 1,000 data sets from the true network. For each data set, the UDG algorithm was used to infer the network topology, and computed the fraction of false edges (those detected that do not exist in the true network) relative to the total number edges detected by the UDG algorithm. The FDR for the network topology, computed as the average fraction for these 1,000 simulations, is 0.014.
The fold changes relative to the untreated hepatocytes for each animal were calculated. An overall ANOVA analysis was performed with gene transcripts nested within subject; interest focused on gene transcript effects and possible gene transcript differences between strains. This analysis showed that glutamine-induced expression change differed by gene (p<0.0001). Significant overall gene transcript effects allows separate transcript-specific paired t-tests between the difference in delta CT values of untreated and glutamine induced gene expression (relative to Actb) in each strain separately. Statistics on these data were analyzed with Prism software version 4.02 (Graph Pad Software) and the aov command in R (www.r-project.org). |
10.1371/journal.pgen.1004360 | PAX6 Regulates Melanogenesis in the Retinal Pigmented Epithelium through Feed-Forward Regulatory Interactions with MITF | During organogenesis, PAX6 is required for establishment of various progenitor subtypes within the central nervous system, eye and pancreas. PAX6 expression is maintained in a variety of cell types within each organ, although its role in each lineage and how it acquires cell-specific activity remain elusive. Herein, we aimed to determine the roles and the hierarchical organization of the PAX6-dependent gene regulatory network during the differentiation of the retinal pigmented epithelium (RPE). Somatic mutagenesis of Pax6 in the differentiating RPE revealed that PAX6 functions in a feed-forward regulatory loop with MITF during onset of melanogenesis. PAX6 both controls the expression of an RPE isoform of Mitf and synergizes with MITF to activate expression of genes involved in pigment biogenesis. This study exemplifies how one kernel gene pivotal in organ formation accomplishes a lineage-specific role during terminal differentiation of a single lineage.
| It is currently poorly understood how a single developmental transcription regulator controls early specification as well as a broad range of highly specialized differentiation schemes. PAX6 is one of the most extensively investigated factors in central nervous system development, yet its role in execution of lineage-specific programs remains mostly elusive. Here, we directly investigated the involvement of PAX6 in the differentiation of one lineage, the retinal pigmented epithelium (RPE), a neuroectodermal-derived tissue that is essential for retinal development and function. We revealed that PAX6 accomplishes its role through a unique regulatory interaction with the transcription factor MITF, a master regulator of the pigmentation program. During the differentiation of the RPE, PAX6 regulates the expression of an RPE-specific isoform of Mitf and importantly, at the same time, PAX6 functions together with MITF to directly activate the expression of downstream genes required for pigment biogenesis. These findings provide comprehensive insight into the gene hierarchy that controls RPE development: from a kernel gene (a term referring to the upper-most gene in the gene regulatory network) that is broadly expressed during CNS development through a lineage-specific transcription factor that together with the kernel gene creates cis-regulatory input that contributes to transcriptionally activate a battery of terminal differentiation genes.
| The retinal pigmented epithelium (RPE) is a monolayer of polarized and highly specialized pigmented cells that are located between the outer segments of the photoreceptors and the choroid layer in the eye. This strategic location demands multiple functions of the RPE during the development and homeostasis of the adjacent tissues, the neuroretina and choroid [1]. The RPE is a major component of the blood retinal barrier and it therefore determines the microenvironment of the photoreceptors. RPE cells are also responsible for photoreceptor outer segment phagocytosis and are directly involved in retinoid metabolism [1]. An important and evolutionarily conserved role of the RPE is the absorption of stray light to increase visual acuity and reduce oxidative damage. This latter activity requires functional melanosomes, which contain enzymes that catalyze the production of melanin (e.g. tyrosinase, TYR; tyrosinase-related protein, TYRP1; and dopachrome tautomerase, DCT) [2], [3]. Melanosomes accumulate in the RPE during cellular differentiation [4]. Defects in any of these complex functions of the RPE may lead to photoreceptor degradation and, eventually, blindness. Considering the importance of the RPE for ocular physiology and the recent breakthroughs in technologies involving gene transfer and cellular based therapies for treating RPE malfunctions, there is a need to understand the molecular and cellular mechanisms that regulate the acquisition of the various specialized functions of this important tissue.
Most pigmented cells in the body originate from the neural crest. In contrast, the RPE is derived from the neural epithelium of the optic vesicles (OV), which are lateral protrusions of the ventral forebrain. The OV undergo patterning and morphogenesis to give rise to the bilayer optic cup (OC) with an inner layer of retinal progenitor cells and an outer layer populated by the progenitors of the pigmented epithelium. The distal regions of the OC differentiate into the epithelial layers of the ciliary body and iris [5]. The partitioning of the optic neuroepithelium into neuronal and pigmented precursors depends on the activity of extrinsic and intrinsic cues such as transforming growth factor-betab (TGFb) and WNT ligands, which promote RPE development, and fibroblast growth factors (FGFs), which play a role in inducing neuronal fates [6]–[13].
Extrinsic cues trigger the expression and activity of intrinsic factors that execute the differentiation program. A pivotal intrinsic mediator of the RPE fate is microphthalmia-associated transcription factor (MITF), a member of the basic-helix-loop-helix leucine zipper family known to be essential for melanin-bearing pigment cells across species and tissue types [14]. MITF binds the DNA as a homodimer to stimulate the expression of its target genes [15], [16]. MITF is also able to form DNA-binding heterodimers with the related factors TFE3, TFEB and TFEC [17]. The Mitf gene encodes a family of at least 10 distinct isoforms generated from a common gene by alternative promoter/exon usage [18], [19]. Of these, M-Mitf is expressed in the neural crest derived pigment cells [20], [21], and A-Mitf, H-Mitf and D-Mitf are highly expressed in the developing RPE where they are equally distributed [18], [22]. The mRNAs of Mitf isoforms M, A, H and D contain different non-coding and coding 5′ sequences and the corresponding proteins thus differ in their N-terminal sequences [18], [22]. However, each of these protein isoforms can regulate the expression of the melanogenic genes [22]–[26].
In melanocytes, PAX3, SOX10, CREB and the canonical WNT3A signaling pathway regulate the expression of the M-Mitf isoform [27]–[31]. In contrast, in the RPE, where Pax3 is not expressed, data suggest that PAX6 plays a role in regulating the onset of Mitf expression [32]. Pax6, the homolog of the eyeless (ey) gene in Drosophila, is pivotal for development of eye cell types derived from the neuroepithelium or from the surface ectoderm (reviewed in [33] and [34]). Moreover, ectopic expression of Pax6 in frog embryos leads to formation of a differentiated eye, thus demonstrating a role for PAX6 in the different ocular lineages including the RPE [35]. The evolutionarily conserved roles of PAX6 and its upstream regulatory functions suggest that in ocular tissue types Pax6 is a kernel gene, a term referring to the hierarchically upper-most gene in a gene regulatory network [36].
Pax6 is abundantly expressed during patterning of the OV and during specification and differentiation of the RPE. Several studies indicate that PAX6 is important in the specification of the RPE. Recently it was shown that a reduction in Pax6 gene dosage leads to development of neuroretina instead of RPE in embryos that are heterozygous for a mutation in Mitf, while embryos with such a mutation and normal Pax6 levels do not exhibit any detectable phenotype [37]. Furthermore, at the OV stage, the redundant activities of PAX6 and PAX2 are required for the early patterning of the OV by regulating Mitf expression [32]. Later in development, PAX6, but not PAX2, is detected in the RPE [32]. These findings establish a role for PAX6 during the RPE specification stage and imply that PAX6 is also important during the differentiation of the RPE, although its role at this stage is still unknown.
The goal of this study was to examine the roles of Pax6 during RPE differentiation, after the specification of the RPE is established. We show that during the onset of RPE differentiation PAX6 regulates the expression of Mitf and at the same time PAX6 functions together with MITF to activate the expression of downstream targets that execute melanogenesis in the RPE. Our findings reveal the molecular mechanism through which a single transcription factor, which is expressed in multiple ocular and non-ocular cell types, controls a highly specialized differentiation program of the neuroepithelium-derived pigmented cells of the eye.
Once the optic cup has formed (around E10.5), RPE progenitors begin to accumulate melanin [38], [39]. During the initiation of the pigmentation program, the expression of PAX6 is detected throughout the RPE layer (E10- E12.5, Figure 1A). In later stages, the expression of PAX6 is gradually reduced, first in the central and subsequently in the peripheral optic cup (Figure S1A-D). PAX6 is eventually maintained in the pigmented cells of the ciliary body (CB) and iris. To study the role of PAX6 in the RPE after its specification and during the first steps of its differentiation, we generated Pax6loxP/loxP;DctCre mice in which loxP sites are located in exon 4 upstream of the initiator ATG and in intron 6 [5], [40], [41]. The Dct promoter is active in the dorsal side of the OV at E9.5 and by E12.5 its activity is detected in the outer layer of the OC in RPE progenitors [5], [41].
Corresponding with DctCre activity and the location of the loxP sequences, the PAX6 paired domain was lost from the optic cup as evident from labeling with an antibody that specifically identifies the N-terminus of PAX6 (E12.5, Figure 1F and Figure S1A-H, red). Nevertheless, a C-terminal fragment of PAX6 was detected in the Pax6loxP/loxP;DctCre mice when using polyclonal antibodies that detect this region of the protein (Figure S1, green). The expression of this variant lacking the paired domain (PD) of PAX6 (PAX6ΔPD) was transient and reliably mimicked that of the full-length PAX6 during development as the PAX6ΔPD gradually disappeared in a central to peripheral pattern and was eventually lost from the RPE at around birth (Figure S1, green).
We further characterized the expression of Pax6ΔPD transcripts in the mutant RPE: a first Pax6ΔPD transcript variant was generated from the P4 promoter and was also detected in control RPE at E16 by RT-PCR and in situ hybridization using a specific probe (Figure S2A,C,E). A second Pax6ΔPD variant was generated due to aberrant splicing between exon 3 and 7 (Figure S2B,D). Nonetheless, we did not detect over-expression of exons 7-8, located upstream of the homeodomain (HD), in the mutated RPE by quantitative real-time PCR (QRT-PCR, Figure S2F). Thus, the Pax6loxP/loxP;DctCre mice constitute a genetic model for determining the role of the full-length PAX6 protein, while not excluding activities mediated by PAX6ΔPD.
The phenotype of the Pax6loxP/loxP;DctCre eyes was evident during embryogenesis as the iris and CB progenitors, which are evident at E19.5 (Figure 1C), did not develop in the Pax6loxP/loxP;DctCre OC (Figure 1H) in agreement with a previous report [5]. In addition, reduced pigmentation in the RPE was noted from early stages of RPE differentiation (E12.5, Figure 1B,G) and was evident when viewing the whole eye of Pax6loxP/loxP;DctCre as compared to control litter mates (E19.5, Figure 1C,H) or in flat mount (E19.5, Figure 1K,M).
Although pigmentation was reduced, the fate of the RPE was maintained in the Pax6-mutant RPE based on the expression of the transcription factors Otx2 and Sox9 (Figure S3). Consistent with maintenance of RPE fate, transmission electron microscope (TEM) analysis conducted on E15.5 control and Pax6loxP/loxP;DctCre eyes (Figure 1D,E,I,J) demonstrated that the typical RPE morphology of a single layer was preserved despite reduction in pigmentation. The adjacent structures of the choriocapillaris and neuroretina maintained normal morphology despite Pax6 loss in the RPE (Figure 1D,I). We next examined actin distribution by phalloidin staining in flat mounts of the RPE and observed that the typical polygonal morphology of the RPE was maintained (Figure 1L,N). Moreover, using QRT-PCR analysis we did not detect significant differences between control and Pax6-deficient RPE in the levels of mRNAs encoding the intercellular junction proteins ZO-1, Connexin-43 and P-cadherin (E15.5, Figure 1O). These findings reveal a role for PAX6 in execution of the pigmentation program, although its absence does not alter the fate and morphology of the RPE at the OC stage.
To determine the global change in gene expression following Pax6 loss in the OC we determined the transcript profile in control and Pax6loxP/loxP;DctCre E15.5 RPE using Affymetrix GeneChip Mouse Gene 1.0 ST arrays. Of the 28,853 genes represented on the microarray, levels of 100 transcripts were significantly altered in mutant RPE, compared to the wild-type (fold change greater than 1.5, p<0.05, Table S1). The expression of 73 of these genes was reduced in the Pax6-deficient RPE. In agreement with the observed phenotype, analysis of enrichment in GO categories revealed significant representation of melanogenic genes (p<0.05; using the ToppGene Suite algorithm; [42]) as summarized in Table 1. The identified pigmentation genes encode key enzymes of melanogenesis (Tyr and Tyrp1), as well as factors involved in melanosome biogenesis (Si, Mlana, and Gpr143) or melanosome transport (RAB27a) and factors implicated in melanosome biogenesis (Gpnmb, Slc45a2, Slc24a5 and Slc3a2). Corresponding to the phenotype observed, the transcriptome analysis indicates an arrest in the melanogenesis program following Pax6 loss.
To validate the microarray results, six melanogenic genes were analyzed by QRT-PCR (Figure 2A). In agreement with the microarray results, transcript levels of Tyr, Tyrp1, Si and Mlana were significantly reduced in the mutant RPE as compared with control, whereas the level of the mRNA encoding the enzyme DCT, which is involved in melanin synthesis, was slightly reduced, and the level of Myo7a mRNA, which encodes a protein involved in cellular transport of melanosomes in the RPE [43], was similar to wild-type. The reductions of Si transcript (Figure 2B,E) and of TYR and TYRP1 proteins (Figure 2C,D,F,G) were validated by in situ hybridization and antibody labeling, respectively. These findings support a role for PAX6 in the proper expression of key melanogenic genes in the RPE.
The transcription factor Mitf is considered the master regulator of all melanin-bearing pigment cells and several melanogenic genes are direct targets of MITF [44], [45]. Out of the 10 melanogenic genes found to be down-regulated following Pax6 loss, seven are known direct targets of MITF: Tyr and Tyrp1 [46], Si and Mlana [47], Gpnmb [48], Rab27a [49] and Gpr143 [50]. Mitf has previously been found to be regulated by PAX6 and PAX2 at the OV stage [32]. We therefore wanted to investigate whether Mitf expression is dependent on PAX6 after RPE specification, when PAX2 is not expressed in the pigmented epithelium [32].
The Mitf gene encodes a family of isoforms generated from a common gene. The isoforms that are predominantly expressed in the RPE are A, H and D [18] (Figure S4A,B). The average response of all the Mitf probes in the GeneChip array revealed a reduction in the transcript levels by 1.37 fold (p = 0.069) in Pax6-deficient RPE compared to the wild-type. By indirect immunofluorescence (IIF) analysis we indeed detected reduced levels of MITF protein in the RPE of Pax6loxP/loxP;DctCre embryos as compared with control Pax6loxP/loxP embryos at E12.5 and at E15.5 (Figure 3A-D). We next determined the expression levels of the specific Mitf isoforms by QRT-PCR of RNA extracted from control and Pax6-deficient RPE (E15.5). This analysis revealed a significant reduction in the expression of D-Mitf, which was over 3-fold lower in the mutants, and slight elevations in levels of A-Mitf and H-Mitf in the mutant RPE, compared to wild-type (Figure 3E). Quantification of a downstream amplicon that is common to all Mitf isoforms revealed a significant 1.45-fold reduction in pan-Mitf transcripts, consistent with the microarray results. In silico analysis of the upstream regulatory region of D-Mitf revealed three putative binding sites for PAX6 PD and four for MITF (Table S2); all are located within the 1200bp preceding the D-Mitf transcription start site (TSS) (Figure 3F). An electrophoretic mobility shift assay (EMSA) revealed that PAX6 binds two of the three sites in vitro (Figure 3G). This binding was specific, as it was competed by a cold probe (Figure 3G). Luciferase reporter assay on regulatory sequences of D-Mitf (between −1,153 and +6 relative to the TSS) was performed using different combinations of Pax6, Pax6ΔPD and Mitf expression vectors (Figure 3H). This analysis revealed synergistic transactivation of the D-Mitf promoter by PAX6 and MITF. The co-transfection of MITF with PAX6ΔPD failed to produce the same result.
In order to identify the regulatory sequence required for PAX6 and MITF transcriptional activity, a series of D-Mitf truncated promoters was analyzed (Figure 3H). The critical regulatory sequence for PAX6 and MITF transactivation is located between −310 and −180 relative to the TSS. This region contains only a PAX6 binding site (site 3: −212 and −194 relative to the TSS) but no known MITF binding site. Taken together, these results suggest that the transactivation of D-Mitf promoter by PAX6 and MITF depends on PAX6 PD and requires the PD binding site. These results also indicate that a self-sustaining PAX6-dependent feedback loop controls Mitf expression.
The abrogated melanogensis in the Pax6-deficient RPE and the corresponding reduction in the D-Mitf isoform suggested that the pigment depletion in the Pax6-deficient RPE could be mediated by D-Mitf. Recently mice with specific deletion of D-Mitf were generated by ablation of 0.2 kb downstream to exon D, exon D and the 5.6 kb preceding sequence (MitfΔD, Figure S4C). In these mice, a slight reduction in pigmentation was observed at E11; however, at later stages the pigmentation was comparable to normal, in contrast to the depigmentation observed in the Pax6loxP/loxP;DctCre mutant eyes [37] (Figure 4A-C). Molecular analysis of MitfΔD/ΔD by IIF and QRT-PCR revealed that pan-Mitf level was similar to that in wild-type mice (Figure 4D,E). Yet, the expression levels of Mitf variants in the MitfΔD/ΔD RPE were similar to the pattern observed in the Pax6loxP/loxP;DctCre mutants: D-Mitf transcript level was completely abolished while A-Mitf and H-Mitf expression levels were elevated (Figure 4E). In addition, transcript quantification of the six melanogenic genes examined in the Pax6-deficient mutants revealed minor reductions in the levels of Tyrp1, Si and Myo7a, but only the reductions of the latter two were significant (Figure 4F). The normal phenotype of MitfΔD/ΔD mice is probably due to redundant activity of the Mitf isoforms expressed in the RPE. Together, the above results reveal that while PAX6 is required for normal levels of expression of D-Mitf, the reduced levels of D-Mitf following Pax6 loss cannot account for the observed arrest in melanogenesis in the Pax6loxP/loxP;DctCre mutants. Moreover, the dramatic loss of pigmentation, while levels of Mitf are partly maintained, indicates that PAX6 has other functions in melanogenesis of the RPE in addition to the regulation of Mitf levels.
The findings above reveal that PAX6 plays a pivotal role in the pigmentation program that goes beyond regulation of D-Mitf expression. This is reminiscent of the activity of the Pax3 gene in melanocyte precursors, where it regulates the onset of Mitf expression as well as the expression of Mitf target genes like Tyrp1 [51]. In order to examine the ability of PAX6 to trans-activate known targets of MITF we performed luciferase reporter assays using the regulatory regions of three pigmentation genes: mTyrp1, hTyr and mMlana (see Tables S3-S5 for details on MITF and PAX6 putative binding sites). We also examined the transcriptional activity of PAX6ΔPD, which was detected in Pax6loxP/loxP;DctCre mice (Figure S1 and Figure S2), on these promoters. On the mMlana promoter there was additive cooperation between the two transcription factors (PAX6: 2.25 fold change, p = 0.02; MITF: 50.3 fold change, p = 0.004; MITF + PAX6: 149 fold change, p≤0.04; n = 3; Figure S5). In contrast to PAX3, PAX6 by itself failed to activate the mTyrp1 promoter either in pigment producing cell lines such as ARPE19 and UACC.62, or in HEK-293T, NIH-3T3 and HeLa cells (Figure 5A and data not shown). However, in the presence of MITF, PAX6 cooperatively and synergistically trans-activated the mTyrp1 promoter (MITF: 5.6 fold change, p = 0.006; MITF + PAX6: 51.9 fold change, p = 0.002; n = 4; Figure 5A). A similar synergistic transactivation pattern was observed using the hTyr promoter (MITF: 12.6 fold change, p = 0.001; MITF + PAX6: 71.3 fold change, p = 0.003; n = 3; Figure 5B). Chromatin immunoprecipitation (ChIP) confirmed the association of PAX6 with the hTyrp1 promoter region in RPE cells derived from human embryonic stem cells [52]. We observed more than 4-fold enrichment of PAX6 in the hTyrp1 proximal promoter compared to a region 2 kb downstream (data not shown and Table S6).
We next examined the contributions of the putative binding sites of PAX6 and the binding sites of MITF (M- and E-box) to the transactivation of the promoters of mTyrp1 and hTyr by PAX6 and MITF. Interestingly, deletion or point mutations in the MITF binding sites dramatically reduced the transactivation observed when MITF and PAX6 were co-expressed. While the wild-type mTyrp1 promoter was trans-activated 51.9 fold in the presence of both factors, compared to their absence, a promoter carrying a deletion or mutations in the M-box was trans-activated only 3.5 fold (Figure 5A). Similarly, the wild-type hTyr promoter was trans-activated 71.3 fold by both PAX6 and MITF, whereas promoters carrying mutations in the M- and E-box sequences were trans-activated 43.3 fold and 13.3 fold, respectively (Figure 5B). In contrast, deletion of the putative binding site for PAX6 in mTyrp1 promoter did not significantly alter the transactivation by MITF and PAX6 (Figure 5A). To examine whether the M-box is sufficient to enhance MITF activity by PAX6, we performed a luciferase reporter assay with PAX6, MITF or both using a reporter with four consecutive M-box elements. As shown in Figure 5C, PAX6 alone did not activate the promoter, MITF alone enhanced the activity by 6.6 fold, and PAX6 and MITF together enhanced the reporter activity by 11.9 fold (p = 0.046, n = 3). These findings suggest that in tissue culture, the MITF binding sites are essential and sufficient for the transactivation of mTyrp1and hTyr promoters by PAX6 and MITF.
The reporter assays revealed that PAX6 transactivation effects are largely dependent on MITF expression and on its binding sites. This mode of action suggests a physical interaction between PAX6 and MITF. We therefore conducted co-immunoprecipitation assays (co-IP) to evaluate this possibility. First, a reciprocal co-IP experiment was performed in ARPE19 cells that endogenously express both Pax6 and Mitf [53]. MITF antibodies co-precipitated PAX6, and immunoprecipitation with PAX6 antibodies resulted in precipitation of MITF (Figure 5D). The enrichment of MITF in the PAX6 immunoprecipitate was very significant as MITF expression was almost below detection in the input sample (Figure 5D, right panel, lane 5). These results support an association of MITF and PAX6 in ARPE19 cells.
To determine whether the PAX6ΔPD variant is capable of physical association with MITF, HeLa cells were transfected with 3xFLAG-PAX6, 3xFLAG-PAX6ΔPD, 3xFLAG-MITF or a combination of 3xFLAG-MITF with each PAX6 protein variant (Figure S6A). Cells were harvested and protein extracts were precipitated using MITF antibodies. Both PAX6 and PAX6ΔPD proteins were enriched in the immunoprecipitates when co-transfected with MITF (Figure S6A, right panel, lanes 9 and 10). These results suggest that the PAX6ΔPD variant is capable of associating with MITF as previously suggested [54]. We next conducted luciferase reporter assays using MITF and PAX6ΔPD. PAX6ΔPD had no transactivation effects on the transcriptional activity of MITF (Figure 5A,B) and did not show a dominant negative effect on the transactivation of the mTyrp1 promoter by MITF and PAX6 (Figure S6B). These results indicate that although the PAX6ΔPD variant is capable of association with MITF, the PD domain is necessary for the PAX6-MITF-mediated transcriptional activation of melanogenic genes.
This study unravels the molecular mechanism through which a single transcription factor, which is expressed in multiple ocular and non-ocular cell types, controls a highly specialized differentiation program of the neuroepithelium-derived pigmented cells of the eye. We show that PAX6 regulates a gene regulatory network central to RPE differentiation. This activity is mediated by a coherent feed-forward loop, by which PAX6 controls the expression of Mitf and jointly with MITF triggers the expression of multiple downstream target genes that are required for the execution of distinct differentiation program of pigment formation (Figure 6). In this mode of action, MITF levels could serve as a sign-sensitive delay for the melanogenesis process in the RPE as transactivation of pigmentation genes by PAX6 depends on Mitf transactivation by PAX6. This type of kinetic mechanism filters out fluctuations in input stimuli since it requires persistent co-expression of both transcription factors. Our data provide an explanation of how PAX6, which is expressed in most ocular lineages, can promote the highly specialized and distinct differentiation program of the RPE.
A role for Pax6 during specification of the OV to the PE lineage was deduced from the analyses of Pax6 mutants that also carry mutations in the transcription factors Pax2 or D-Mitf. The PE of the transgenic Pax2−/−;Pax6+/− and MitfΔD/ΔD;Pax6+/− mice develops into a second neuroretina [32], [37]. In contrast, RPE cells that lost the expression of Pax6 after specification maintained their morphology of a single layer of polygonal epithelium (Figure 1). Accordingly, we did not detect changes in the expression of several epithelial markers (Figure 1O) or elevated expression of the neuronal gene CHX10 (Figure 3A-D) in the Pax6-mutant RPE. Although the mutant RPE transcriptome data did not reveal overt elevations in neuronal genes, we did detect alterations in the expression levels of MITF-regulators, both key RPE-specification factors such as Wnt2b ([6]; −1.67, p = 0.00005) and Gli2 ([55]; −1.1, p = 0.003) and of the retinal promoting gene Msx2 ([56]; +1.8, p = 0.016). These alterations in gene expression suggest partial changes in the differentiation program of the Pax6-deficient RPE and point to additional regulators of MITF that are controlled by PAX6. However, these changes were not sufficient to completely disrupt RPE differentiation, in contrast to the complete disruption observed following inhibition of the Wnt/β-catenin pathway in specified RPE [57]. A previous study showed that PAX6 together with PAX2 is required for expression of MITF during the specification stage, and that the former two proteins regulate the expression of the A-Mitf isoform in vitro [32]. In our analysis, the loss of Pax6 during RPE differentiation resulted in up-regulation of A-Mitf. These findings suggest stage-dependent roles for PAX6 during various stages of RPE development, from pattering to differentiation.
We show that Pax6 is essential for the proper expression of Mitf and its melanogenic target genes. This activity requires both PAX6 and MITF to act synergistically, as shown by luciferase reporter assays on the promoters of mD-Mitf, mTyrp1 and hTyr (Figure 3H and Figure 5A,B). Although PAX6 binding sites were identified in each of these three promoters (Tables S2-S4), the deletion of the PAX6 binding site in the Tyrp1 promoter did not reduce the transcriptional activity of PAX6 and MITF, while mutations in MITF binding sites in either mTyrp1 or hTyr promoters hampered their activity (Figure 5A,B). In contrast, the PAX6 binding site in the D-Mitf promoter (site 3, Figure 3F,G), but not the putative MITF binding sites, was essential for PAX6 and MITF transcriptional activity (Figure 3H). Together with the observation that PAX6 and MITF are capable of physical association (Figure 5D), these results suggest that the PAX6-MITF complex may trans-activate promoters through either PAX6 or MITF binding sites. This mode of action of PAX6 may account for the broad spectrum of PAX6 transcriptional targets. ChIP-Seq studies on embryonic RPE or hES-RPE cells are underway to determine the promoter occupancy of PAX6 and MITF on RPE genes.
The Mitf gene encodes at least 10 isoforms with alternative promoter or exon usage. The three RPE-specific isoforms (D-, A- and H-Mitf; [18], [19]) differ only in the N-terminal sequences [22]. The fact that all Mitf isoforms have different promoter sequences and predicted transcription factor binding sites suggests different regulation mechanisms [58]. Our data show that PAX6 is specifically required for the normal expression of D-Mitf. In both mouse mutant lines – Pax6loxP/loxP;DctCre and Mitf ΔD/ΔD –we observed an up-regulation of A-Mitf and H-Mitf isoforms (Figure 3E and Figure 4E). However, while in the RPE of MitfΔD/ΔD mice the total transcript level of Mitf was similar to that in the wild-type RPE, in the RPE of Pax6loxP/loxP;DctCre the level of pan-Mitf was 1.45-fold lower than the wild-type RPE. We therefore suggest that there is a feedback mechanism that balances the total level of MITF protein and that this mechanism requires full-length PAX6.
The observation that A- and H-Mitf are capable of compensating for the absence of D-Mitf activity in the MitfΔD/ΔD transgenic mice, but not in the Pax6loxP/loxP;DctCre mutants, suggests that reduction in pigmentation in the RPE of Pax6loxP/loxP;DctCre transgenic mice might be caused by down-regulation of a MITF co-factor, either PAX6 itself or another protein. Other than MITF, the only transcription factor that has been demonstrated to have a role in regulation of RPE melanogenesis is OTX2. OTX2 plays an important role in RPE development [59], [60] by trans-activating the melanogenic enzymes-encoding genes Tyr, Tyrp1 and Dct [61]–[63]. Since we did not detect a significant change in the expression pattern of Otx2 in the RPE of Pax6loxP/loxP;DctCre transgenic mice (Figure S3), it is unlikely that changes in its expression mediate the reduction in the expression of the melanogenic genes and pigmentation observed following Pax6 loss.
Another candidate that might be responsible for the reduced pigmentation in the Pax6loxP/loxP;DctCre is the bHLH leucine-zipper transcription factor TFEC. The amino acid sequences of TFEC and MITF bHLH leucine-zipper show high similarity [64] and these two proteins bind to an E-box as heterodimer complex [17]. Bharti et al. (2012) showed that PAX6 trans-activates the expression of Tfec and that Tfec can rescue eye defects in mice with a mutation in the Mitf gene [37]. The transcript level of Tfec was indeed reduced in Pax6loxP/loxP;DctCre mutants (Figure S7A: −1.23 fold change, p = 0.075, n = 6), and this minor down-regulation may have also contributed to the overall reduction in pigmentation, as TFEC is capable of trans-activating mTyrp1 and hTyr promoters alone and synergistically with PAX6 (Figure S7B and data not shown). Thus, in addition to its known role during PE specification, TFEC may also have a role in RPE differentiation where it acts like an additional isoform of MITF [17].
Association between PAX6 and MITF was previously shown in vitro by Planque et al. (2001). However, in that study transfection of the two proteins caused a reduction in MITF transactivation of Tyr promoter [54]. The discrepancy with our results could be explained by the different ratio of Pax6/Mitf levels used in the reporter assays, as Planque et al. used a ratio of 20∶1, whereas in our study the ratio was 1∶1. The importance of the ratio between PAX6 and MITF has also been demonstrated in vivo. The transcript levels of Tyr and Tyrp1 were lower in mice that over-express Pax6 (i.e. Pax6Yac/Yac mice) on the background of a MitfΔD/ΔD transgenic mice compared to either Pax6Yac/Yac or MitfΔD/ΔD alone [37]. In these experiments, RPE pigmentation levels were consistent with the altered expression levels of Tyr and Tyrp1 ([37] and Bharti, unpublished results). Interestingly, while during embryonic development PAX6 is eliminated from the RPE in a proximal to distal gradient, MITF and its downstream pigmentation genes are expressed along the entire length of the RPE. Thus, we infer that PAX6 is involved in the initiation of the pigmentation program but not in its maintenance.
The somatic mutation induced by the DctCre transgene deleted exons 4−6, which encode the initiation codon and the PD of PAX6. Interestingly, while the PD was eliminated from the Pax6loxP/loxP;DctCre embryos, a truncated transcript of Pax6 that gave rise to a Pax6ΔPD variant was identified. The PAX6ΔPD variant was not previously noted in somatic mutations of the Pax6loxP allele [40], , probably because in some tissues, such as the lens placode and the peripheral optic cup, but not in the RPE, the expression of Pax6 depends on full-length PAX6 protein [70], which is absent due to the Cre-mediated deletion. In addition, it is possible that RPE-specific post-transcriptional mechanisms that alter splicing and RNA stability lead to more prominent accumulation of the PAX6ΔPD in the RPE.
While the physiological activity of PAX6ΔPD in the eye is still unknown, its over-expression results in microphthalmia due to aberrant lens and corneal development [71], [72]. Thus, although we did not detect over-expression of the homeodomain of Pax6, we should consider the possible contribution of the over-expression of the PAX6ΔPD isoform and the disruption of the PAX6/PAX6ΔPD ratio to the pigment phenotype of the Pax6loxP/loxP;Dct-Cre RPE. There are several lines of evidence that rule out a major effect of the PAX6ΔPD isoform in the Pax6loxP/loxP;Dct-Cre mutants: First, we detected a Pax6ΔPD transcript in the control RPE, which was initiated from the P4 promoter (Figure S2A,C,E). Thus, the Pax6ΔPD transcript is expressed during normal differentiation and onset of pigmentation in the RPE. Second, Pax6ΔPD is expressed in the progenitors of the CB and is maintained there in the adult, both in the pigmented and non-pigmented epithelium [72]. Yet, mice carrying 10 copies of the Pax6 locus and over-expressing Pax6ΔPD do not display any alteration in the pigmentation of the CB [72]. Therefore the PAX6ΔPD isoform is unlikely to interfere with the pigmentation program. Third, we did not detect reduced pigmentation in the RPE of Pax6loxP/+;DctCre heterozygous mice, thus further arguing against a dominant-negative effect of this truncated product ([5] and data not shown). Finally, even though the PAX6ΔPD isoform was able to associate with MITF in a co-IP assay (Figure S6A), it had no repressive or inductive effects on the promoters of mTyrp1 and hTyr either alone or when co-expressed with MITF or together with MITF and PAX6 (Figure 5A,B and Figure S6B).
Although there is little evidence for independent eye invention events during metazoan species evolution [34], there is strong argument in favor of a common molecular network controlling the development of the metazoan eye, in which Pax genes were redundantly employed and were later on variably adapted for eye development in different animal taxa [34], . According to this hypothesis an ancestor of a Pax6 gene was at the node of a gene regulatory network that controlled the morphogenesis of a primitive eye composed of a photoreceptor cell that contained pigment granules, as in Palaemonetes pugio [76]. The evolutionarily earliest gene regulatory networks were likely to be hierarchically shallow and, as animal body parts gradually elaborated and gained more complex regional subdivision of the developing embryo, the underlying regulatory networks became hierarchically deeper and were terminally fixed into kernel genes, in which any minor change would lead to extremely harmful consequences [36], [77]. In such a scenario, the development of the vertebrate eye into a complex structure that includes PE and multilayered neuroretina would require different cell specific transcription factors that in combination with PAX6 generate different cis-regulatory input functions that result in execution of distinct and highly specified differentiation programs. In this model, PAX6 acts as an accelerator directed by its tissue-specific partner to a specific transcriptional program.
The mouse lines employed in this study, MitfΔD [37], Pax6loxP [40] and DctCre [5] have been previously described. The latter two were used to establish Pax6loxP/loxP;DctCre somatic mutants. Pax6loxP/loxP littermates were used as controls. The genetic background of all mice used in this study was C57BL/6J, except for in situ staining, for which mice of the outbred ICR genetic background were used. All animal work was conducted according to national and international guidelines and approved by the Tel Aviv University review board.
All data were examined using two-tailed Student's t-test.
Immunofluorescence analysis was performed on 10 µm paraffin sections as previously described [40], using the following primary antibodies: rabbit anti-PAX6 (1∶400, Covance, prb-278b), mouse anti-PAX6 (1∶25, Santa Cruz, sc-32766), rabbit anti-SOX9 (1∶200, Chemicon, ab5535), sheep anti-CHX10 (1∶1000, Exalpha, X1180P), rabbit anti-OTX2 (1∶50, Millipore, AB9566), rabbit anti-MITF [18], rabbit anti-TYR (1∶1,000, a gift from the Vincent Hearing lab, NCI), rabbit anti-TYRP1 (1∶1,000, a gift from the Vincent Hearing lab, NCI). Secondary antibodies were donkey anti-rabbit conjugated to alexa594 (1∶1000, Invitrogen, A21207) and alexa488 donkey anti mouse/sheep (1∶1000, Invitrogen, A21202/A11015).
In situ hybridization (ISH) was performed on 14 µm cryo-sections using DIG-labeled RNA probes as previously described [78]. The Pax6 intron 7 probe was generated from a 849bp PCR fragment (forward, 5′-TTTGGAGCCCTCCATCTTTCTC-3′; reverse, 5′- TGCACACTTTCGGGCAAGG-3′). Plasmid for antisense transcription of silver was kindly provided by the laboratory of Dr. William Pavan (NIH) [79].
Flat-mount samples were prepared as follows: Eyes at E19.5 were enucleated and immediately fixed in 4% paraformaldehyde for 30 minutes. The RPE was carefully dissected from the rest of eye structures, sliced radially to four pieces and flattened on membrane filters (Schleicher& Schull, 0.45 µm D-37582). Samples were blocked and stained with phalloidin (1∶100, Invitrogen, A12379). Thereafter, RPE was flattened on its basal side on a slide and sealed for observation.
The heads of E15.5 embryos and perforated eyes of P1 neonatal mice were fixed in 0.1 M cacodylate-buffered fixative containing 2.5% paraformaldehyde and 2% glutaraldehyde and further processed as described previously [80]. Ultrathin sections were cut with a Leica Ultramicrotome UCT (Leica Microsystems), stained with uranyl acetate and lead citrate and analyzed with a H7600 transmission electron microscope (Hitachi).
Exact timed matings were performed by overnight cohabitation of an inbred Pax6loxP/loxP;DctCre male with Pax6loxP/loxP females. Pregnant females were harvested on day E15.5 and the RPE of the embryos was dissected as previously described [18]. RPEs were pooled into two separate tubes according to their pigmentation intensity, and tubes were stored at −80°C. Tail cuts of the embryos were collected for genotype verification. Each tube was considered as one biological repeat. RNA was extracted using the QIAshredder and the RNeasy kits (QIAGEN). RNA isolated from three control and three mutant samples was processed for microarray analysis using the Affymetrix GeneChip 1.0ST as described previously [81]. Differentially expressed genes with p-values lower than 0.05 and with a fold-change cutoff of 1.5 are listed in Table S1. The expression data were submitted to the NCBI Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo) under series accession no. GSE56548.
Reverse transcription of 1 µg of RNA from each sample was performed using the SuperScript III First Strand kit (Invitrogen). cDNA was amplified using the Power SYBR Green Mix (Applied Biosystems) in a 384-well optical reaction plate using ABI Prism 7000 Sequence Detection System (Applied Biosystems). All primer pairs were first tested for specificity and amplicon size using end-point PCR. Formation of a dimer structure was refuted by analyzing the dissociation curve at the end of each amplification reaction. Results were calibrated in relation to an average of two house-keeping genes, Ppia and Tbp, after verifying that their levels were consistent in normal and mutant RPE. Raw data was processed using the comparative Ct method by the formula 2−ΔΔCT. Each amplification reaction was performed in triplicate using 20 ng of cDNA for each sample. Primers used to amplify and sequence the two Pax6ΔPD transcript variants are listed in Table S6.
Reporter assays were performed in HeLa cells using the Dual-Luciferase Reporter Assay System (Promega). Cells were seeded in a 24-well plate and 24 hours later were transfected using jetPEI DNA transfection reagent (Polyplus-transfection). Each well was co-transfected with three types of vectors in a total amount of 1210 ng of DNA: 1) 400 ng of a luciferase reporter vector (pGL3 basic) under the regulation of the examined promoter; 2) A total of 800 ng of expression vector (p3XFlag-CMV-10), either carrying no insert or containing an insert encoding the ORF of Pax6, Pax6ΔPD or A-Mitf, 400 ng of each; 3) 10 ng normalizing vector (pRL-TK). Cells were harvested 48 hours after transfection and luminescence was evaluated. Each treatment was carried out in duplicate, and each assay was repeated at least three times.
End-point PCR of 17 cycles was performed using oligonucleotides containing the desired mutated sites (Table S6, mutated nucleotides are in lower case) and the wild-type promoter reporter plasmid (pGL3 basic) as template. The PCR products were treated with 12U DpnI restriction enzyme (Fermentas) for 1 hour at 37°C, and 5 µl of the DNA was transformed into E. coli XL-1Blue strain, followed by colony-picking mini-prep and midi-prep plasmid purification (Qiagen). All plasmids were verified by sequencing.
Transfection into HeLa cells was performed as described in the reporter assay section. Cells were seeded in 90-mm dishes and transfected with total of 10 µg of DNA. Cells were washed with phosphate buffered saline (PBS), scraped in 1 ml lysis buffer (10 mM HEPES pH 8.0, 100 mM NaCl, 1 mM MgCl2, 0.5% NP-40) containing protease inhibitor (Roche, complete Mini EDTA-free) and incubated on ice for 30 minutes. Extracts were clarified by centrifugation at 10,000 g for 15 minutes at 4°C. To avoid nonspecific binding of proteins to the beads, extracts were subjected to pre-clearing using 15 µl of protein A agarose beads (Millipore, 16-157) for 2 hours at 4°C, followed by centrifugation at 10,000 g for 1 minute at 4°C. Input samples (50 µl of the supernatant) were kept at −20°C for input analysis and the cleared extracts were incubated with 5 µl of rabbit anti-MITF (kindly provided by David E. Fisher, MGH [82]) with rotation overnight at 4°C. The resulting immuno-complexes were incubated with 30 µl of protein A beads for 2 hours at 4°C. The beads were then washed four times with RIPA buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 1% NP-40, 0.5% Na-deoxycholate, 0.1% SDS) and the complexes were eluted in SDS sample buffer by boiling for 5 minutes. Samples were subjected to SDS–polyacrylamide gel electrophoresis. Separated proteins were transferred to nitrocellulose membrane and reacted with mouse anti-FLAG antibody (1∶10,000, Sigma F3165) followed by anti-mouse horseradish peroxidase-conjugated secondary antibody. The reaction was examined by enhanced chemiluminescence detection kit (Biological Industries).
Co-immunoprecipitation from ARPE19 cells was performed essentially as described above, except cells were scraped in RIPA buffer containing protease inhibitor (Roche, complete Mini EDTA-free). Antibodies used for IP were either rabbit anti-PAX6 (Millipore, AB2237) or rabbit anti-MIT (kindly provided by David E. Fisher, MGH). Mouse anti-PAX6 (Santa Cruz, sc-3276) and mouse anti-MITF (kindly provided by David E. Fisher, MGH [82]) antibodies were used for immunoblot.
ChIP was performed as previously described [83], [84]. Briefly, hES-RPE cells were grown as described [52]. Fixed chromatin was extracted from 2×107 cells and immunoprecipitated using rabbit anti-PAX6 (Millipore, AB2237) or non-immune rabbit IgG (Rockland) as a negative control. The primers used for ChIP analysis are listed in Table S6.
HEK-293T cells were transfected with p3XFlag-CMV-10 encoding the ORF of full-length Pax6. Nuclear extracts were obtained as previously described [85]. Nuclear extract (1 ml) or 1∶10 diluted nuclear extract was incubated for 10 minutes on ice in 8.5 mM HEPES pH 7.9, 30 mM KCl, 1.5 mM MgCl2, 0.4 mM DTT, 2 mg polydI/dC (Sigma). Binding with 1 ml double-stranded 59-c-ATP-labeled probe (30,000 cpm) was performed at room temperature for 20 minutes and 200 ng of ‘‘cold’’ PAX6 consensus site (PAX6CON) was used for competition [86].
|
10.1371/journal.pcbi.1003047 | Improving Breast Cancer Survival Analysis through Competition-Based Multidimensional Modeling | Breast cancer is the most common malignancy in women and is responsible for hundreds of thousands of deaths annually. As with most cancers, it is a heterogeneous disease and different breast cancer subtypes are treated differently. Understanding the difference in prognosis for breast cancer based on its molecular and phenotypic features is one avenue for improving treatment by matching the proper treatment with molecular subtypes of the disease. In this work, we employed a competition-based approach to modeling breast cancer prognosis using large datasets containing genomic and clinical information and an online real-time leaderboard program used to speed feedback to the modeling team and to encourage each modeler to work towards achieving a higher ranked submission. We find that machine learning methods combined with molecular features selected based on expert prior knowledge can improve survival predictions compared to current best-in-class methodologies and that ensemble models trained across multiple user submissions systematically outperform individual models within the ensemble. We also find that model scores are highly consistent across multiple independent evaluations. This study serves as the pilot phase of a much larger competition open to the whole research community, with the goal of understanding general strategies for model optimization using clinical and molecular profiling data and providing an objective, transparent system for assessing prognostic models.
| We developed an extensible software framework for sharing molecular prognostic models of breast cancer survival in a transparent collaborative environment and subjecting each model to automated evaluation using objective metrics. The computational framework presented in this study, our detailed post-hoc analysis of hundreds of modeling approaches, and the use of a novel cutting-edge data resource together represents one of the largest-scale systematic studies to date assessing the factors influencing accuracy of molecular-based prognostic models in breast cancer. Our results demonstrate the ability to infer prognostic models with accuracy on par or greater than previously reported studies, with significant performance improvements by using state-of-the-art machine learning approaches trained on clinical covariates. Our results also demonstrate the difficultly in incorporating molecular data to achieve substantial performance improvements over clinical covariates alone. However, improvement was achieved by combining clinical feature data with intelligent selection of important molecular features based on domain-specific prior knowledge. We observe that ensemble models aggregating the information across many diverse models achieve among the highest scores of all models and systematically out-perform individual models within the ensemble, suggesting a general strategy for leveraging the wisdom of crowds to develop robust predictive models.
| Breast cancer remains the most common malignancy in females, with more than 200,000 cases of invasive breast cancer diagnosed in the United States annually [1]. Molecular profiling research in the last decade has revealed breast cancer to be a heterogeneous disease [2]–[4], motivating the development of molecular classifiers of breast cancer sub-types to influence diagnosis, prognosis, and treatment.
In 2002, a research study reported a molecular predictor of breast cancer survival [5] based on analysis of gene expression profiles from 295 breast cancer patients with 5 year clinical follow-up. Based on these results, two independent companies developed the commercially available MammaPrint [6] and Oncotype DX [7] assays, which have both been promising in augmenting risk prediction compared to models based only on clinical data. However, their role in clinical decision-making is still being debated.
Based on the success of these initial molecular profiles, a large number of additional signatures have been proposed to identify markers of breast cancer tumor biology that may affect clinical outcome [8]–[13]. Meta-analyses indicate that many of them perform very similarly in terms of risk prediction, and can often be correlated with markers of cell proliferation [14], a well-known predictor of patient outcome [15], especially for ER+ tumors [16], [17]. Therefore, it is much more challenging to identify signatures that provide additional independent and more specific risk prediction performance once accounting for proliferation and clinical factors. Recent studies have even suggested that most random subsets of genes are significantly associated with breast cancer survival, and that the majority (60%) of 48 published signatures did not perform significantly better than models built from the random subsets of genes [18]. Correcting for the confounding effect of proliferation based on an expression marker of cell proliferation removes most of the signal from the 48 published signatures [18].
The difficulties in reaching community consensus regarding the best breast cancer prognosis signatures illustrates a more intrinsic problem whereby researchers are responsible for both developing a model and comparing its performance against alternatives [19]. This phenomenon has been deemed the “self-assessment trap”, referring to the tendency of researchers to unintentionally or intentionally report results favorable to their model. Such self-assessment bias may arise, for example, by choosing assessment statistics for which their model is likely to perform well, selective reporting of performance in the modeling niche where their method is superior, or increased care or expertise in optimizing performance of their method compared to others.
In this work, we explore the use of a research strategy of collaborative competitions as a way to overcome the self-assessment trap. In particular, the competitive component formally separates model development from model evaluation and provides a transparent and objective mechanism for ranking models. The collaborative component allows models to evolve and improve through knowledge sharing, and thereby emphasizes correct and insightful science as the primary objective of the study.
The concept of collaborative competitions is not without precedent and is most evident in crowd-sourcing efforts for harnessing the competitive instincts of a community. Netflix [20] and X-Prize [21] were two early successes in online hosting of data challenges. Commercial initiatives such as Kaggle [22] and Innocentive [23] have hosted many successful online modeling competitions in astronomy, insurance, medicine, and other data-rich disciplines. The MAQC-II project [24] employed blinded evaluations and standardized datasets in the context of a large consortium-based research study to assess modeling factors related to prediction accuracy across 13 different phenotypic endpoints. Efforts such as CASP [25], DREAM [26], and CAFA [27] have created communities around key scientific challenges in structural biology, systems biology, and protein function prediction, respectively. In all cases it has been observed that the best crowd-sourced models usually outperform state-of-the-art off-the-shelf methods.
Despite their success in achieving models with improved performance, existing resources do not provide a general solution for hosting open-access crowd-sourced collaborative competitions due to two primary factors. First, most systems provide participants with a training dataset and require them to submit a vector of predictions for evaluation in the held-out dataset [20], [22], [24], [26], often requiring (only) the winning team to submit a description of their method and sometimes source code to verify reproducibility. While this achieves the goal of objectively assessing models, we believe it fails to achieve an equally important goal of developing a transparent community resource where participants work openly to collaboratively share and evolve models. We overcome this problem by developing a system where participants submit models as re-runnable source code by implementing a simple programmatic API consisting of a train and predict method. Second, some existing systems are designed primarily to leverage crowd-sourcing to develop models for a commercial partner [22], [23] who pays to run the competition and provides a prize to the developer of the best-performing model. Although we support this approach as a creative and powerful method for advancing commercial applications, such a system imposes limitations on the ability of participants to share models openly as well as intellectual property restrictions on the use of models. We overcome this problem by making all models available to the community through an open source license.
In this study, we formed a research group consisting of scientists from 5 institutions across the United States and conducted a collaborative competition to assess the accuracy of prognostic models of breast cancer survival. This research group, called the Federation, was set up as a mechanism for advancing collaborative research projects designed to demonstrate the benefit of team-oriented science. The rest of our group consisted of the organizers of the DREAM project, the Oslo team from the Norwegian Breast Cancer study, and leaders of the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC), who provided a novel dataset consisting of nearly 2,000 breast cancer samples with median 10-year follow-up, detailed clinical information, and genome-wide gene expression and copy number profiling data. In order to create an independent dataset for assessing model consistency, the Oslo team generated novel copy number data on an additional 102 samples (the MicMa cohort), which was combined with gene expression and clinical data for the same samples that was previously put in the public domain by the same research group [4], [28].
The initial study using the METABRIC data focused on unsupervised molecular sub-class discovery [29]. Although some of the reported sub-classes do correlate with survival, the goal of this initial work was not to build prognostic models. Indeed, the models developed in the current study provide more accurate survival predictions than those trained using molecular sub-classes reported in the original work. Therefore, the current study represents the first large-scale attempt to assess prognostic models based on a dataset of this scale and quality of clinical information.
The contributions of this work are two-fold. First, we conducted a detailed post-hoc analysis of all submitted models to determine model characteristics related to prognostic accuracy. Second, we report the development of a novel computational system for hosting community-based collaborative competitions, providing a generalizable framework for participants to build and evaluate transparent, re-runnable, and extensible models. Further, we suggest elements of study design, dataset characteristics, and evaluation criteria used to assess whether the results of a competition-style research study improve on standard approaches. We stress that the transparency enabled by making source code available and providing objective pre-defined scoring criteria allow researchers in future studies to verify reproducibility, improve on our findings, and assess their generalizability in future applications. Thus the results and computational system developed in this work serve as a pilot study for an open community-based competition on prognostic models of breast cancer survival. More generally, we believe this study will serve as the basis for additional competition-based research projects in the future, with the goal of promoting increased transparency and objectivity in genomics research (and other applications) and providing an open framework to collaboratively evolve complex models leading to patient benefit, beyond the sum of the individual efforts, by leveraging the wisdom of crowds.
We used the METABRIC dataset as the basis of evaluating prognostic models in this study. This dataset contains a total of nearly 2,000 breast cancer samples. 980 of these samples (excluding those with missing survival information) were available for the duration of the collaborative competition phase of this study. An additional 988 samples became available after we had concluded our evaluation in the initial dataset and, fortunately, served as a large additional dataset for assessing the consistency of our findings. For each sample, the dataset contains median 10 year follow-up, 16 clinical covariates (Table 1), and genome-wide gene expression and copy number profiling data, normalized as described in [29], resulting in 48,803 gene expression features and 31,685 copy number features summarized at the gene level (see Methods).
Initial analysis was performed to confirm that the data employed in the competition were consistent with previously published datasets and to identify potential confounding factors such as internal subclasses. Data-driven, unsupervised hierarchical clustering of gene expression levels revealed the heterogeneity of the data and suggested that multiple subclasses do exist (not shown) [29]. However, for the current analysis we decided to focus on the well established separation into basal, luminal, and HER2 positive subclasses, as previously defined [2], [30]. These subclasses are known to closely match clinical data in the following way: most triple-negative samples belong to the basal subclass; most ER positive samples belong to the luminal subclass; and most ER negative HER2 positive samples belong to the HER2 subclass. To ensure that this holds in the current dataset, the 50 genes that best separate the molecular subclasses in the Perou dataset [31] (PAM50) were used for hierarchical clustering of the METABRIC data and compared with a similar clustering of the Perou dataset (Figure 1A). The results of the supervised clustering reveal similar subclasses with similar gene expression signatures as those presented by Perou et al, and were also consistent with the clinical definitions as presented above. Finally, the 3 subclasses show a distinct separation in their Kaplan-Meier overall survival plots for the three subtypes defined by the clinical data, where the HER2 subclass has the worst prognosis, followed by the basal subclass, and the luminal subclass has the best prognosis, as expected (Figure 1B). This analysis shows that sub-classification based on ER (IHC), PR (gene expression), and HER2 (copy number) should capture the major confounding factors that may be introduced by the heterogeneity of the disease.
Multiple individual clinical features exhibit high correlation with survival for non-censored patients, and have well documented prognostic power (Table 1, Figure 1C), while others have little prognostic power (Figure 1D). To demonstrate that the competition data is consistent in this respect, a Cox proportional hazard model was fit to the overall survival (OS) of all patients using each one of the clinical covariates individually. As expected, the most predictive single clinical features are the tumor size, age at diagnosis, PR status, and presence of lymph node metastases (Table 1). To assess the redundancy of the clinical variables, an additional multivariable Cox proportional hazard model was fit to the overall survival (OS) of all patients using all clinical features. The remaining statistically significant covariates were patient age at diagnosis (the most predictive feature), followed by tumor size, presence of lymph node metastases, and whether the patient received hormone therapy.
Participants from our 5 research groups were provided data from 500 patient samples used to train prognostic models. These models were submitted as re-runnable source code and participants were provided real-time feedback in the form of a “leaderboard” based on the concordance index of predicted survival versus the observed survival in the 480 held-out samples.
Participants independently submitted 110 models to predict survival from the supplied clinical and molecular data (Table S1), showing a wide variability in their performance, which was expected since there were no constraints on the submissions. Post-hoc analysis of submitted models revealed 5 broad classes of modeling strategies based on if the model was trained using: only clinical features (C); only molecular features (M); molecular and clinical features (MC); molecular features selected using prior knowledge (MP); molecular features selected using prior knowledge combined with clinical features (MPC) (Table 2). The complete distribution of the performance of all the models, evaluated using concordance index, and classified into these categories is shown in Figure 2.
Analysis of the relative performance among model categories suggested interesting patterns related to criteria influencing model performance. The traditional method for predicting outcome is Cox regression on the clinical features [32]. This model, which used only clinical features, served as our baseline, and obtained a concordance index of 0.6347 on the validation set. Models trained on the clinical covariates using state-of-the-art machine learning methods (elastic net, lasso, random forest, boosting) achieved notable performance improvements over the baseline Cox regression model (Figure 2, category ‘C’).
Two submitted models were built by naively inputting all molecular features into machine learning algorithms (i.e. using all gene expression and CNA features and no clinical features). These models (our category ‘M’) both performed significantly worse than the baseline clinical model (median concordance index of 0.5906). Given that our training set contains over 80,000 molecular features and only 500 training samples, this result highlights the challenges related to overfitting due to the imbalance between the number of features and number of samples, also known as the curse of dimensionality [33], [34].
Models trained using molecular feature data combined with clinical data (category ‘MC’) outperformed the baseline clinical model in 10 out of 28 (36%) submissions, suggesting there is some difficulty in the naïve incorporation of molecular feature data compared to using only clinical information. In fact, the best MC model attributed lower weights to molecular compared to clinical features by rank-transforming all the features (molecular and clinical) and training an elastic net model, imposing a penalty only on the molecular features and not on the clinical ones, such that the clinical features are always included in the trained model. This model achieved a concordance index of 0.6593, slightly better than the best-performing clinical only model.
One of the most successful approaches to addressing the curse of dimensionality in genomics problems has been to utilize domain-specific prior knowledge to pre-select features more likely to be associated with the phenotype of interest [35]. Indeed, the majority of submitted models (66 of 110, 60%) utilized a strategy of pre-selecting features based on external prior knowledge. Interestingly, analysis of model submission dates indicates that participants first attempted naïve models incorporating all molecular features, and after achieving small performance improvements over clinical only models, evolved to incorporate prior information as the dominant modeling strategy in the later phase of the competition (Figure 2B). This observation is consistent with previous reports highlighting the importance of real-time feedback in motivating participants to build continuously improving models [36].
All models trained on only the molecular features (i.e. excluding the clinical features) and incorporating prior knowledge (MP category) performed worse than the baseline model, with the highest concordance index being 0.5947, further highlighting the difficultly in using molecular information alone to improve prognostic accuracy compared to clinical data.
Twenty-four models outperformed the baseline by combining clinical features with molecular features selected by prior knowledge (MPC category). The overall best-performing model attained a concordance index of 0.6707 by training a machine learning method (boosted regression) on a combination of: 1) clinical features; 2) expression levels of genes selected based on both data driven criteria and prior knowledge of their involvement in breast cancer (the MASP feature selection strategy, as described in Methods); 3) an aggregated “genomic instability” index calculated from the copy number data (see Methods).
The wide range of concordance index scores for models in the MPC category raises the question of whether the improved performance of the best MPC models are explained by the biological relevance of the selected features or simply by random fluctuations in model scores when testing many feature sets. Due to the uncontrolled experimental design inherent in accepting unconstrained model submissions, additional evaluations are needed to assess the impact of different modeling choices in a controlled experimental design. We describe the results of this experiment next.
We analyzed the modeling strategies utilized in the original “uncontrolled” model submission phase and designed a “controlled” experiment to assess the associations of different modeling choices with model performance. We determined that most models developed in the uncontrolled experiment could be described as the combination of a machine learning method with a feature selection strategy. We therefore tested models trained using combinations of a discrete set of machine learning methods crossed with feature selection strategies using the following experimental design:
This experiment design resulted in a total of 60 models based on combinations of modeling strategies from the uncontrolled experiment (Table S4), plus 20 models using ensemble strategies. This controlled experimental design allowed us to assess the effect of different modeling choices while holding other factors constant.
Following an approach suggested in the MAQC-II study [24], we designed negative and positive control experiments to infer bounds on model performance in prediction problems for which models should perform poorly and well, respectively. As a negative control, we randomly permuted the sample labels of the survival data, for both the training and test datasets, and computed the concordance index of each model trained and tested on the permuted data. To evaluate how the models would perform on a relatively easy prediction task, we conducted a positive control experiment in which all models were used to predict the ER status of the patients based on selected molecular features (excluding the ER expression measurement). We found that all negative control models scored within a relatively tight range of concordance indices centered around 0.5 (minimum: 0.468, maximum: 0.551), significantly lower than the lowest concordance index (0.575) of any model trained on the real data in this experiment. Conversely, all ER-prediction models scored highly (minimum: 0.79, maximum: 0.969), suggesting that the scores achieved by our survival models (maximum: 0.6707) are not due to a general limitation of the selected modeling strategies but rather the difficulty of modeling breast cancer survival.
Overall, we found that the predictive performance of the controlled experiment models (Figure 3A) was significantly dependent on the individual feature sets (P = 1.02e-09, F-test), and less dependent on the choice of the statistical learning algorithm (P = 0.23, F-test). All model categories using clinical covariates outperformed all model categories trained excluding clinical covariates, based on the average score across the 4 learning algorithms. The best-performing model category selected features based on marginal correlation with survival, further highlighting the difficulty in purely data-driven approaches, and the need to incorporate prior knowledge to overcome the curse of dimensionality.
The best-performing model used a random survival forest algorithm trained by combining the clinical covariates with a single additional aggregate feature, called the genomic instability index (GII), calculated as the proportion of amplified or deleted sites based on the copy number data. This result highlights the importance of evaluating models using a controlled experimental design, as the best-performing method in the uncontrolled experiment combined clinical variables with GII in addition to selected gene expression features (clinical variables plus only GII was not evaluated), and the controlled experiment pointed to isolating GII as the modeling insight associated with high prediction accuracy.
The random survival forest trained using clinical covariates and GII was significantly better than a random survival forest trained using clinical covariates alone (P = 2e-12 by paired Wilcoxon signed rank test based on 100 bootstrap samples with replacement from the test dataset). We also tested if inclusion of the GII feature improved model performance beyond a score that could be obtained by chance based on random selection of features. We trained 100 random survival forest models and 100 boosting models, each utilizing clinical information in addition to random selections of 50 molecular features (corresponding to the number of features used based on the MASP strategy, which achieved the highest score of all feature selection methods). The best-performing model from our competition (trained using clinical covariates and GII) achieved a higher score than each of these 100 models for both learning algorithms (P< = .01).
The use of the aggregate GII feature was based on previous reports demonstrating the association between GII and poor prognosis breast cancer subtypes like Luminal B, HER2+ and Basal-like tumors [37]. We found that HER2+ tumors had the strongest association with the GII score (P = 1.65e-12, t-test) which partly explains why it performs so well considering none of the patients were treated with compounds that target the HER2 pathway (e.g. Herceptin). Samples with high GII scores were also associated with high-grade tumors (P = 7.13e-13, t-test), further strengthening its credential as a good survival predictor. However, despite these strong associations, the genomic instability index provided an added value to the strength of predictions even as clinical covariates histologic grade and HER2 status are used in the models.
Boosting was the best-performing method on average. Elastic net and lasso exhibited stable performance across many feature sets. Random survival forests performed very well when trained on a small number of features based on clinical information and the genomic instability index. However, their performance decreased substantially with the inclusion of large molecular feature sets.
Ensemble methods trained by averaging predicted ranks across multiple methods systematically performed better than the average concordance index scores of the models contained in the ensemble, consistent with previously reported results [38]. Strikingly, an ensemble method aggregating all 60 models achieved a concordance index score of .654, significantly greater than the average of all model scores (.623) (Figure 3B). The ensemble performed better than the average model score for each of 100 resampled collections of 60 models each, using bootstrapping to sample with replacement from all 60 models (P< = .01). The ensemble model scored better than 52 of the 60 (87%) models that constituted the ensemble. We note that 2 of the algorithms (boosting and random forests) utilize ensemble learning strategies on their own. For both of the other 2 algorithms (lasso and elastic net) the method trained on an ensemble of the 15 feature sets scored higher than each of the 15 models trained on the individual feature sets (Figure 3B). Consistent with previous reports, the systematic outperformance of ensemble models compared to their constituent parts suggests that ensemble approaches effectively create a consensus that enhances the biologically meaningful signals captured by multiple modeling approaches. As previously suggested in the context of the DREAM project [38]–[41], our finding further reinforces the notion that crowd-sourced collaborative competitions are a powerful framework for developing robust predictive models by training an ensemble model aggregated across diverse strategies employed by participants.
In the first round of the competition, we did not restrict the number of models a participant could submit. This raises the possibility of model overfitting to the test set used to provide real-time feedback. We therefore used 2 additional datasets to evaluate the consistency of our findings. The first dataset, which we called METABRIC2, consisted of the 988 samples (excluding those with missing survival data) from the METABRIC cohort that were not used in either the training dataset or the test dataset used for real-time evaluation. The second dataset, called MicMa, consisted of 102 samples with gene expression, clinical covariates, and survival data available [4], [28] and copy number data presented in the current study (see Methods). We used the models from our controlled experiment, which were trained on the original 500 METABRIC samples, and evaluated the concordance index of the survival predictions of each model compared to observed survival in both METABRIC2 and MicMa.
The concordance index scores across models from the original evaluation were highly consistent in both METABRIC2 and MicMa. The 60 models evaluated in the controlled experiment (15 feature sets used in 4 learning algorithms) had Pearson correlations of .87 (P<1e-10) compared to METABRIC2 (Figure 4A) and .76 (P<1e-10) compared to MicMa (Figure 4C), although we note that p-values may be over-estimated due to smaller effective sample sizes due to non-independence of modeling strategies. Model performance was also strongly correlated for each different algorithm across the feature sets for both METABRIC2 (Figure 4B) and MicMa (Figure 4D).
Consistent with results from the original experiment, the top scoring model, based on average concordance index of the METABRIC2 and MicMa scores, was a random survival forest trained using clinical features in combination with the GII. The second best model corresponded to the best model from the uncontrolled experiment (3rd best model in the controlled experiment), and used clinical data in combination with GII and the MASP feature selection strategy, and was trained using a boosting algorithm. A random forest trained using only clinical data achieve the 3rd highest score. The top 39 models all incorporated clinical data.
As an additional comparison, we generated survival predictions based on published procedures used in the clinically approved MammaPrint [6] and Oncotype DX [7] assays. We note that these assays are designed specifically for early stage, invasive, lymph node negative breast cancers (in addition ER+ in the case of Oncotype DX) and use different scores calculated from gene expression data measured on distinct platforms. It is thus difficult to reproduce exactly the predictions provided by these assays or to perform a fair comparison to the present methods on a dataset that includes samples from the whole spectrum of breast tumors. The actual Oncotype DX score is calculated from RT-PCR measurements of the mRNA levels of 21 genes. Using z-score normalized gene expression values from METABRIC2 and MicMa datasets, together with their published weights, we recalculated Oncotype DX scores in an attempt to reproduce the actual scores as closely as possible. We then scored the resulting predictions against the two datasets and obtained concordance indices of 0.6064 for METABRIC2 and 0.5828 for MicMa, corresponding to the 81st ranked model based on average concordance index out of all 97 models tested, including ensemble models and Oncotype DX and MammaPrint feature sets incorporated in all learning algorithms (see Table S5). Similarly, the actual MammaPrint score is calculated based on microarray gene expression measurements, with each patient's score determined by the correlation of the expression of 70 specific genes to the average expression of these genes in patients with good prognosis (defined as those who have no distant metastases for more than five years, ER+ tumors, age less than 55 years old, tumor size less than 5 cm, and are lymph node negative). Because of limitations in the data, we were not able to compute this score in exactly the same manner as the original assay (we did not have the metastases free survival time, and some of the other clinical features were not present in the validation datasets). We estimated the average gene expression profile for the 70 MammaPrint genes based on all patients who lived longer than five years (with standardized gene expression data), then computed each patient's score as their correlation to this average good prognosis profile. We scored the predictions against the two validation datasets and observed concordance indices of 0.602 in METABRIC2 and 0.598 in MicMa, corresponding to the 78th ranked out of 97 models based on average concordance index.
We were able to significantly improve the scores associated with both MammaPrint and Oncotype DX by incorporating the gene expression features utilized by each assay as feature selection criteria in our prediction pipelines. We trained each of the 4 machine learning algorithms with clinical features in addition to gene lists from MammaPrint and Oncotype DX. The best-performing models would have achieved the 8th and 26th best scores, respectively, based on average concordance index in METABRIC2 and MicMa. We note that using the ensemble strategy of combining the 4 algorithms, the model trained using Mammaprint genes and clinical data performed better than clinical data alone, and achieved the 5th highest average model score, including the top score in METABRIC2, slightly (.005 concordance index difference) better than the random forest model using clinical data combined with GII, though only the 17st ranked score in MicMa. This result suggests that incorporating the gene expression features identified by these clinically implemented assays into the prediction pipeline described here may improve prediction accuracy compared to current analysis protocols.
An ensemble method, aggregating results across all learning algorithms and feature sets, performed better than 71 of the 76 models (93%) that constituted the ensemble, consistent with our finding that the ensemble strategy achieves performance among the top individual approaches. For the 19 feature selection strategies used in the METABRIC2 and MicMa evaluations, an ensemble model combining the results of the 4 learning algorithms performed better than the average of the 4 learning algorithms in 36 out of 38 cases (95%). Also consistent with our previous result, for both algorithms that did not use ensemble strategies themselves (elastic net and lasso), an ensemble model aggregating results across the 19 feature sets performed better than each of the individual 19 feature sets for both METABRIC2 and MicMa. Taken together, the independent evaluations in 2 additional datasets are consistent with the conclusions drawn from the original real-time feedback phase of the completion, regarding improvements gained from ensemble strategies and the relative performance of models.
“Precision Medicine”, as defined by the Institute of Medicine Report last year, proposes a world where medical decisions will be guided by molecular markers that ensure therapies are tailored to the patients who receive them [42]. Moving towards this futuristic vision of cancer medicine requires systematic approaches that will help ensure that predictive models of cancer phenotypes are both clinically meaningful and robust to technical and biological sources of variation.
Despite isolated successful developments of molecular diagnostic and personalized medicine applications, such approaches have not translated to routine adoption in standard-of-care protocols. Even in applications where successful molecular tests have been developed, such as breast cancer prognosis [5], [6], a plethora of research studies have claimed to develop models with improved predictive performance. Much of this failure has been attributed to “difficulties in reproducibility, expense, standardization and proof of significance beyond current protocols” [43]. The propensity of researchers to over-report the performance of their own approaches has been deemed the “self-assessment trap” [19].
We propose community-based collaborative competitions [43]–[49] as a general framework to develop and evaluate predictive models of cancer phenotypes from high-throughput molecular profiling data. This approach overcomes limitations associated with the design of typical research studies, which may conflate self-assessment with methodology development or, even more problematic, with data generation. Thus competition-style research may promote transparency and objective assessment of methodologies, promoting the emergence of community standards of methodologies most likely to yield translational clinical benefit.
The primary challenge of any competition framework is to ensure that mechanisms are in place to prevent overfitting and fairly assess model performance, since performance is only meaningful if models are ranked based on their ability to capture some underlying signal in the data. For example, such an approach requires datasets affording sufficient sample sizes and statistical power to make meaningful comparisons of many models across multiple training and testing data subsets. We propose several strategies for assessing if the results obtained from a collaborative competition are likely to generalize to future applications and improve on state-of-the art methodologies that would be employed by an expert analyst.
First, baseline methods should be provided as examples of approaches an experienced analyst may apply to the problem. In our study, we employed a number of such methods for comparison, including methodologies used in clinical diagnostic tests and multiple state-of-the-art machine learning methods trained using only clinical covariates.
Second, performance of models should be evaluated in multiple rounds of independent validation. In this study, we employed a multi-phase strategy suggested by previous researchers [50] in which a portion of the dataset is held back to provide real-time feedback to participants on model performance and another portion of the dataset is held back and used to score the performance of all models, such that participants cannot overfit their models to the test set. If possible, we recommend an additional round of validation using a dataset different from the one used in previous rounds, in order to test against the possibility that good performance is due to modeling confounding variables in the original dataset. This experimental design provides 3 independent rounds of model performance assessment, and consistent results across these multiple evaluations provides strong evidence that performance of the best approaches discovered in this experimental design are likely to generalize in additional datasets.
Finally, statistical permutation tests can provide useful safeguards against the possibility that improved model performance is attributable to random fluctuations based on evaluation of many models. Such tests should be designed carefully based on the appropriate null hypothesis. A useful, though often insufficient, test is to utilize a negative control null model, for example by permuting the sample labels of the response variable. We suggest that additional tests may be employed as post-hoc procedures designed specifically to provide falsifiable hypotheses that may provide alternative explanations of model performance. For example, in this study we assessed the performance of many models trained using the same learning algorithm (random survival forest) and the same clinical features as used in the top scoring model, but using random selections of molecular features instead of the GII feature. This test was designed to falsify the hypothesis that model performance is within the range of likely values based on random selection of features, as has been a criticism of previously reported models [18].
We suggest that the guidelines listed above provide a useful framework in reporting the results of a collaborate competition, and may even be considered necessary criteria to establish the likelihood that findings will generalize to future applications. As with most research studies, a single competition cannot comprehensively assess the full extent to which findings may generalize to all potentially related future applications. Accordingly, we suggest that a collaborative competition should indeed report the best forming model, provided it meets the criteria listed above, but need not focus on declaring a single methodology as conclusively better than all others. By analogy to athletic competitions such as an Olympic track race, a gold medal is given to the runner with the fastest time, even if by a fraction of a second. Judgments of superior athletes emerge through integrating multiple such data points across many races against different opponents, distances, weather conditions, etc., and active debate among the community. A research study framed as a collaborative competition may facilitate the transparency, reproducibility, and objective evaluation criteria that provide the framework on which future studies may build and iterate towards increasingly refined assessments through a continuous community-based effort.
Within several months we developed and evaluated several hundred modeling approaches. Our research group consisted of experienced analysts trained as both data scientists and clinicians, resulting in models representing state-of-the art approaches employed in both machine learning and clinical cancer research (Table 3). By conducting detailed post-hoc analysis of approaches developed by this group, we were able to design a controlled experiment to isolate the performance improvements attributable to different strategies, and to potentially combine aspects of different approaches into a new method with improved performance.
The design of our controlled experiment builds off pioneering work by the MAQC-II consortium, which compiled 6 microarray datasets from the public domain and assessed modeling factors related to the ability to predict 13 different phenotypic endpoints. MAQC-II classified each model based on several factors (type of algorithm, normalization procedure, etc), allowing analysis of the effect of each modeling factor on performance. Our controlled experiment follows this general strategy, and extends it in several ways.
First, MAQC-II, and most competition-base studies [20], [22], [26], accept submissions in the form of prediction vectors. We developed a computational system that accepts models as re-runnable source code implementing a simple train and predict API. Source code for all submitted models are stored in the Synapse compute system [51] and are freely available to the community. Thus researchers may reproduce reported results, verify fair play and lack of cheating, learn from the best-performing models, reuse submitted models in related applications (e.g. building prognostic models in other datasets), build ensemble models by combining results of submitted models, and combine and extend innovative ideas to develop novel approaches. Moreover, storing models as re-runnable source code is important in assessing the generalizability and robustness of models, as we are able to re-train models using different splits or subsets of the data to evaluate robustness, and we (or any researcher) can evaluate generalizability by assessing the accuracy of a model's predictions in an independent dataset, such as existing related studies [5] or emerging clinical trial data [52]. We believe this software system will serve as a general resource that is extended and re-used in many future competition-based studies.
Second, MAQC-II conducted analysis across multiple phenotypic endpoints, which allowed models to be re-evaluated in the context of many prediction problems. However, this design required models to be standardized across all prediction problems and did not allow domain-specific insights to be assessed for each prediction problem. By contrast, our study focused on the single biomedical problem of breast cancer prognosis, and allowed clinical research specialists to incorporate expert knowledge into modeling approaches. In fact, we observed that feature selection strategies based on prior domain-specific knowledge had a greater effect on model performance than the choice of learning algorithm, and learning algorithms that did not incorporate prior knowledge were unable to overcome challenges with incorporating high-dimensional feature data. In contrast to previous reports that have emphasized abstracting away domain-specific aspects of a competition in order to attract a broader set of analysis [50], in real-word problems, we emphasize the benefit of allowing researchers to apply domain-specific expertise and objectively test the performance of such approaches against those of analysts employing a different toolbox of approaches.
Finally, whereas MAQC-II employed training and testing splits of datasets for model evaluation, our study provides an additional level of evaluation in a separate, independent dataset generated on a different cohort and using different gene expression and copy number profiling technology. Consistent with findings reported by MAQC-II, our study demonstrates strong consistency of model performance across independent evaluations and provides an important additional test of model generalizability that more closely simulates real-world clinical applications, in which data is generated separately from the data used to construct models. More generally, whereas MAQC-II evaluated multiple prediction problems in numerous datasets with gene expression data and samples numbers from 70 to 340, our study went deeper into a evaluating a single prediction problem, utilizing copy number and clinical information in addition to gene expression, and with a dataset of 2,000 samples in addition to an independently-generated dataset with 102 samples.
The model achieving top performance in both the initial evaluation phase and the evaluation in additional datasets combined a state-of-the-art machine learning approach (random survival forest) with a clinically motivated feature selection strategy that used all clinical features together with an aggregate genomic instability index. Interestingly, this specific model was not tested in the uncontrolled phase, and was the result of the attempt to isolate and combine aspects of different modeling approaches in a controlled experiment. The genomic instability index measure may serve as a proxy for the degree to which DNA damage repair pathways (including, for instance, housekeeping genes like p53 and RB) have become dysregulated [37].
Beyond the specifics of the top performing models, we believe the more significant contribution of this work is as a building block, providing a set of baseline findings, computational infrastructure, and proposed research methodologies used to assess breast cancer prognosis models, and extending in the future to additional phenotype prediction problems. Towards this end, we have recently extended this work into an open collaborative competition through which any researcher can freely register and evaluate the performance of submitted models against all others submitted throughout the competition. Though this expanded breast cancer competition, and future phenotype prediction competitions to be hosted as extensions of the current work, we invite researchers to improve, refute, and extend our findings and research methodologies to accelerate the long arc of cumulative progress made by the community through a more transparent and objectively assessed process.
Our competition was designed to assess the accuracy of predicting patient survival (using the overall survival metric, median 10 year follow-up) based on feature data measured in the METABRIC cohort of 980 patients, including gene expression and copy number profiles and 16 clinical covariates (Table 1). Participants were given a training dataset consisting of data from 500 samples, and data from the remaining 480 were hidden from participants and used as a validation dataset to evaluate submitted models.
We developed the computational infrastructure to support the competition within the open-source Sage Synapse software platform. Detailed documentation is available on the public competition website: https://sagebionetworks.jira.com/wiki/display/BCC/Home. The system is designed to generalize to support additional community-based competitions and consists of the following components (Figure 5):
All models are available with downloadable source code using the Synapse IDs displayed in Table S1 and Table S4. An automated script continuously monitored for new submissions, which were sent to worker nodes in a computational cluster for scoring. Each worker node ran an evaluation script, which called the submitted model's customPredict method with arguments corresponding to the gene expression, copy number, and clinical covariate values in the held-out validation dataset. This function returns a vector of predicted survival times in the validation dataset, which were used to calculate the concordance index as a measure of accuracy compared to the measured survival times for the same samples. Concordance index scores were shown in a real-time leaderboard, similar to the leaderboards displaying the models scores shown in Table S1 and Table S4.
Concordance index (c-index) is the standard metric for evaluation of survival models [53]. The concordance index ranges from 0 in the case of perfect anti-correlation between the rank of predictions and the rank of actual survival time through 0.5 in the case of predictions uncorrelated with survival time to 1 in the case of exact agreement with rank of actual survival time. We implemented a method to compute the exact value of the concordance index by exhaustively sampling all pairwise combinations of samples rather than the usual method of stochastically sampling pairwise samples. This method overcomes the stochastic sampling used in standard packages for concordance index calculation and provides a deterministic, exact statistic used to compare models.
Data on the original 980 samples were obtained for this study in early January, 2012. Study design and computational infrastructure were developed from then until March 14th, at which point participants were given access to the 500 training samples and given 1 month to develop models in the “uncontrolled experiment” phase. During this time, participants were given real-time feedback on model performance evaluated against the held-out test set of 480 samples. After this 1-month model development phase, all models were frozen and inspected by the group to conduct post-hoc model evaluation and identify modeling strategies used to design the controlled evaluation. All models in the controlled evaluation were re-trained on the 500 training samples and re-evaluated on the 480 test samples. After all evaluation was completed based on the original 980 samples, the METABRIC2 and MicMa datasets became available, and were used to perform additional evaluations of all models, which was conducted between January 2013–March 2013. For the new evaluation, all data was renormalized to the gene level, as described below, in order to allow comparison of models across datasets performed on different platforms. Models were retrained using the re-normalized data for the same 500 samples in the original training set.
All model source code is available in the subfolders of Synapse ID syn160764, and specific Synapse IDs for each model are listed in Table S1 and Table S4. Data stored in Synapse may be accessed using the Synapse R client (https://sagebionetworks.jira.com/wiki/display/SYNR/Home) or by clicking the download icon on the web page corresponding to each model, allowing the user to download a Zip archive containing the source files contained in the submission.
The METABRIC dataset used in the competition contains gene expression data from the Illumina HT 12v3 platform and copy number data derived from experiments performed on the Affymetrix SNP 6.0 platform. In the initial round of analysis, the first 980 samples data was normalized as described in [29], corresponding to the data available in the European Genome-Phenome Archive (http://www.ebi.ac.uk/ega), accession number EGAS00000000083. Copy number data was summarized to the gene level by calculating the mean value of the segmented regions overlapping a gene. Data for use in our study are available in the Synapse software system (synapse.sagebase.org) within the folder with accession number syn160764 (https://synapse.prod.sagebase.org/#Synapse:syn160764), subject to terms of use agreements described below. Data may be loaded directly in R using the Synapse R client or downloaded from the Synapse web site.
Patients treated for localized breast cancer from 1995 to 1998 at Oslo University Hospital were included in the MicMa cohort, and 123 of these had available fresh frozen tumor material [4], [28]. Gene expression data for 115 cases obtained from an Agilent whole human genome 4×44 K one color oligo array was available (GSE19783) [54]. Novel SNP-CGH data from 102 of the MicMa samples were obtained using the Illumina Human 660k Quad BeadChips according to standard protocol. Normalized LogR values summarized to gene level were made available and are accessible in Synapse (syn1588686).
All data used for the METABRIC2 and MicMa analyses are available as subfolders of Synapse ID syn1588445. For comparison of METABRIC2 and MicMa, we standardized all clinical variables, copy number, and gene expression data across both datasets. Clinical variables were filtered out that were not available in both datasets. Data on clinical variables used in this comparison are available in Synapse.
All gene expression datasets were normalized according the supervised normalization of microarrays (snm) framework and Bioconductor package [55], [56]. Following this framework we devised models for each dataset that express the raw data as functions of biological and adjustment variables. The models were built and implemented through an iterative process designed to learn the identity of important variables. Once these variables were identified we used the snm R package to remove the effects of the adjustment variables while controlling for the effects of the biological variables of interest.
SNP6.0 copy number data was also normalized using the snm framework, and summarization of probes to genes was done as follows. First, probes were mapped to genes using information obtained from the pd.genomewidesnp.6 Bioconductor package [57]. For genes measured by two probes we define the gene-level values as an unweighted average of the probes' data. For genes measured by a single probe we define the gene-level values as the data for the corresponding probe. For those measured by more than 2 probes we devised an approach that weights probes based upon their similarity to the first eigengene. This is accomplished by taking a singular value decomposition of the probe-level data for each gene. The percent variance explained by the first eigengene is then calculated for each probe. The summarized values for each gene are then defined as the weighted mean with the weights corresponding to the percent variance explained.
For Illumina 660k data we processed the raw files using the crlmm bioconductor R package [58]. The output of this method produces copy number estimates for more than 600k probes. Next, we summarized probes to Entrez gene ids using a mapping file obtained from the Illumina web site. For genes measured by more than two probes we selected the probe with the largest variance.
Feature selection strategies used in the controlled experiment (identified through post-hoc analysis of the uncontrolled experiment) are described briefly in Table 3. Specific genes used in each category are available within Synapse ID syn1643406 and can be downloaded as R binaries via the Synapse web client or directly loaded in R using the Synapse R client. Most feature selection strategies are sufficiently described in Table 3, and we provide additional details on 2 methods below.
The MASP (Marginal Association with Subsampling and Prior Knowledge) algorithm employs the following procedure: all genes were first scored for association with survival (using Cox regression) in chunks of 50 randomly selected gene expression samples. This process was repeated 100 times which resulted in an overall survival association score where is the p-value associated with the Cox regression on the expression of gene i in sample set j. All genes were sorted in descending order by their survival association score and the top 50 oncogenes and transcription factors were kept. A list of human transcription factors was obtained from [59] and a list of oncogenes was compiled by searching for relevant keywords against the Entrez gene database.
GII is a measure of the proportion of amplified or deleted genomic loci, calculated from the copy number data. Copy number values are presented as segmented log-ratios with respect to normal controls. Amplifications and deletions are thus counted when or and devided by the total number of loci .
The data used in this study were collected and analyzed under approval of an IRB [29]. The MicMa study was approved by the Norwegian Regional Committee for medical research ethics, Health region II (reference number S-97103). All patients have given written consent for the use of material to research purposes.
|
10.1371/journal.pgen.1008251 | Recurrent gene co-amplification on Drosophila X and Y chromosomes | Y chromosomes often contain amplified genes which can increase dosage of male fertility genes and counteract degeneration via gene conversion. Here we identify genes with increased copy number on both X and Y chromosomes in various species of Drosophila, a pattern that has previously been associated with sex chromosome drive involving the Slx and Sly gene families in mice. We show that recurrent X/Y co-amplification appears to be an important evolutionary force that has shaped gene content evolution of sex chromosomes in Drosophila. We demonstrate that convergent acquisition and amplification of testis expressed gene families are common on Drosophila sex chromosomes, and especially on recently formed ones, and we carefully characterize one putative novel X/Y co-amplification system. We find that co-amplification of the S-Lap1/GAPsec gene pair on both the X and the Y chromosome occurred independently several times in members of the D. obscura group, where this normally autosomal gene pair is sex-linked due to a sex chromosome—autosome fusion. We explore several evolutionary scenarios that would explain this pattern of co-amplification. Investigation of gene expression and short RNA profiles at the S-Lap1/GAPsec system suggest that, like Slx/Sly in mice, these genes may be remnants of a cryptic sex chromosome drive system, however additional transgenic experiments will be necessary to validate this model. Regardless of whether sex chromosome drive is responsible for this co-amplification, our findings suggest that recurrent gene duplications between X and Y sex chromosomes could have a widespread effect on genomic and evolutionary patterns, including the epigenetic regulation of sex chromosomes, the distribution of sex-biased genes, and the evolution of hybrid sterility.
| Sex chromosomes are hot spots for genetic conflict, and selfish genetic elements that increase their transmission are prone to originate on sex chromosomes. Previous work has shown that genes that co-amplify on both the X and Y chromosome are involved in sex chromosome drive in mice. Here, we use bioinformatic approaches to identify co-amplified genes in 26 Drosophila species, and we characterize a novel co-amplified gene family in D. pseudoobscura using functional genomic approaches. Our comparative genomic analysis suggests that co-amplification of genes on sex chromosomes is rampant, and we detect co-amplified X/Y genes in dozens of fly species investigated, especially those with young sex chromosomes. We find that co-amplified genes are often derived from well-characterized meiosis genes that are necessary for proper segregation, and are highly expressed in testis. Finally, we show that co-amplified genes often produce antisense transcripts and are targeted by small RNAs. Functional enrichment and expression patterns are consistent with a model where co-amplification of these genes is driven by intragenomic conflicts over transmission of the X and Y chromosome, and that RNAi mechanisms are utilized to launch evolutionary responses to counter sex ratio distortion. This would imply a novel role for the RNAi pathway to defend the genome against selfish elements that try to manipulate fair transmission.
| Selfish genetic elements whose evolutionary trajectories are in conflict with those of their host were first described almost 100 years ago [1]. However, only in recent decades has it become apparent that the antagonistic coevolution resulting from genetic conflict has shaped genome content and structure across the tree of life, from bacteria to plants and animals [2]. Antagonistic coevolution can occur between organisms, as in the evolutionary “arms race” experienced between pathogens and their hosts, or within genomes among genetic elements with different inheritance patterns (such as mobile elements, or X and Y chromosomes [3]). For instance, selfish genetic elements can manipulate meiosis (or gametogenesis) so that they are transmitted to more than 50% of offspring (so called “segregation distorters” or “meiotic drivers”)[4]. These processes can leave behind a variety of distinct genetic signatures. For example, genes involved in pathogen virulence and host resistance consistently show elevated levels of amino acid substitutions (i.e. dN/dS) whereas genes involved in intra-genomic conflict, such as segregation distorters and their suppressors, often have high rates of lineage-specific duplications and gene amplifications [5,6]. A well-studied cryptic sex chromosome drive system in mouse involves the convergent acquisition and amplification of the same gene families (Slx/Sly) on both the X and Y chromosome, and careful experimentation has shown that the co-amplified genes are in a co-evolutionary battle over sex chromosome transmission, whereby the X-and Y-linked copies of a gene family directly compete with each other [7,8]. Sly knockdowns show female-biased sex ratios, while Slx deficiency causes a sex ratio distortion towards males. A similar mechanisms of cryptic segregation distortion has been implicated in the Stellate/ Suppressor of Stellate (Ste/Su(Ste)) system in D. melanogaster, where the expression of the X-linked gene Ste leads to the production of defective sperm, and Su(Ste), which is a multi-gene copy of Ste that moved to the Y-chromosome, silences Ste [9,10].
Although rapid rates of amino acid substitution, gene duplication, and gene amplification are all characteristics of evolutionary conflict, these processes are also associated with strong selection in the absence of conflict, or in some cases, even neutral evolution. This makes identification of conflict from genomic data alone difficult [11]. For example, recent studies have shown that the Y chromosomes of many organisms contain testes-specific genes that have amplified in copy number [12–14]. Some of these Y-linked gene families, such as those in mice, have been shown to be involved in sex chromosome drive, whereas for other gene families, the extra copies may either act to increase gene dosage or prevent degeneration by providing a substrate for non-allelic gene conversion [15]. Alternatively, the extra copies may be neutral or even slightly deleterious, yet they remain on the Y due to the reduced efficiency of selection on this non-recombining chromosome [16]. One key signature that appears to be unique to Y-amplified genes involved in sex chromosome drive is that their X-linked homologs have duplicated as well. This pattern is consistent with antagonistic co-evolution resulting from repeated bouts of sex ratio distortion and suppression (see Discussion).
Here, we use bioinformatics and functional genomic analyses to assess the prevalence of sex chromosome gene amplification across Drosophila species. Consistent with a role for Y-amplified genes unrelated to genetic conflict (see Discussion), we find hundreds of genes that appear to be present in multiple copies on the Y chromosomes of many Drosophila species. However, we also find a second category of Y-amplified genes whose X homolog has been duplicated as well. We show that species with young sex chromosomes have repeatedly evolved genes that have co-amplified on the X and the Y and show functions and expression patterns that are consistent with genetic conflict. We explore a variety of evolutionary scenarios that could give rise to this pattern of X-Y co-amplification based on detailed investigation of the S-Lap and GAPsec genes that have been independently co-amplified on the X and Y chromosomes of multiple species in the obscura group. We find that gene expression levels and small RNA production from these co-amplified genes are most consistent with a cryptic sex ratio drive system, however additional experiments are necessary to test these claims. We develop a model for how such a system could have evolved and present evidence suggesting that the same genes appear to have become involved in a meiotic conflict independently among multiple species of this group.
To analyze gene content evolution and identify amplified X- and Y-linked genes, we sequenced both male and female genomic DNA in 26 Drosophila species from across the Drosophila phylogeny (S1 Table). Roughly half of the species considered (11 out of 26) harbor the typical sex chromosome complement of Drosophila (that is, a single pair of ancient sex chromosomes, shared by all members of Drosophila). In addition to the ancestral pair of sex chromosomes, the other 15 species have a younger pair of “neo-sex” chromosomes, which formed when an autosome became fused to one or both of the ancient X and Y chromosomes (S1 Fig). These younger “neo-sex” chromosomes are at various stages of evolving the typical properties of ancestral sex chromosomes, with neo-Y chromosomes losing their original genes and acquiring a genetically inert heterochromatic appearance, and neo-X chromosomes acquiring their unique gene content and sex-specific expression patterns [16,17]. We identified putative Y-amplified genes based on male and female gene coverage without relying on a genome assembly (S2 Fig, see Methods) and validated our approach using a high-quality genome assembly from D. miranda (S3 Fig). Using this approach, we identify (depending on our cutoffs) 100s of genes that have multiple copies on the Y across the 26 species investigated (S2 and S3 Tables, S4 Fig). Genes might amplify on the Y for a variety of reasons, but co-amplification of testis genes may be a defining feature of genes evolving under genetic conflict (see Discussion). Here we define co-amplified genes as being amplified on the Y, based on male and female gene coverage, and as having at least two copies on the X chromosome in our genome assemblies.
Among these multi-copy gene families on the Y, we found 35 amplified Y-linked genes with co-amplified X homologs in 10 species (Fig 1, S4 Table). We infer that the copy number of these co-amplified X/Y genes ranges from 8 copies on the Y up to 297 Y-linked copies (for an uncharacterized testis gene in D. melanogaster that amplified on the Y of D. robusta), with a mean copy number of 58 (S5 Table). We detect between 2–4 X-linked copies in our assemblies for these co-amplified X/Y genes (S5 Table). The number of assembled X copies is likely an underestimate since recent gene duplicates are typically collapsed in assemblies derived from short read sequencing data, but investigations of high-quality genome assemblies derived from long-read technologies confirm that co-amplified genes have considerably fewer copies on the X than the Y chromosome (S6 Table; see also Discussion).
We next sought to investigate the putative functions of these co-amplified genes. We found that many are expressed in reproductive tissues in D. melanogaster (S4 Table). Of the candidate genes that we find, 76% are expressed in the testes in D. melanogaster (versus 56% genome-wide, FlyAtlas data) which is significantly more than expected by chance (one-sided Fisher’s Exact Test P = 0.011). Several of the genes have meiosis-related functions (Fig 1, S4 Table). For example, we identify genes that are associated with spindle assembly involved in male meiosis (fest), chromosome segregation (mars), or male meiosis cytokinesis (scra), amongst others (Fig 1). Indeed, GO enrichment analysis reveals the following terms to be enriched among co-amplified X/Y genes: sperm chromatin condensation, spindle organization, cell cycle process, and mitotic spindle organization (S5A Fig, S7 Table; note that the nominal P-values are significant but not after correcting for multiple hypothesis testing). Genes only amplified on the Y chromosome, on the other hand, show GO enrichment for different categories of metabolic processes, translation, and biosynthetic processes (S5B Fig, S7 Table).
Amplified Y genes were detected in each species investigated (S2 Table). Interestingly, however, co-amplification of genes is much more common in species with recently added neo-sex chromosomes: of the 10 species where we found co-amplified genes, nine harbor neo-sex chromosomes (Fig 1), and in the vast majority of cases the amplified genes were ancestrally present on the chromosome that formed the neo-sex chromosomes (Fig 1).
We decided to more carefully characterize two co-amplified genes in D. pseudoobscura, a species with a high quality PacBio-based genome assembly. D. pseudoobscura currently lacks an assembled Y chromosome, but we inferred Y-linkage of contigs based on male and female read coverage using Illumina data (see Methods). We identified two adjacent genes that exist in multiple copies on the X and Y chromosome of D. pseudoobscura: S-Lap1 (Dpse\GA19547) and GAPsec (Dpse\GA28668). S-Lap1 is a member of a leucyl aminopeptidase gene family that encodes the major protein constituents of Drosophila sperm [18], while GAPsec is a GTPase activating protein. This situation is reminiscent of the Segregation distorter meiotic drive system in D. melanogaster, where the distorter is a truncated tandem duplication of RanGAP, which is also a GTPase activator [19]. Both S-Lap1 and GAPsec show partial tandem duplications on the X (Fig 2A), and we detect roughly 100 (partial and full-length) copies of both S-Lap1 and GAPsec on the Y chromosome (the Y-linked contigs contain 127 copies of S-Lap1 and 91 copies of GAPsec; Figs 2B and 3). S-Lap1 and S-Lap2 are present in all Drosophila species investigated (Fig 2A), and probably originated in an ancestor of Drosophila; phylogenetic clustering of S-Lap1 and S-Lap2 in certain species groups (Fig 2C) probably resulted from gene conversion homogenizing gene duplicates within a clade [20]. In most species of the Drosophila clade, S-Lap1 and S-Lap2 are similar in size; in D. pseudoobscura and its sister D. persimilis, however, S-Lap2 has acquired a large deletion, removing more than half of the 3’ end of the gene (Fig 2A and 2C; S6A Fig). The partial duplication of GAPsec, on the other hand, is only found in D. pseudoobscura and its close relative D. persimilis (Fig 2A and 2C; S6B Fig). S-Lap1 and GAPsec probably dispersed onto the Y chromosome simultaneously, as there are multiple locations on the Y that preserve their X orientation (Fig 2B); note however, that the amplified copies on the Y do not include the tandemly duplicated copies. For both S-Lap1 and GAPsec, the X- and Y-linked copies are highly expressed in testis of D. pseudoobscura (S4 and S8 Tables, Fig 4A).
We used stranded RNA-seq and small RNA profiles from wildtype D. pseudoobscura testes, to obtain insights into the evolutionary mechanism responsible for the co-amplification of S-Lap1 and GAPsec. Interestingly, we detect both sense and antisense transcripts and small RNAs derived from S-Lap1 (Fig 4A, S8 Table; see S7 Fig for the size distribution of small RNAs mapping and S8 Fig for cross-mapping of RNA-seq and small RNA reads between different gene copies). In particular, stranded RNA-seq data reveal that the X-linked copy of S-Lap1 duplicate produces both sense and anti-sense transcripts, resulting in the production of small RNAs (see Fig 4A, S8 Table). Close inspection of this genomic region in D. pseudoobscura shows that the duplicated GAPsec gene is directly adjacent to where the S-Lap1 duplicate antisense transcript begins (Fig 4A). Intriguingly, this segment scores highly as a potential promoter sequence when using the Berkeley Drosophila Genome Project (BDGP) neural network promoter prediction algorithm [21] (score = 0.89, highest possible score = 1). Thus, this suggests that the partial duplication of GAPsec provided a promoter-like sequence in D. pseudoobscura for antisense transcription of S-Lap1 duplicate. Note that this putative promoter sequence is not part of the Y copies of S-Lap1 (which lack the GAPsec duplicate), and we detect virtually no antisense transcripts that originate from the Y-amplified copies of S-Lap1 (S8 Table). CAGE-seq data support that the GAPsec duplicate generated a new TSS resulting in antisense transcription of S-Lap1 duplicate (Fig 4A).
How unusual is antisense RNA expression, and the production of small RNAs for testis genes? To see if these features of S-Lap1 are commonly observed for other genes in D. pseudoobscura testis, we used our RNA-seq and small RNA data to identify additional genes that are expressed in testis (rpkm> = 2), show antisense expression (at least 75% of sense expression), and the production of small RNAs (rpkm> = 100). In addition to the S-Lap1 gene, this screen revealed 6 additional genes that produced antisense transcripts and small RNAs in the testes of D. pseudoobscura (S9 Fig, S9 Table). Interestingly, all of the identified genes are members of gene families (i.e. we detect at least 2 gene copies in the D. pseudoobscura assembly), and for 5 out of 6 genes, at least one copy is located on one of the sex chromosomes, and another copy is found on the other sex chromosome, or an autosome.
In most Drosophila species, S-Lap1 and GAPsec are located on an autosome (chromosome 3L in D. melanogaster). In the D. pseudoobscura and affinis group, however, this chromosome arm fused with the sex chromosomes about 15MY ago, causing S-Lap1 and GAPsec to become sex-linked. Intriguingly, patterns of molecular evolution at S-Lap1 and GAPsec suggest that they may have independently co-amplified in several members of the pseudoobscura species group. We used high-quality PacBio genome assemblies for two additional members of that species group [22], D. miranda, which diverged form D. pseudoobscura about 2–4 MY ago, and D. athabasca, which diverged 10–15 MY ago [23]. While our Illumina sequencing-based approach failed to detect co-amplified X and Y genes in these species (they have similar M/F coverage), examination of the assembled PacBio genomes revealed that both gene pairs independently amplified on the sex chromosomes of both D. miranda and D. athabasca (see Fig 5). We identify tandem duplications of the entire genomic region containing a total of 11 copies of S-Lap1 and 6 copies of GAPsec on chromosome XR in D. miranda, and these two genes have amplified 5 and 4 times, respectively, on the neo-Y chromosome of D. miranda (Fig 5A). Both the nature of the duplication event and patterns of sequence evolution suggest that co-amplification of S-Lap1 and GAPsec occurred independently in D. miranda. Here, the XR copies arose from individual duplications of these two genes followed by three tandem duplications of the entire genomic region encompassing S-Lap1 and GAPsec, producing a total of 11 copies of S-Lap1 and 6 copies of GAPsec. All six of the X-linked copies of GAPsec are highly similar to each other (>99% identical), and more similar to their Y-linked paralogs than they are to D. pseudoobscura (Fig 5C). Also, S-Lap1 and GAPsec appear to have moved only to a single location on the neo-Y of D. miranda, instead of being dispersed all across the Y, as in D. pseudoobscura (Fig 5A). Patterns of gene expression and short RNA production in D. miranda mimic that of D. pseudoobscura, with SLap-1 (and GAPsec) transcripts being produced from both strands, and small RNAs are generated across that genomic region (Fig 4B). The mechanism of antisense production appears to differ from D. pseudoobscura (Fig 4). In particular, transcriptional read-through at both S-Lap1 and GAPsec appear to generate anti-sense transcripts of both genes. Close inspection of this genomic region in D. miranda reveals sequence differences between the X and Y copies that may account for antisense production at X-linked gene copies. In particular, we detect a polyadenylation signal (AATAAA) for GAPsec that is present in most (3 of the 4) Y copies, and in the homologous copy on the X in D. pseudoobscura, but which is missing in the D. miranda X-linked copies of GAPsec. This mutational event could account for the creation of read-through transcripts on the X of D. miranda, leading to production of antisense transcripts for S-Lap1 and initiation of RNAi, analogous to the model proposed for D. pseudoobscura.
Likewise, we infer independent amplifications of S-Lap1 and GAPsec on the X chromosome of D. athabasca, another member of the pseudoobscura group for which we have generated a high-quality female genome assembly (Fig 5B). We detect 6 copies of GAPsec and 5 copies of S-Lap1 on the X chromosome of D. athabasca, and genomic coverage analysis suggests a similar number of copies on the Y chromosome (i.e. males show similar coverage of S-Lap1 and GAPsec as females, suggesting similar copy numbers on the X and Y). This suggests that S-Lap1 and GAPsec are independently involved in co-amplification in species where this locus has become sex-linked.
Our comparative analysis shows that co-amplification of genes on sex chromosomes is common in Drosophila. Note, however, that our method for identifying co-amplified X/Y genes is conservative, and we might greatly underestimate the true magnitude of co-amplification. On one hand, our approach for detecting amplified Y-genes requires them to have much higher coverage in male than female genomic reads (i.e. 2.5-fold higher coverage), and can thus only detect genes that have acquired considerably more copies on the Y chromosome relative to the X or autosomes. Indeed, our recent careful examination of gene family evolution on the fully sequenced and assembled neo-Y of D. miranda confirms that the true number of co-amplified X/Y gene families is much higher than what we can detect here: Direct sequence inspection revealed that at least 94 genes co-amplified on the X and Y of D. miranda [24], while we could only identify 15 genes with our methodology. In addition, we only probed for genes that are present in the D. melanogaster annotation. Most of the species that we surveyed here are only distantly related to D. melanogaster, and many genes from other species may simply not have a homolog in D. melanogaster. Indeed, about 1/3 of the co-amplified X/Y genes that we identified in D. miranda did not have an ortholog in D. melanogaster [24]. Finally, we required X-linked co-amplified gene copies to be present in our Illumina assemblies; however, recent gene duplicates are often collapsed is such assemblies [22]. Thus, our current list of co-amplified X/Y genes may only be the tip of the iceberg, and careful examination of high-quality genome sequences of X and Y chromosomes in many taxa may reveal the true extent of gene (co)-amplification on sex chromosomes.
Genes may amplify on the Y chromosome for a variety of reasons, and our current data do not allow us to evaluate their relative importance. In particular, multi-copy genes may simply arise on the Y at a higher rate, since the high repeat content on the Y facilitates structural re-arrangements that can promote gene family expansion [25]. Additionally, the efficacy of natural selection is reduced on the non-recombining Y, and Y chromosomes across diverse taxa accumulate functionless and deleterious repetitive DNA [16]. Amplified Y genes thus may either provide no benefit for their carriers, or could in fact be slightly deleterious, yet natural selection is unable to remove them [26]. Heterochromatin formation on the Y may further dampen any functional consequences of gene family expansion, and multi-copy Y genes may simply be more tolerated on the silenced Y. Finally, some multi-copy Y genes may actually contribute to male fitness and fertility [12–14,27]. Gene family expansion on the Y chromosome may help to compensate for reduced gene dose on the heterochromatic and transcriptionally repressed Y chromosome [16,17]. Y chromosomes are transmitted from father to son, and are thus an ideal genomic location for genes that specifically enhance male fitness [28]. Y chromosomes of several species, including mammals and Drosophila, have been shown to contain multi-copy gene families that are expressed in testis and contribute to male fertility [12–14]. Our analysis shows that multi-copy Y genes are common across flies, and it will be of great interest to identify the diverse evolutionary processes driving their amplification.
Co-amplification of X/Y genes, on the other hand, is more difficult to explain under scenarios that do not involve genetic conflict. Several factors that may explain accumulation of genes on the Y do not apply to the X: The repeat content of X chromosomes is comparable to that of autosomes [22]; natural selection efficiently purges deleterious mutations from the recombining X; and transcription of the X chromosome in Drosophila males is increased in somatic tissues, rather than reduced [29]. In addition, co-amplified X and Y genes are enriched for meiosis functions (see also [24]), and the X-linked copies of co-amplified genes are highly expressed in testis [24]. Functions in chromatin formation and chromosome segregation might be expected for selfish genes that are trying to interfere with proper condensation of the heterochromatic Y chromosome, or with fair segregation of homologous chromosomes. Testis expression of co-amplified X-linked genes is unusual, as testis-expressed genes are underrepresented on the X chromosome of Drosophila ([30], but also see [31]), but can be understood under intragenomic conflict models [32–36]. Most importantly, production of double-stranded RNA and triggering of the RNAi pathway is inconsistent with gene amplification boosting gene product, but instead has the opposite effect and results in transcriptional down-regulation of co-amplified X/Y genes.
Could other evolutionary forces or properties of sex chromosomes account for co-amplification of sex-linked genes? In several species, X chromosomes are down-regulated during spermatogenesis. While there has been considerable debate about the exact mechanisms of male germline X inactivation in Drosophila, testis genes appear transcriptionally repressed during spermatogenesis on old X chromosomes in Drosophila [37–39]. Co-amplification of testis genes could compensate for reduced expression of inactivated sex-linked genes and may thus be an adaptation to counter silencing of sex-linked genes in spermatogenesis. However, this model would not explain why co-amplified genes are frequent targets by endo-siRNA [24], which instead indicates a conflict between the X- and Y-linked copies, and not coordinated selection for their up-regulation. Also, many copies of co-amplified genes are truncated (as we observe for S-Lap or GAPsec, but also for others; see S6 Table), suggesting that the duplicated copies do not have the same function as their parent copies. Gene amplification to counter male germline X inactivation would also predict that (co)-amplified genes are more abundant on older sex chromosomes, where inactivation of the X in spermatogenesis is complete. In contrast, we find that co-amplified genes are more common in species with young neo-sex chromosomes (Fig 1). In particular, the young neo-X chromosome of D. miranda shows no signs of reduced expression in testis [40], yet we detect the largest numbers of co-amplified genes in this species (Fig 1).
Co-amplification of X and Y-linked genes could also allow meiotic pairing between diverging sex chromosomes. In particular, the ribosomal RNA gene cluster is present on both the X and the Y in D. melanogaster and functions as an X-Y pairing site during male meiosis [41]. This model, however, would not explain meiosis-specific function and testis-expression of co-amplified genes, or their targeting by endo-siRNA. Also, acquiring homologous pairing sites should also be more important on more divergent, heteromorphic sex chromosomes, counter to our finding of co-amplified genes being more common on young neo-sex chromosomes.
Co-amplification of genes on young sex chromosomes with meiosis-related functions, expression in testis, and targeting by endo-siRNAs can all be understood under a model of RNAi mediated cryptic sex chromosome drive. How would co-amplification of meiosis-related genes on the X and Y cause meiotic drive and its suppression? If amplified Y genes are involved in a battle with the X over fair transmission, changes in gene copy number may tip the balance over inclusion into functional sperm, and could result in repeated co-amplification of distorters and suppressors on the sex chromosomes (Fig 6A). In particular, an X-linked gene involved in chromosome segregation may evolve a duplicate that acquires the ability to incapacitate Y-bearing sperm (Fig 6A). Invasion of this sex-ratio distorter skews the population sex ratio and creates a selective advantage to evolve a Y-linked suppressor that is resistant to the distorter. Suppression may be achieved at the molecular level by increased copy number of the wildtype function or by inactivation of X-linked drivers using RNAi [33,34,42]. If both driver and suppressor are dosage sensitive, they would undergo iterated cycles of expansion, resulting in rapid co-amplification of both driver and suppressor on the X and Y chromosome [43]. Such a model is consistent with what we observe for the S-Lap and GAPsec genes in D. pseudoobscura (Fig 6B). S-Lap1 is the most abundant sperm protein in D. melanogaster [18] but its function is poorly characterized. If this protein is crucial for generating Y-bearing sperm, depletion of S-Lap1 during spermatogenesis would result in drive. S-Lap1 was duplicated in an ancestor of D. pseudoobscura, and a partial duplication of GAPsec (and truncation of S-Lap1-duplicate) created a TSS for anti-sense transcription of S-Lap1 duplicate. Anti-sense production of S-Lap1-duplicate transcript may trigger siRNA production and silencing of S-Lap1, which could result in elimination of Y-bearing sperm. Acquisition of multiple copies of S-Lap1 on the Y chromosome could restore S-Lap1 function, and create a cryptic drive system in D. pseudoobscura (Fig 6B). It is of course also possible that the Y-linked copies of S-Lap1 interfere with the production of X-linked sperm (i.e. that the Y chromosome is the driver), and S-Lap1-duplicate on the X silences Y-copies through production of antisense RNA and RNAi. A similar model of cryptic drive could also explain patterns of molecular evolution and gene expression at S-Lap and GAPsec in D. miranda, where read-through transcription generates anti-sense transcripts that trigger RNAi. Detailed molecular testing will be necessary to characterize the wildtype function of S-Lap1, and the cellular basis of the putative drive phenotype and its suppression. Below, we discuss how several aspects of the co-amplified genes that we have identified would make sense under a model of sex chromosome drive.
Most of the species where we identify co-amplified X/Y genes harbor neo-sex chromosomes. Under a drive model, this would make sense if sex ratio distorters have repeatedly evolved to exploit genomic vulnerabilities associated with the formation of new sex chromosomes. Different features of young vs. old sex chromosomes create different susceptibilities to sex chromosome drive. Old Y chromosomes are typically highly repetitive and heterochromatic, a feature that may easily be exploited by a driver on the X. Also, old sex chromosomes show much higher levels of sequence divergence, which makes identification and targeting of the homolog by a driver easier. Yet, young Y chromosomes typically contain many more genes that can evolve to cheat meiosis, thereby increasing the chances of a Y-linked driver. Finally, young X chromosomes may not yet be transcriptionally inactive during spermatogenesis and thus express more drivers. In many species, including Drosophila, expression from the X chromosome is reduced during spermatogenesis [37]. Low gene number and high repeat content makes Y chromosomes especially vulnerable to meiotic drive, and silencing of the X during spermatogenesis may have evolved as a genome defense against driving X’s [32]. Suppression of transcription during spermatogenesis may not have fully evolved on young X chromosomes, allowing the expression of more X-linked drivers. This may account for the prevalence of co-amplified X/Y genes in species with recently formed neo-sex chromosomes.
The RNAi pathway could be utilized in different ways to either create a meiotic driver, or to suppress it. For example, a gene on the X (or it’s duplicate) may gain a novel function that disrupts segregation of the Y chromosome. The homologous Y gene (or duplicates of it) may then silence the driving X by producing anti-sense transcripts that generate dsRNAs and launch the RNAi response, to silence the X-linked driver. This scenario resembles the Winters sex ratio system, even though the suppressors of X-linked drive are autosomal and not Y-linked [42]. The RNAi machinery can also be hijacked to create a driving X. In particular, if an X-linked gene is required for producing Y-bearing sperm, an X chromosome that silences this gene could evolve a drive phenotype. It could do so by antisense RNA production of this X-linked gene (or duplicates of it), thereby triggering the RNAi response to inactivate the gene. The organism could restore the wildtype function of this gene by increasing its dose through its amplification on the Y, and even non-functional copies may act as a decoy to soak up endo-siRNA that are targeting this locus. This pathway may underlie the putative GAPsec /S-Lap1 drive system in D. pseudoobscura, where the X-linked duplicates of S-Lap1 produce the vast majority of antisense transcripts (roughly 95%, see S8 Table), while the Y-linked S-Lap1 copies predominantly generate sense RNA (>99.9%). Most of the small RNAs are produced from S-Lap1 duplicate (the putative driver) and the Y-linked copies of S-Lap1 (about 96% in total), consistent with the idea that amplification of this spermatogenesis gene allows restoration of wildtype function, possibly by acting as a decoy to dilute RNAi induced silencing triggered by antisense transcripts of S-Lap1 duplicate. Under either scenario, if both the driver and suppressor are dosage-sensitive, this can lead to the repeated invasion of driving and suppressing chromosomes through co-amplification of genes on the X and Y chromosome.
In order to trigger the RNAi response, the production of dsRNA is required. This can be achieved in multiple ways. In the D. simulans Winters system, the two suppressor genes both encode related long inverted repeats that can form hairpin RNAs (hpRNAs), which are then processed by the RNAi machinery to generate siRNAs that repress the paralogous distorters [42]. Alternatively, the production of dsRNA can occur through anti-sense transcription of the target genes, and this mechanism is creating siRNAs in the putative drive involving GAPsec and S-Lap1, both in D. pseudoobscura and D. miranda.
Our data further support growing evidence that the production of antisense transcripts, hairpin RNAs and small RNAs may underlie some silenced meiotic drive systems [24,42]. RNA interference (RNAi) related pathways provide defense against viruses and transposable elements, and have been implicated in the suppression of meiotic drive elements [42]. Intriguingly, genes in these pathways often evolve rapidly, and show frequent gene duplication and loss over long evolutionary time periods. Argonaute 2 (Ago2), for example, is one of the key RNAi genes in insects, and has repeatedly formed new testis-specific duplicates in the recent history of the Drosophila obscura group [44]. Analysis of additional RNAi-pathway genes confirms that they undergo frequent independent duplications and that their history has been particularly labile within the Drosophila obscura group [45]. Our finding suggests that the presence of young sex chromosomes in this species group makes them especially vulnerable to the invasion of meiotic drive elements, and may thus drive the rapid evolution of RNAi genes in this clade. It will be of interest to study the dynamics of RNAi genes in other species groups that have gained novel sex chromosomes, to see if diversification of RNAi genes is correlated with the emergence of new sex chromosomes.
It is important to note that RNAi is only one means by which co-amplified genes could compete with each other to create and suppress sex chromosome drive. In particular, drive and silencing for the well-studied, co-amplified Slx/Sly genes in mice involves competition of SLX and SLY at the protein level for entry to the nucleus and for nuclear binding sites [7,8][46]. Also, not all drive systems lead to gene amplification. The Winters sex-ratio driver in D. simulans encodes a duplicate X-linked distorter (Dox/Nmy) that is silenced by their paralogous autosomal suppressors Nmy and Tmy through RNAi [33,34,42], and the Paris sex-ratio drive in D. simulans drive is caused by deficient alleles of a fast-evolving X-linked heterochromatin protein, showing that the rapid evolution of genes involved in heterochromatin structure can fuel intragenomic conflict [47]. Careful molecular dissection of several drive systems is necessary to establish general characteristics of segregation distorters.
Most co-amplified genes have considerably fewer copies on the X than the Y chromosome (S6 Table). Our approach is biased towards finding genes that are more highly amplified on the Y relative to the X, since we require genes to have increased M/F coverage ratios to be classified as Y-amplified in the first place. However, analyses of high-quality genome assemblies of D. pseudoobscura and D. miranda confirm that co-amplified genes indeed have many fewer copies on the X than on the Y, in both species (average copy number is 3 on the X, versus 44 on the Y). Thus, this difference in copy number of co-amplified genes on the X and the Y is not simply an artifact and surprising under a simple model of dose-dependent co-amplification of meiotic drivers and their suppressors. The reasons for this difference are not clear, but could include the following: Expression on the heterochromatic Y is generally dampened, and disproportionately more copies of a gene are needed to balance amplified X genes. This is supported by gene expression data from S-Lap1, where one parental copy on XR produces almost as many transcripts as we observe from the dozens of Y-linked copies combined (see S8 Table). Additionally, high copy-number gene arrays may be more difficult to maintain on the recombining X chromosome [48]. Also, RNAi induced drive models may not be stoichiometric. If X-linked drive operates by inactivation of a gene essential for Y chromosomes through antisense RNA production, a much larger number of sense Y-linked transcripts may be required to dilute silencing antisense transcripts. Interestingly, co-amplified genes on the mouse sex chromosomes are also much more abundant on the Y relative to the X; Sly/Slx have 126 copies on the Y and 39 on the X, Ssty/Sstx have 306 copies on the Y and 11 on the X, and Srsy/Srsx have 197 copies on the Y and 14 on the X [49]. Thus, higher copy number on the Y appears to be a general feature of co-amplified X/Y genes.
It is intriguing that S-Lap and GAPsec have repeatedly and independently co-amplified on the X and Y of multiple species in the D. obscura group, where Muller element D became sex-linked. In both D. pseudoobscura and D. miranda, two species for which we have stranded testis RNA-seq data and small RNA profiles, both genes produce endo-siRNA, indicating their involvement in a genomic conflict (see above). Understanding the molecular basis of this putative drive system will require detailed experimental work to characterize the wild-type function of these genes during spermatogenesis in the D. obscura group, and careful manipulation of the co-amplified X and Y copies. The function of S-Lap is poorly studied, but a recent paper suggests that all S-Lap genes in D. melanogaster are structural components of the mitochondrial paracrystalline material in sperm tails [50]. Sperm tail morphology varies dramatically between D. melanogaster and D. pseudoobscura, and unlike D. melanogaster, D. pseudoobscura has two types of sperm: long fertile eusperm (which are roughly 400μm long), and short infertile parasperm (roughly 100μm long; [51] [52]). While sperm heteromorphism is an intriguing phenomena, and may be exploited for drive, it complicates comparisons between sperm morphology and function between D. melanogaster and D. pseudoobscura. Intriguingly, GAPsec is a GTPase activating protein (GAPse is a Rab-GTPase), similar to the well-characterized Sd gene in D. melanogaster, which is a truncated duplicate of the RanGAP gene (which is a Ran-GTPase, [19]). Thus, either (or both) gene(s) may be considered a good candidate to be involved in meiotic drive, and it will be of great interest to study the wildtype function of these genes during spermatogenesis in D. pseudoobscura and its relatives. We focused our analysis in D. pseudoobscura on S-Lap1, since its duplicate copy on the X is a readily detectable and annotated gene that is transcribed. The duplicate of GAPsec, on the other hand, is highly degraded in D. pseudoobscura, not annotated as a gene, and it shows low levels of transcription. However, the fact that GAPsec is a GTPase activator that is independently co-amplifying with S-Lap on both the X and the Y in different species is captivating, and we readily detect antisense transcripts and small RNAs from both genes in D. miranda. This could mean that antisense transcription of either gene or possibly both is required for drive, and it could also imply that the cryptic drive system in D. pseudoobscura is now defunct. Higher copy number of both S-Lap and GAPsec on the Y, and more divergence among Y copies in D. pseudoobscura compared to D. miranda is consistent with this cryptic drive being older in D. pseudoobscura. Enough time may have passed in D. pseudoobscura for the drive to be fully silenced, after which point there is no selection to retain the driver, and it may start accumulating deactivating mutations.
To conclude, our comparative analysis suggests that co-amplification of genes on X and Y chromosomes may be relatively common in Drosophila, especially on young sex chromosomes, and we have shown that the same genes have been independently co-amplified in multiple species from the obscura group. We considered several evolutionary scenarios that would explain such amplifications, including compensation for male germline X inactivation, the formation of gene arrays to aid in meiotic chromosome pairing, and sex chromosome drive. We believe that sex chromosome drive is the most likely explanation for this pattern, for reasons discussed above, however, proof of this hypothesis will require careful experimental validation. The fact that these genes exist in multiple copies, are highly similar on the X and Y, and were all found in non-model Drosophila species that lack transgenic resources will make experimental validation of cryptic drive a very difficult task. That said, future characterization of the putative drive systems identified here would provide a full picture of how distorting elements manipulate and cheat meiosis, what molecular pathways or developmental processes are particularly vulnerable, and how the genome has launched evolutionary responses to counter distortion.
Strains were acquired from the Drosophila Species Stock Center (UC San Diego) or the EHIME stock center (Ehime University, Japan) as indicated in S1 Table. For each strain, DNA was extracted from a single male and a single female, using the Qiagen Gentra Puregene cell kit. The Illumina TruSeq Nano DNA library preparation kit was used to prepare 100 bp paired-end sequencing libraries for all species except D. robusta, D. melanica, and D. willistoni. For these species, the Illumina Nextera DNA library preparation kit was used to prepare 150 bp paired-end sequencing libraries. The genome assemblies produced for this study are noted in S1 Table. Assemblies were produced from the female data: reads were error-corrected using BFC [53] and assembled using IDBA-UD [54] with default parameters.
X chromosome/autosome fusions were identified in two steps [55]. For each species, genomic scaffolds were assigned to Muller elements based on their gene content, inferred from the results of a translated BLAST search of D. melanogaster peptides to the assembly of interest. Scaffolds smaller than 5kb were excluded. Next, the male and female Illumina data were separately mapped to the female assembly using Bowtie2 [56] and excluding alignments with mapping quality less than 20. The coverage ratio (M/F) was calculated for each scaffold that was assigned to a Muller element. The distribution of coverage ratios for each Muller element (S1 Fig) was then examined to determine if any of the ancestral autosomes had become X-linked. The raw (un-normalized) ratios are reported in S1 Fig. Most libraries were sequenced with similar number of reads for both males and females but for others, there was more data for males. Regardless of the value itself, the M/F values for an X-linked chromosome should be approximately half of the Y-linked values.
To characterize co-amplified genes on the sex chromosomes, we first identify genes amplified on the Y. For each species, male and female Illumina reads were separately aligned to a filtered version of the D. melanogaster peptide set, where only the longest isoform of each gene was retained. To generate these alignments, the DIAMOND software package [57] was used to perform a translated search of each Illumina read to the peptide set. Read coverage for each peptide sequence was calculated in 30 amino acid non-overlapping windows and normalized by dividing by the total number of mapped reads. The M/F coverage ratio was computed by dividing the median male coverage by the median female coverage, for each peptide. We required that potentially Y-amplified genes have a normalized M/F coverage ratio of at least 2.5 and only retained genes whose parent copy was X-linked in the species of interest. We searched for X-linked duplicates in the female genome assemblies by first using Exonerate [58] to extract the coding sequence of the best hit between the D. melanogaster peptide and the female assembly. We then used BLASTN [59] to obtain a stringent (E-value threshold = 1e-20) list of all non-overlapping hits between each exon of the coding sequence and the genome assembly. We considered a gene to be duplicated in females if at least 25% of the parent coding sequence aligned to more than one location in the genome assembly.
The Muller-D copies of S-Lap1 and GAPsec were identified in the 12 Drosophila genomes [60] by synteny with D. melanogaster and their coding sequences were downloaded from FlyBase [61]. The PRANK software package [62] was used to generate codon-aware alignments of coding sequences for each gene. The resulting alignment was trimmed using trimAl [63] and RaxML [64] was used to infer a maximum likelihood phylogeny (100 bootstrap replicates). D. pseudoobscura Y-linked contigs were identified using read coverage information from male versus female genomic sequencing data. Exonerate [58] was used to determine the location of the amplified copies of S-Lap1 and GAPsec on these scaffolds with the D. pseudoobscura S-Lap1 (FBpp0285960) and GAPsec (FBpp0308917) peptide sequences as queries. The D. pseudoobscura Y copies of each gene were aligned using MAFFT [65], trimmed with trimAl [63], and a Y consensus sequence for each gene was generated using PILER [66].
We dissected testes from 3–8 day old virgin males of D. pseudoobscura (strain MV25) reared at 18°C on Bloomington food. We used Trizol (Invitrogen) and GlycoBlue (Invitrogen) to extract and isolate total RNA. D. pseudoobscura CAGE-seq data were obtained from the ModEncode project [67]. We resolved 20 μg of total RNA on a 15% TBE-Urea gel (Invitrogen) and size selected 19–29 nt long RNA, and used Illumina’s TruSeq Small RNA Library Preparation Kit to prepare small RNA libraries, which were sequenced on an Illumina HiSeq 4000 at 50 nt read length (single-end). We used to Ribo-Zero to deplete ribosomal RNA from total RNA, and used Illumina’s TruSeq Stranded Total RNA Library Preparation Kit to prepare stranded testis RNA libraries, which were sequenced on an Illumina HiSeq 4000 at 100 nt read length (paired-end). Total RNA data were aligned to the D. pseudoobscura reference genome using HISAT2 [68], whereas Bowtie2 [56] (seed length: 18) was used to align small RNA and CAGE-seq data. In all cases, alignments with mapping quality less than 20 were discarded.
|
10.1371/journal.pcbi.1004390 | The Invariance Hypothesis Implies Domain-Specific Regions in Visual Cortex | Is visual cortex made up of general-purpose information processing machinery, or does it consist of a collection of specialized modules? If prior knowledge, acquired from learning a set of objects is only transferable to new objects that share properties with the old, then the recognition system’s optimal organization must be one containing specialized modules for different object classes. Our analysis starts from a premise we call the invariance hypothesis: that the computational goal of the ventral stream is to compute an invariant-to-transformations and discriminative signature for recognition. The key condition enabling approximate transfer of invariance without sacrificing discriminability turns out to be that the learned and novel objects transform similarly. This implies that the optimal recognition system must contain subsystems trained only with data from similarly-transforming objects and suggests a novel interpretation of domain-specific regions like the fusiform face area (FFA). Furthermore, we can define an index of transformation-compatibility, computable from videos, that can be combined with information about the statistics of natural vision to yield predictions for which object categories ought to have domain-specific regions in agreement with the available data. The result is a unifying account linking the large literature on view-based recognition with the wealth of experimental evidence concerning domain-specific regions.
| Domain-specific regions, like the fusiform face area, are a prominent feature of ventral visual cortex organization. Despite decades of interest from a large number of investigators employing diverse methods, there has been surprisingly little theoretical work on “why” the ventral stream may adopt this modular organization. In this study we propose a computational account of the role played by domain-specific regions in ventral stream function. It follows from a new theoretical analysis of the recognition problem which highlights the importance of building representations that are robust to class-specific transformations. These results provide a unifying account linking neuroimaging and neuropsychology-based ideas of domain-specific regions to the psychophysics and electrophysiology-oriented literature on view-based object recognition and invariance.
| The discovery of category-selective patches in the ventral stream—e.g., the fusiform face area (FFA)—is one of the most robust experimental findings in visual neuroscience [1–6]. It has also generated significant controversy. From a computational perspective, much of the debate hinges on the question of whether the algorithm implemented by the ventral stream requires subsystems or modules dedicated to the processing of a single class of stimuli [7, 8]. The alternative account holds that visual representations are distributed over many regions [9, 10], and the clustering of category selectivity is not, in itself, functional. Instead, it arises from the interaction of biological constraints like anatomically fixed inter-region connectivity and competitive plasticity mechanisms [11, 12] or the center-periphery organization of visual cortex [13–17].
The interaction of three factors is thought to give rise to properties of the ventral visual pathway: (1) The computational task; (2) constraints of anatomy and physiology; and (3) the statistics of the visual environment [18–22]. Differing presuppositions concerning their relative weighting lead to quite different models of the origin of category-selective regions. If the main driver is thought to be the visual environment (factor 3), then perceptual expertise-based accounts of category selective regions are attractive [23–25]. Alternatively, mechanistic models show how constraints of the neural “hardware” (factor 2) could explain category selectivity [12, 26, 27]. Contrasting with both of these, the perspective of the present paper is one in which computational factors are the main reason for the clustering of category-selective neurons.
The lion’s share of computational modeling in this area has been based on factors 2 and 3. These models seek to explain category selective regions as the inevitable outcome of the interaction between functional processes; typically competitive plasticity, wiring constraints, e.g., local connectivity, and assumptions about the system’s inputs [12, 26–28]. Mechanistic models of category selectivity may even be able to account for the neuropsychology [29, 30] and behavioral [31, 32] results long believed to support modularity.
Another line of evidence seems to explain away the category selective regions. The large-scale topography of object representation is reproducible across subjects [33]. For instance, the scene-selective parahippocampal place area (PPA) is consistently medial to the FFA. To explain this remarkable reproducibility, it has been proposed that the center-periphery organization of early visual areas extends to the later object-selective regions of the ventral stream [13–15, 17]. In particular, the FFA and other face-selective region are associated with an extension of the central representation, and PPA with the peripheral representation. Consistent with these findings, it has also been argued that real-world size is the organizing principle [16]. Larger objects, e.g., furniture, evoke more medial activation while smaller objects, e.g., a coffee mug, elicit more lateral activity.
Could category selective regions be explained as a consequence of the topography of visual cortex? Both the eccentricity [15] and real-world size [16] hypotheses correctly predict that houses and faces will be represented at opposite ends of the medial-lateral organizing axis. Since eccentricity of presentation is linked with acuity demands, the differing eccentricity profiles across object categories may be able to explain the clustering. However, such accounts offer no way of interpreting macaque results indicating multi-stage processing hierarchies [17, 34]. If clustering was a secondary effect driven by acuity demands, then it would be difficult to explain why, for instance, the macaque face-processing system consists of a hierarchy of patches that are preferentially connected with one another [35].
In macaques, there are 6 discrete face-selective regions in the ventral visual pathway, one posterior lateral face patch (PL), two middle face patches (lateral- ML and fundus- MF), and three anterior face patches, the anterior fundus (AF), anterior lateral (AL), and anterior medial (AM) patches [2, 36]. At least some of these patches are organized into a feedforward hierarchy. Visual stimulation evokes a change in the local field potential ∼ 20 ms earlier in ML/MF than in patch AM [34]. Consistent with a hierarchical organization involving information passing from ML/MF to AM via AL, electrical stimulation of ML elicits a response in AL and stimulation in AL elicits a response in AM [35]. In addition, spatial position invariance increases from ML/MF to AL, and increases further to AM [34] as expected for a feedforward processing hierarchy. The firing rates of neurons in ML/MF are most strongly modulated by face viewpoint. Further along the hierarchy, in patch AM, cells are highly selective for individual faces and collectively provide a representation of face identity that tolerates substantial changes in viewpoint [34].
Freiwald and Tsao argued that the network of face patches is functional. Response patterns of face patch neurons are consequences of the role they play in the algorithm implemented by the ventral stream. Their results suggest that the face network computes a representation of faces that is—as much as possible—invariant to 3D rotation-in-depth (viewpoint), and that this representation may underlie face identification behavior [34].
We carry out our investigation within the framework provided by a recent theory of invariant object recognition in hierarchical feedforward architectures [37]. It is broadly in accord with other recent perspectives on the ventral stream and the problem of object recognition [22, 38]. The full theory has implications for many outstanding questions that are not directly related to the question of domain specificity we consider here. In other work, it has been shown to yield predictions concerning the cortical magnification factor and visual crowding [39]. It has also been used to motivate novel algorithms in computer vision and speech recognition that perform competitively with the state-of-the-art on difficult benchmark tasks [40–44]. The same theory, with the additional assumption of a particular Hebbian learning rule, can be used to derive qualitative receptive field properties. The predictions include Gabor-like tuning in early stages of the visual hierarchy [45, 46] and mirror-symmetric orientation tuning curves in the penultimate stage of a face-specific hierarchy computing a view-tolerant representation (as in [34]) [46]. A full account of the new theory is outside the scope of the present work; we refer the interested reader to the references—especially [37] for details.
Note that the theory only applies to the first feedforward pass of information, from the onset of the image to the arrival of its representation in IT cortex approximately 100 ms later. For a recent review of evidence that the feedforward pass computes invariant representations, see [22]. For an alternative perspective, see [11]. Though note also, contrary to a claim in that review, position dependence is fully compatible with the class of models we consider here (including HMAX). [39, 47] explicitly model eccentricity dependence in this framework.
Our account of domain specificity is motivated by the following questions: How can past visual experience be leveraged to improve future recognition of novel individuals? Is any past experience useful for improving at-a-glance recognition of any new object? Or perhaps past experience only transfers to similar objects? Could it even be possible that past experience with certain objects actually impedes the recognition of others?
The invariance hypothesis holds that the computational goal of the ventral stream is to compute a representation that is unique to each object and invariant to identity-preserving transformations. If we accept this premise, the key question becomes: Can transformations learned on one set of objects be reliably transferred to another set of objects? For many visual tasks, the variability due to transformations in a single individual’s appearance is considerably larger than the variability between individuals. These tasks have been called “subordinate level identification” tasks, to distinguish them from between-category (basic-level) tasks. Without prior knowledge of transformations, the subordinate-level task of recognizing a novel individual from a single example image is hopelessly under-constrained.
The main thrust of our argument—to be developed below—is this: The ventral stream computes object representations that are invariant to transformations. Some transformations are generic; the ventral stream could learn to discount these from experience with any objects. Translation and scaling are both generic (all 2D affine transformations are). However, it is also necessary to discount many transformations that do not have this property. Many common transformations are not generic; 3D-rotation-in-depth is the primary example we consider here (see S1 Text for more examples). It is not possible to achieve a perfectly view-invariant representation from one 2D example. Out-of-plane rotation depends on information that is not available in a single image, e.g. the object’s 3D structure. Despite this, approximate invariance can still be achieved using prior knowledge of how similar objects transform. In this way, approximate invariance learned on some members of a visual category can facilitate the identification of unfamiliar category members. But, this transferability only goes so far.
Under this account, the key factor determining which objects could be productively grouped together in a domain-specific subsystem is their transformation compatibility. We propose an operational definition that can be computed from videos of transforming objects. Then we use it to explore the question of why certain object classes get dedicated brain regions, e.g., faces and bodies, while others (apparently) do not.
We used 3D graphics to generate a library of videos of objects from various categories undergoing rotations in depth. The model of visual development (or evolution) we consider is highly stylized and non-mechanistic. It is just a clustering algorithm based on our operational definition of transformation compatibility. Despite its simplicity, using the library of depth-rotation videos as inputs, the model predicts large clusters consisting entirely of faces and bodies.
The other objects we tested—vehicles, chairs, and animals—ended up in a large number of small clusters, each consisting of just a few objects. This suggests a novel interpretation of the lateral occipital complex (LOC). Rather than being a “generalist” subsystem, responsible for recognizing objects from diverse categories, our results are consistent with LOC actually being a heterogeneous region that consists of a large number of domain-specific regions too small to be detected with fMRI.
These considerations lead to a view of the ventral visual pathway in which category-selective regions implement a modularity of content rather than process [48, 49]. Our argument is consistent with process-based accounts, but does not require us to claim that faces are automatically processed in ways that are inapplicable to objects (e.g., gaze detection or gender detection) as claimed by [11]. Nor does it commit us to claiming there is a region that is specialized for the process of subordinate-level identification—an underlying assumption of some expertise-based models [50]. Rather, we show here that the invariance hypothesis implies an algorithmic role that could be fulfilled by the mere clustering of selectivity. Consistent with the idea of a canonical cortical microcircuit [51, 52], the computations performed in each subsystem may be quite similar to the computations performed in the others. To a first approximation, the only difference between ventral stream modules could be the object category for which they are responsible.
To make the invariance hypothesis precise, let gθ denote a transformation with parameter θ. Two images I, I′ depict the same object whenever ∃θ, such that I′ = gθ I. For a small positive constant ε, the invariance hypothesis is the claim that the computational goal of the ventral stream is to compute a function μ, called a signature, such that
| μ ( g θ I ) - μ ( I ) | ≤ ϵ . (1)
We say that a signature for which Eq (1) is satisfied (for all θ) is ϵ-invariant to the family of transformations {gθ}. An ϵ-invariant signature that is unique to an object can be used to discriminate images of that object from images of other objects. In the context of a hierarchical model of the ventral stream, the “top level” representation of an image is its signature.
One approach to modeling the ventral stream, first taken by Fukushima’s Neocognitron [53], and followed by many other models [54–58], is based on iterating a basic module inspired by Hubel and Wiesel’s proposal for the connectivity of V1 simple (AND-like) and complex (OR-like) cells. In the case of HMAX [55], each “HW”-module consists of one C-unit (corresponding to a complex cell) and all its afferent S-units (corresponding to simple cells); see Fig 1B. The response of an S-unit to an image I is typically modeled by a dot product with a stored template t, indicated here by ⟨I, t⟩. Since ⟨I, t⟩ is maximal when I = t (assuming that I and t have unit norm), we can think of an S-unit’s response as a measure of I’s similarity to t. The module corresponding to Hubel and Wiesel’s original proposal had several S-units, each detecting their stored template at a different position. Let g x ⃗ be the translation operator: when applied to an image, g x ⃗ returns its translation by x ⃗. This lets us write the response of the specific S-unit which signals the presence of template t at position x ⃗ as ⟨ I , g x ⃗ t ⟩. Then, introducing a nonlinear pooling function, which for HMAX would be the max function, the response C(I) of the C-unit (equivalently: the output of the HW-module, one element of the signature) is given by
C ( I ) = max i ( ⟨ I , g x → i t ⟩ ) (2)
where the max is taken over all the S-units in the module. The region of space covered by a module’s S-units is called its pooling domain and the C-unit is said to pool the responses of its afferent S-units. HMAX, as well as more recent models based on this approach typically also pool over a range of scales [56–58]. In most cases, the first layer pooling domains are small intervals of translation and scaling. In the highest layers the pooling domains are usually global, i.e. over the entire range of translation and scaling that is visible during a single fixation. Notice also that this formulation is more general than HMAX. It applies to a wide class of hierarchical models of cortical computation, e.g., [53, 58–60]. For instance, t need not be directly interpretable as a template depicting an image of a certain object. A convolutional neural network in the sense of [61, 62] is obtained by choosing t to be a “prototype” obtained as the outcome of a gradient descent-based optimization procedure. In what follows we use the HW-module language since it is convenient for stating the domain-specificity argument.
HW-modules can compute approximately invariant representations for a broad class of transformations [37]. However, and this is a key fact: the conditions that must be met are different for different transformations. Following Anselmi et al. [37], we can distinguish two “regimes”. The first regime applies to the important special case of transformations with a group structure, e.g., 2D affine transformations. The second regime applies more broadly to any locally-affine transformation.
For a family of transformations {gθ}, define the orbit of an image I to be the set OI = {gθ I, θ ∈ ℝ}. Anselmi et al. [37] proved that HW-modules can pool over other transformations besides translation and scaling. It is possible to pool over any transformation for which orbits of template objects are available. A biologically-plausible way to learn the pooling connections within an HW-module could be to associate temporally adjacent frames of the video of visual experience (as in e.g., [63–68]). In both regimes, the following condition is required for the invariance obtained from the orbits of a set of template objects to generalize to new objects. For all gθ I ∈ OI there is a corresponding gθ′ t ∈ Ot such that
⟨ g θ I , t ⟩ = ⟨ I , g θ ′ t ⟩ (3)
In the first regime, Eq (3) holds regardless of the level of similarity between the templates and test objects. Almost any templates can be used to recognize any other images invariantly to group transformations (see S1 Text). Note also that this is consistent with reports in the literature of strong performance achieved using random filters in convolutional neural networks [69–71]. Fig 1A illustrates that the orbit with respect to in-plane rotation is invariant.
In the second regime, corresponding to non-group transformations, it is not possible to achieve a perfect invariance. These transformations often depend on information that is not available in a single image. For example, rotation in depth depends on an object’s 3D structure and illumination changes depend on its material properties (see S1 Text). Despite this, approximate invariance to smooth non-group transformations can still be achieved using prior knowledge of how similar objects transform. Second-regime transformations are class-specific, e.g., the transformation of object appearance caused by a rotation in depth is not the same 2D transformation for two objects with different 3D structures. However, by restricting to a class where all the objects have similar 3D structure, all objects do rotate (approximately) the same way. Moreover, this commonality can be exploited to transfer the invariance learned from experience with (orbits of) template objects to novel objects seen only from a single example view.
The theory makes two core predictions:
Learned invariance to group transformations should be transferable from any set of stimuli to any other. For non-group transformations, approximate invariance will transfer within certain object classes. In the case of 3D depth-rotation, it will transfer within classes for which all members share a common 3D structure.
Both core predictions were addressed with tests of transformation-tolerant recognition based on a single example view. Two image sets were created to test the first core prediction: (A) 100 faces derived from the Max-Planck institute face dataset [72]. Each face was oval-cropped to remove external features and normalized so that all images had the same mean and variance over pixels (as in [73]). (B) 100 random greyscale noise patterns. 29 images of each face and random noise pattern were created by placing the object over the horizontal interval from 40 pixels to the left of the image’s center up to 40 pixels to the right of the image’s center in increments of 5 pixels. All images were 256 × 256 pixels.
Three image sets were created to test the second core prediction: (A) 40 untextured face models were rendered at each orientation in 5° increments from −95° to 95°. (B) 20 objects sharing a common gross structure (a conical shape) and differing from one another by the exact placement and size of smaller bumps. (C) 20 objects sharing gross structure consisting of a central pyramid on a flat plane and two walls on either side. Individuals differed from one another by the location and slant of several additional bumps. The face models were generated using Facegen [74]. Class B and C models were generated with Blender [75]. All rendering was also done with Blender and used perspective projection at a resolution of 256 × 256 pixels.
The tests of transformation-tolerant recognition from a single example were performed as follows. In each “block”, the model was shown a reference image and a set of query images. The reference image always depicted an object under the transformation with the median parameter value. That is, for rotation in depth of faces, it was a frontal face (0°) and for translation, the object was located in the center of the visual field. Each query image either depicted the same object as the reference image (target case) or a different object (distractor case). In each block, each query image was shown at each position or angle in the block’s testing interval. All testing intervals were symmetric about 0. Using a sequence of testing intervals ordered by inclusion, it was possible to investigate how tolerance declines with increasingly demanding transformations. The radius of the testing interval is the abscissa of the plots in Figs 2 and 3.
For each repetition of the translation experiments, 30 objects were randomly sampled from the template class and 30 objects from the testing class. For each repetition of the depth-rotation experiments, 10 objects were sampled from template and testing classes that were always disjoint from one another.
Networks consisting of K HW-modules were constructed where K was the number of sampled template objects. The construction followed the procedure described in the method section below. Signatures computed by these networks are vectors with K elements. In each block, the signature of the reference image was compared to the signature of each query image by its Pearson correlation and ranked accordingly. This ranked representation provides a convenient way to compute the ROC curve since it admits acceptance thresholds in terms of ranks (as opposed to real numbers). Thus, the final measure of transformation tolerance reported on the ordinate of the plots in Figs 2 and 3 is the mean area under the ROC curve (AUC) over all choices of reference object and repetitions of the experiment with different training / test set splits. Since AUC is computed by integrating over acceptance thresholds, it is a bias free statistic. In this case it is analogous to d′ for the corresponding 2AFC same-different task. When performance is invariant, AUC as a function of testing interval radius will be a flat line.
If there is imperfect invariance (ϵ-invariance), then performance will decline as the radius of the testing interval is increased. To assess imperfect invariance, it is necessary to compare with an appropriate baseline at whatever performance level would be achieved by similarity in the input. Since any choice of input encoding induces its own similarity metric, the most straightforward way to obtain interpretable results is to use the raw pixel representation as the baseline (red curves in Figs 2 and 3). Thus, a one layer architecture was used for these simulations: each HW-module directly receives the pixel representation of the input.
The first core prediction was addressed by testing translation-tolerant recognition with models trained using random noise templates to identify faces and vice versa (Fig 2). The results in the plots on the diagonal for the view-based model (blue curve) indicate that face templates can indeed be used to identify other faces invariantly to translation; and random noise templates can be used to identify random noise invariantly to translation. The key prediction of the theory concerns the off-diagonal plots. In those cases, templates from faces were used to recognize noise patterns and noise was used to recognize faces. Performance was invariant in both cases; the blue curves in Fig 2 were flat. This result was in accord with the theory’s prediction for the group transformation case: the templates need not resemble the test images.
The second core prediction concerning class-specific transfer of learned ϵ-invariance for non-group transformations was addressed by analogous experiments with 3D depth-rotation. Transfer of invariance both within and between classes was assessed using 3 different object classes: faces and two synthetic classes. The level of rotation tolerance achieved on this difficult task was the amount by which performance of the view-based model (blue curve) exceeded the raw pixel representation’s performance for the plots on the diagonal of Fig 3. The off-diagonal plots show the deleterious effect of using templates from the wrong class.
There are many other non-group transformations besides depth-rotation. S1 Text describes additional simulations for changes in illumination. These depend on material properties. It also describes simulations of pose (standing, sitting, etc)-invariant body recognition.
How can object experience—i.e., templates—be assigned to subsystems in order to facilitate productive transfer? If each individual object is assigned to a separate group, the negative effects of using templates from the wrong class are avoided; but past experience can never be transferred to new objects. So far we have only said that “3D structure” determines which objects can be productively grouped together. In this section we derive a more concrete criterion: transformation compatibility.
Given a set of objects sampled from a category, what determines when HW-modules encoding templates for a few members of the class can be used to approximately invariantly recognize unfamiliar members of the category from a single example view? Recall that the transfer of invariance depends on the condition given by Eq (3). For non-group transformations this turns out to require that the objects “transform the same way” (see S1 Text for the proof; the notion of a “nice class” is also related [76, 77]). Given a set of orbits of different objects (only the image sequences are needed), we would like to have an index ψ ¯ that measures how similarly the objects in the class transform. If an object category has too low ψ ¯, then there would be no gain from creating a subsystem for that category. Whenever a category has high ψ ¯, it is a candidate for having a dedicated subsystem.
The transformation compatibility of two objects A and B is defined as follows. Consider a smooth transformation T parameterized by i. Since T may be class-specific, let TA denote its application to object A. One of the requirements that must be satisfied for ϵ-invariance to transfer from an object A to an object B is that TA and TB have equal Jacobians (see S1 Text). This suggests an operational definition of the transformation compatibility between two objects ψ(A, B).
Let Ai be the ith frame of the video of object A transforming and Bi be the ith frame of the video of object B transforming. The Jacobian can be approximated by the “video” of difference images: JA(i) = ∣Ai − Ai+1∣ (∀i). Then define the “instantaneous” transformation compatibility ψ(A, B)(i): = ⟨JA(i), JB(i)⟩. Thus for a range of parameters i ∈ R = [−r, r], the empirical transformation compatibility between A and B is
ψ ( A , B ) : = 1 | R | ∑ i = - r r ⟨ J A ( i ) , J B ( i ) ⟩ . (4)
The index ψ ¯ that we compute for sets of objects is the mean value of ψ(A, B) taken over all pairs A, B from the set. For very large sets of objects it could be estimated by randomly sampling pairs. In the present case, we were able to use all pairs in the available data.
For the case of rotation in depth, we used 3D modeling / rendering software [75] to obtain (dense samples from) orbits. We computed the transformation compatibility index ψ ¯ for several datasets from different sources. Faces had the highest ψ ¯ of any naturalistic category we tested—unsurprising since recognizability likely influenced face evolution. A set of chair objects (from [78]) had very low ψ ¯ implying no benefit would be obtained from a chair-specific region. More interestingly, we tested a set of synthetic “wire” objects, very similar to those used in many classic experiments on view-based recognition e.g. [79–81]. We found that the wire objects had the lowest ψ ¯ of any category we tested; experience with familiar wire objects does not transfer to new wire objects. Therefore it is never productive to group them into a subsystem.
The above considerations suggest an unsupervised strategy for sorting object experience into subsystems. An online ψ-based clustering algorithm could sort each newly learned object representation into the subsystem (cluster) with which it transforms most compatibly. With some extra assumptions beyond those required for the main theory, such an algorithm could be regarded as a very stylized model of the development (or evolution) of visual cortex. In this context we asked: Is it possible to derive predictions for the specific object classes that will “get their own private piece of real estate in the brain” [8] from the invariance hypothesis?
The extra assumptions required at this point are as follows.
Cortical object representations (HW-modules) are sampled from the distribution D of objects and their transformations encountered under natural visual experience. Subsystems are localized on cortex. The number of HW-modules in a local region and the proportion belonging to different categories determines the predicted BOLD response for contrasts between the categories. For example, a cluster with 90% face HW-modules, 10% car HW-modules, and no other HW-modules would respond strongly in the faces—cars contrast, but not as strongly as it would in a faces—airplanes contrast. We assume that clusters containing very few HW-modules are too small to be imaged with the resolution of fMRI—though they may be visible with other methods that have higher resolution.
Any model that can predict which specific categories will have domain-specific regions must depend on contingent facts about the world, in particular, the—difficult to approximate—distribution D of objects and their transformations encountered during natural vision. Consider the following: HW-modules may be assigned to cluster near one another on cortex in order to maximize the transformation compatibility ψ ¯ of the set of objects represented in each local neighborhood. Whenever a new object is learned, its HW-module could be placed on cortex in the neighborhood with which it transforms most compatibly. Assume a new object is sampled from D at each iteration. We conjecture that the resulting cortex model obtained after running this for some time would have a small number of very large clusters, probably corresponding to faces, bodies, and orthography in a literate brain’s native language. The rest of the objects would be encoded by HW-modules at random locations. Since neuroimaging methods like fMRI have limited resolution, only the largest clusters would be visible to them. Cortical regions with low ψ ¯ would appear in neuroimaging experiments as generic “object regions” like LOC [82].
Since we did not attempt the difficult task of sampling from D, we were not able to test the conjecture directly. However, by assuming particular distributions and sampling from a large library of 3D models [74, 78], we can study the special case where the only transformation is rotation in depth. Each object was rendered at a range of viewpoints: −90° to 90° in increments of 5 degrees. The objects were drawn from five categories: faces, bodies, animals, chairs, and vehicles. Rather than trying to estimate the frequencies with which these objects occur in natural vision, we instead aimed for predictions that could be shown to be robust over a range of assumptions on D. Thus we repeated the online clustering experiment three times, each using a different object distribution (see S2 Table, and S6, S7, S8, S9, and S10 Figs).
The ψ-based clustering algorithm we used can be summarized as follows: Consider a model consisting of a number of subsystems. When an object is learned, add its newly-created HW-module to the subsystem with which its transformations are most compatible. If the new object’s average compatibility with all the existing subsystems is below a threshold, then create a new subsystem for the newly learned object. Repeat this procedure for each object—sampled according to the distribution of objects encountered in natural vision (or whatever approximation is available). See S1 Text for the algorithm’s pseudocode. Fig 4 shows example clusters obtained by this method.
Robust face and body clusters always appeared (Fig 5, S8, S9, and S10 Figs). Due to the strong effect of ψ ¯, a face cluster formed even when the distribution of objects was biased against faces as in Fig 5. Most of the other objects ended up in very small clusters consisting of just a few objects. For the experiment of Figs 4 and 5, 16% of the bodies, 64% of the animals, 44% of the chairs, and 22% of the vehicles were in clusters consisting of just one object. No faces ended up in single-object clusters.
To confirm that ψ-based clustering is useful for object recognition with these images, we compared the recognition performance of the subsystems to the complete system that was trained using all available templates irrespective of their cluster assignment. We simulated two recognition tasks: one basic-level categorization task, view-invariant cars vs. airplanes, and one subordinate-level task, view-invariant face recognition. For these tests, each “trial” consisted of a pair of images. In the face recognition task, the goal was to respond ‘same’ if the two images depicted the same individual. In the cars vs. airplanes case, the goal was to respond ‘same’ if both images depicted objects of the same category. In both cases, all the objects in the cluster were used as templates; the test sets were completely disjoint. The classifier was the same as in Figs 2 and 3. In this case, the threshold was optimized on a held out training set.
As expected from the theory, performance on the subordinate-level view-invariant face recognition task was significantly higher when the face cluster was used (Fig 5B). The basic-level categorization task was performed to similar accuracy using any of the clusters (Fig 5C). This confirms that invariance to class-specific transformations is only necessary for subordinate level tasks.
We explored implications of the hypothesis that achieving transformation invariance is the main goal of the ventral stream. Invariance from a single example could be achieved for group transformations in a generic way. However, for non-group transformations, only approximate invariance is possible; and even for that, it is necessary to have experience with objects that transform similarly. This implies that the optimal organization of the ventral stream is one that facilitates the transfer of invariance within—but not between—object categories. Assuming that a subsystem must reside in a localized cortical neighborhood, this could explain the function of domain-specific regions in the ventral stream’s recognition algorithm: to enable subordinate level identification of novel objects from a single example.
Following on from our analysis implicating transformation compatibility as the key factor determining when invariance can be productively transferred between objects, we simulated the development of visual cortex using a clustering algorithm based on transformation compatibility. This allowed us to address the question of why faces, bodies, and words get their own dedicated regions but other object categories (apparently) do not [8]. This question has not previously been the focus of theoretical study.
Despite the simplicity of our model, we showed that it robustly yields face and body clusters across a range of object frequency assumptions. We also used the model to confirm two theoretical predictions: (1) that invariance to non-group transformations is only needed for subordinate level identification; and (2) that clustering by transformation compatibility yields subsystems that improve performance beyond that of the system trained using data from all categories. These results motivate the the next phase of this work: building biologically-plausible models that learn from natural video. Such models automatically incorporate a better estimate of the natural object distribution. Variants of these models may be able to quantitatively reproduce human level performance on simultaneous multi-category subordinate level (i.e., fine-grained) visual recognition tasks and potentially find application in computer vision as well as neuroscience. In [42], we report encouraging preliminary results along these lines.
Why are there domain-specific regions in later stages of the ventral stream hierarchy but not in early visual areas [2, 3]? The templates used to implement invariance to group transformations need not be changed for different object classes while the templates implementing non-group invariance are class-specific. Thus it is efficient to put the generic circuitry of the first regime in the hierarchy’s early stages, postponing the need to branch to different domain-specific regions tuned to specific object classes until later, i.e., more anterior, stages. In the macaque face-processing system, category selectivity develops in a series of steps; posterior face regions are less face selective than anterior ones [34, 83]. Additionally, there is a progression from a view-specific face representation in earlier regions to a view-tolerant representation in the most anterior region [34]. Both findings could be accounted for in a face-specific hierarchical model that increases in template size and pooling region size with each subsequent layer (e.g., [41, 42, 84, 85]). The use of large face-specific templates may be an effective way to gate the entrance to the face-specific subsystem so as to keep out spurious activations from non-faces. The algorithmic effect of large face-specific templates is to confer tolerance to clutter [41, 42]. These results are particularly interesting in light of models showing that large face templates are sufficient to explain holistic effects observed in psychophysics experiments [73, 86].
As stated in the introduction, properties of the ventral stream are thought to be determined by three factors: (1) computational and algorithmic constraints; (2) biological implementation constraints; and (3) the contingencies of the visual environment [18–22]. Up to now, we have stressed the contribution of factor (1) over the others. In particular, we have almost entirely ignored factor (2). We now discuss the role played by anatomical considerations in this account of ventral stream function.
That the the circuitry comprising a subsystem must be localized on cortex is a key assumption of this work. In principle, any HW-module could be anywhere, as long as the wiring all went to the right place. However, there are several reasons to think that the actual constraints under which the brain operates and its available information processing mechanisms favor a situation in which, at each level of the hierarchy, all the specialized circuitry for one domain is in a localized region of cortex, separate from the circuitry for other domains. Wiring length considerations are likely to play a role here [87–90]. Another possibility is that localization on cortex enables the use of neuromodulatory mechanisms that act on local neighborhoods of cortex to affect all the circuitry for a particular domain at once [91].
There are other domain-specific regions in the ventral stream besides faces and bodies; we consider several of them in light of our results here. It is possible that even more regions for less-common (or less transformation-compatible) object classes would appear with higher resolution scans. One example may be the fruit area, discovered in macaques with high-field fMRI [3].
Lateral Occipital Complex (LOC) [82] These results imply that LOC is not really a dedicated region for general object processing. Rather, it is a heterogeneous area of cortex containing many domain-specific regions too small to be detected with the resolution of fMRI. It may also include clusters that are not dominated by one object category as we sometimes observed appearing in simulations (see Fig 4 and S1 Text). The Visual Word Form Area (VWFA) [4] In addition to the generic transformations that apply to all objects, printed words undergo several non-generic transformations that never occur with other objects. We can read despite the large image changes occurring when a page is viewed from a different angle. Additionally, many properties of printed letters change with typeface, but our ability to read—even in novel fonts—is preserved. Reading hand-written text poses an even more severe version of the same computational problem. Thus, VWFA is well-accounted for by the invariance hypothesis. Words are frequently-viewed stimuli which undergo class-specific transformations. This account appears to be in accord with others in the literature [92, 93]. Parahippocampal Place Area (PPA) [94] A recent study by Kornblith et al. describes properties of neurons in two macaque scene-selective regions deemed the lateral and medial place patches (LPP and MPP) [95]. While homology has not been definitively established, it seems likely that these regions are homologous to the human PPA [96]. Moreover, this scene-processing network may be analogous to the face-processing hierarchy of [34]. In particular, MPP showed weaker effects of viewpoint, depth, and objects than LPP. This is suggestive of a scene-processing hierarchy that computes a representation of scene-identity that is (approximately) invariant to those factors. Any of them might be transformations for which this region is compatible in the sense of our theory. One possibility, which we considered in preliminary work, is that invariant perception of scene identity despite changes in monocular depth signals driven by traversing a scene (e.g., linear perspective) could be discounted in the same manner as face viewpoint. It is possible that putative scene-selective categories compute depth-tolerant representations. We confirmed this for the special case of long hallways differing in the placement of objects along the walls: a view-based model that pools over images of template hallways can be used to recognize novel hallways [97]. Furthermore, fast same-different judgements of scene identity tolerate substantial changes in perspective depth [97]. Of course, this begs the question: of what use would be a depth-invariant scene representation? One possibility could be to provide a landmark representation suitable for anchoring a polar coordinate system [98]. Intriguingly, [95] found that cells in the macaque scene-selective network were particularly sensitive to the presence of long straight lines—as might be expected in an intermediate stage on the way to computing perspective invariance.
Is this proposal at odds with the literature emphasizing the view-dependence of human vision when tested on subordinate level tasks with unfamiliar examples—e.g. [72, 79, 99]? We believe it is consistent with most of this literature. We merely emphasize the substantial view-tolerance achieved for certain object classes, while they emphasize the lack of complete invariance. Their emphasis was appropriate in the context of earlier debates about view-invariance [100–103], and before differences between the view-tolerance achieved on basic-level and subordinate-level tasks were fully appreciated [104–106].
The view-dependence observed in experiments with novel faces [72, 107] is consistent with the predictions of our theory. The 3D structure of faces does not vary wildly within the class, but there is still some significant variation. It is this variability in 3D structure within the class that is the source of the imperfect performance in our simulations. Many psychophysical experiments on viewpoint invariance were performed with synthetic “wire” objects defined entirely by their 3D structure e.g., [79–81]. We found that they were by far, the least transformation-compatible (lowest ψ ¯) objects we tested (Table 1). Thus our proposal predicts particularly weak performance on viewpoint-tolerance tasks with novel examples of these stimuli and that is precisely what is observed [80].
Tarr and Gauthier (1998) found that learned viewpoint-dependent mechanisms could generalize across members of a homogenous object class [106]. They tested both homogenous block-like objects, and several other classes of more complex novel shapes. They concluded that this kind of generalization was restricted to visually similar objects. These results seem to be consistent with our proposal. Additionally, our hypothesis predicts better within-class generalization for object classes with higher ψ ¯. That is, transformation compatibility, not visual similarity per se, may be the factor influencing the extent of within-class generalization of learned view-tolerance. Though, in practice, the two are usually correlated and hard to disentangle. In a related experiment, Sinha and Poggio (1996) showed that the perception of an ambiguous transformation’s rigidity could be biased by experience [108]. View-based accounts of their results predict that the effect would generalize to novel objects of the same class. Since this effect can be obtained with particularly simple stimuli, it might be possible to design them so as to separate specific notions of visual similarity and transformation compatibility. In accord with our prediction that group transformations ought to be discounted earlier in the recognition process, [108] found that their effect was spared by presenting the training and test objects at different scales.
Many authors have argued that seemingly domain-specific regions are actually explained by perceptual expertise [24–27, 109]. Our account is compatible with some aspects of this idea. However, it is largely agnostic about whether the sorting of object classes into subsystems takes place over the course of evolution or during an organism’s lifetime. A combination of both is also possible—e.g. as in [110]. That said, our proposal does intersect this debate in several ways.
Our theory agrees with most expertise-based accounts that subordinate-level identification is the relevant task. The expertise argument has always relied quite heavily on the idea that discriminating individuals from similar distractors is somehow difficult. Our account allows greater precision: the precise component of difficulty that matters is invariance to non-group transformations. Our theory predicts a critical factor determining which objects could be productively grouped into a module that is clearly formulated and operationalized: the transformation compatibility ψ ¯.
Under our account, domain-specific regions arise because they are needed in order to facilitate the generalization of learned transformation invariance to novel category-members. Most studies of clustering and perceptual expertise do not use this task. However, Srihasam et al. tested a version of the perceptual expertise hypothesis that could be understood in this way [111]. They trained macaques to associate reward amounts with letters and numerals (26 symbols). In each trial, a pair of symbols were displayed and the task was to pick the symbol associated with greater reward. Importantly, the 3-year training process occurred in the animal’s home cage and eye tracking was not used. Thus, the distance and angle with which the monkey subjects viewed the stimuli was not tightly controlled during training. The symbols would have projected onto their retina in many different ways. These are exactly the same transformations that we proposed are the reason for the VWFA. In accord with our prediction, Srihasam et al. found that this training experience caused the formation of category-selective regions in the temporal lobe. Furthermore, the same regions were activated selectively irrespective of stimulus size, position, and font. Interestingly, this result only held for juvenile macaques, implying there may be a critical period for cluster formation [111].
Our main prediction is the link between transformation compatibility and domain-specific clustering. Thus one way to test whether this account of expertise-related clustering is correct could be to train monkeys to recognize individual objects of unfamiliar classes invariantly to 3D rotation in depth. The task should involve generalization from a single example view of a novel exemplar. The training procedure should involve exposure to videos of a large number of objects from each category undergoing rotations in depth. Several categories with different transformation compatibilities should be used. The prediction is that after training there will be greater clustering of selectivity for the classes with greater average transformation compatibility (higher ψ ¯). Furthermore, if one could record from neurons in the category-selective clusters, the theory would predict some similar properties to the macaque face-processing hierarchy: several interconnected regions progressing from view-specificity in the earlier regions to view-tolerance in the later regions. However, unless the novel object classes actually transform like faces, the clusters produced by expertise should be parallel to the face clusters but separate from them.
How should these results be understood in light of recent reports of very strong performance of “deep learning” computer vision systems employing apparently generic circuitry for object recognition tasks e.g., [62, 112]? We think that exhaustive greedy optimization of parameters (weights) over a large labeled data set may have found a network similar to the architecture we describe since all the basic structural elements (neurons with nonlinearities, pooling, dot products, layers) required by our theory are identical to the elements in deep learning networks. If this were true, our theory would also explain what these networks do and why they work.
An HW-architecture refers to a feedforward hierarchical network of HW-layers. An HW-layer consists of K HW-modules arranged in parallel to one another (see Fig 1B). For an input image I, the output of an HW-layer is a vector μ(I) with K elements. If I depicts a particular object, then μ(I) is said to be the signature of that object.
The parameters (weights) of the k-th HW-module are uniquely determined by its template book T k = { t k 1 , ⋯ , t k m } . (5)
For all simulations in this paper, the output of the k-th HW-module is given by
μ k ( I ) = max t ∈ T k ( ⟨ I , t ⟩ ∥ I ∥ ∥ t ∥ ) . (6)
We used a nonparametric method of training HW-modules that models the outcome of temporal continuity-based unsupervised learning [42, 67]. In each experiment, the training data consisted of K videos represented as sequences of frames. Each video depicted the transformation of just one object. Let G0 be a family of transformations, e.g., a subset of the group of translations or rotations. The set of frames in the k-th video was Otk = {gtk ∣ g ∈ G0}.
In each simulation, an HW-layer consisting of K HW-modules was constructed. The template book Tk of the k-th HW-module was chosen to be
T k : = O t k = { g t k | g ∈ G 0 } . (7)
Note that HW-architectures are usually trained in a layer-wise manner (e.g., [57]). That is, layer ℓ templates are encoded as “neural images” using the outputs of layer ℓ − 1. However, in this paper, all the simulations use a single HW-layer.
One-layer HW-architectures are a particularly stylized abstraction of the ventral stream hierarchy. With our training procedure, they have no free parameters at all. This makes them ideal for simulations in which the aim is not to quantitatively reproduce experimental phenomena, but rather to study general principles of cortical computation that constrain all levels of the hierarchy alike.
|
10.1371/journal.pcbi.1004693 | Molecular Mechanisms of Glutamine Synthetase Mutations that Lead to Clinically Relevant Pathologies | Glutamine synthetase (GS) catalyzes ATP-dependent ligation of ammonia and glutamate to glutamine. Two mutations of human GS (R324C and R341C) were connected to congenital glutamine deficiency with severe brain malformations resulting in neonatal death. Another GS mutation (R324S) was identified in a neurologically compromised patient. However, the molecular mechanisms underlying the impairment of GS activity by these mutations have remained elusive. Molecular dynamics simulations, free energy calculations, and rigidity analyses suggest that all three mutations influence the first step of GS catalytic cycle. The R324S and R324C mutations deteriorate GS catalytic activity due to loss of direct interactions with ATP. As to R324S, indirect, water-mediated interactions reduce this effect, which may explain the suggested higher GS residual activity. The R341C mutation weakens ATP binding by destabilizing the interacting residue R340 in the apo state of GS. Additionally, the mutation is predicted to result in a significant destabilization of helix H8, which should negatively affect glutamate binding. This prediction was tested in HEK293 cells overexpressing GS by dot-blot analysis: Structural stability of H8 was impaired through mutation of amino acids interacting with R341, as indicated by a loss of masking of an epitope in the glutamate binding pocket for a monoclonal anti-GS antibody by L-methionine-S-sulfoximine; in contrast, cells transfected with wild type GS showed the masking. Our analyses reveal complex molecular effects underlying impaired GS catalytic activity in three clinically relevant mutants. Our findings could stimulate the development of ATP binding-enhancing molecules by which the R324S mutant can be repaired extrinsically.
| Glutamine synthetase (GS) catalyzes the ATP-dependent ligation of ammonia and glutamate to glutamine, which makes the enzyme essential for human nitrogen metabolism. Three mutations in human GS, R324C, R324S, and R341C, had been identified previously that lead to a glutamine deficiency, resulting in neonatal death in the case of R324C and R341C. However, the molecular mechanisms underlying this impairment of GS activity have remained elusive. Our results from computational biophysics approaches suggest that all three mutants influence the first step of GS’ catalytic cycle, namely ATP or glutamate binding. The analyses reveal a complex set of effects including the loss of direct interactions to substrates, the involvement of water-mediated interactions that alleviate part of the mutation effect, and long-range effects between the catalytic site and structural parts distant from it. As to the latter, experimental validation is in line with our prediction of a significant destabilization of helix H8 in the R341C mutant, which should negatively affect glutamate binding. Finally, our findings could stimulate the development of ATP-binding enhancing molecules for the R324S mutant, which has been suggested to have residual activity, that way extrinsically “repairing” the mutant.
| Glutamine synthetase (GS, glutamate ammonia ligase, EC 6.3.1.2) catalyzes the ATP-dependent ligation of glutamate and ammonia to glutamine [1]. GS is ubiquitously expressed in human tissues. High expression levels of GS are found in astrocytes in brain tissues [2], where it is part of glutamate-glutamine cycling [3], and in perivenous hepatocytes, where it is part of the intercellular glutamine cycle and essential for ammonia detoxification by the liver [4–6]. Glutamate clearance, ammonia detoxification, and glutamine formation make GS essential for the human nitrogen metabolism [7, 8] and for neurological functionality. Accordingly, several links between changes in GS activity and neurological disorders have been described, including Alzheimer’s disease [9, 10], schizophrenia [11], hepatic encephalopathy [12–14] and epilepsy [15, 16]. In particular, two mutations in the GS gene (R324C in patient 1 and R341C in patient 2; throughout the manuscript, the sequence numbering of human GS is used) have been linked to congenital human GS deficiency with severe brain malformations resulting in multiorgan failure and neonatal death [17, 18]. In immortalized lymphocytes, R324C GS activity was reduced to about 12% of that found in wild type controls [17]. In fibroblasts from the father of patient 2, a 50% drop in specific GS activity was found, which may have been compensated for by a parallel increase in GS expression [17]. In a single case known to date, another GS mutation (R324S) was identified in a boy, now seven years old (patient 3), who is neurologically compromised due to the lack of ammonia detoxification and glutamine synthesis [19]. A plausible but not proven explanation for the survival of this patient would be the assumption of a higher level of GS residual activity compared to the other two GS mutants [20]. However, the molecular mechanisms for how these mutations lead to glutamine deficiency have not been understood.
Human GS belongs to class II of GS enzymes [21] and forms a homodecamer [22] in which two pentameric rings stack to each other; a bifunnel-shaped catalytic site is located in each interface formed by two adjacent subunits, resulting in ten catalytic sites in total (Fig 1A and 1B). For glutamine formation by GS, a two-step mechanism has been suggested [23, 24]: In the first step adenosine triphosphate (ATP) binds to GS, which induces conformational changes to enable binding of glutamate [24]. After glutamate bound to the complex, the terminal phosphate group of ATP is transferred to the γ-carboxylate function of glutamate yielding adenosine diphosphate (ADP) and γ-glutamyl phosphate (GGP), a reactive acyl-phosphate intermediate. In the second step, an ammonium ion binds to a negatively charged pocket formed by D63, S66, Y162, and E305. The ammonium ion transfers a proton to D63 to yield ammonia. Subsequently, ammonia attacks GGP, which results in inorganic phosphate and a tetrahedral, positively charged reaction intermediate that is stabilized by E305 via a salt bridge interaction. E305 then gets protonated, which destabilizes the salt bridge, leading to the opening of the glutamate binding site and glutamine release.
Residue R324, which is mutated to cysteine (R324C) or serine (R324S), is located in the catalytic site and forms an ionic salt bridge with the β-phosphate group of ADP in the crystal structure (Fig 1C) [22]. It is reasonable to assume that R324 interacts analogously with the GS substrate ATP, although no crystal structure of human GS with ATP or an ATP analog is available. Residue R341 is located 10 Å away from the catalytic site (Fig 1C). No explanation has been put forward how the R341C mutation influences GS’ catalytic activity over that distance.
Here, we investigated changes in GS structure, dynamics, and energetics at the atomistic level due to the three GS mutations R324C, R324S, and R341C by molecular dynamics (MD) simulations, rigidity analysis [25, 26], and free energy calculations. Our data show direct effects of the R324C/S mutations on the ATP binding, which are attenuated in the case of R324S due to the emergence of water-mediated interactions to ATP. In contrast, for R341C, we demonstrate a long-range influence on both ATP and glutamate binding: First, R341 indirectly influences ATP binding as a stabilizing element in an amino acid triplet; second, R341 connects two topologically separated regions between which information transmission is essential for glutamate binding. In vitro studies on the GS mutant H281A-H284A-Y288A (HHY), predicted to mimic the loss of interactions in the R341C mutant, provide evidence for this influence. These results can semi-quantitatively explain the observed GS deficiencies linked to the three mutations [17–19] and provide a basis for investigations how to counteract the effect due to the R324S mutation.
We performed molecular dynamics (MD) simulations of the wild type GS and the three GS mutants, R324C, R342S, and R341C. Coordinates of human GS were obtained from a crystal structure available from the Protein Data Bank (PDB) [27] as PDB entry 2QC8 [22] solved at 2.6 Å resolution. Human GS is a homodecamer with ten identical subunits, each consisting of 373 amino acids. As MD simulations of the GS decamer are computationally highly expensive, we considered a dimeric model system containing only two adjacent subunits forming a single catalytic site. The dimeric model was generated by extracting two adjacent monomers from the GS crystal structure (chains A and B). The validity of the dimeric model was checked by comparative MD simulations of the GS wild type decamer and the GS wild type dimer. Both systems were simulated in the presence of bound ADP, the intermediate GGP, and magnesium ions (Mg2+).
Using the dimeric model we investigated the influence of the three mutations on four different states according to the suggested mechanism of glutamine formation [24]: GS without a ligand (GSAPO), with bound ATP (GSATP), with bound ATP and glutamate (GSATP+GLU), and with bound ADP and GGP (GSADP+GGP). All states were modelled for wild type GS and the three GS mutants R324C, R342S, and R341C. Models of GS mutants were obtained by amino acid exchanges in the wild type dimer using the SwissPDBViewer [28]. For all mutants the best ranked side chain rotamers were used as starting conformations.
The GS crystal structure contains non-covalently bound ADP, the inhibitor L-methionine-S-sulfoximine phosphate (MSO-P), manganese ions (Mn2+), chloride ions, and crystal water [22]. For the GSATP and GSATP+GLU states, ADP was changed to ATP by adding the missing atoms with the LEaP program [29] of AmberTools 1.4 [30] according to the library of Meagher et al. [31]. In the case of GSADP+GGP, ADP coordinates were taken directly from the crystal structure, and hydrogen atoms were added according to the library of Meagher et al [31]. To generate GSATP+GLU and GSADP+GGP, glutamate and GGP were manually modelled based on the coordinates of the structurally similar inhibitor MSO-P present in the crystal structure. Structurally bound Mn2+ ions were changed into Mg2+ ions, for which well-validated simulation parameters [32] are available. Moreover, GS is catalytically active with Mg2+ ions [33]. Magnesium ions were present in all states GSAPO, GSATP, GSATP+GLU, and GSADP+GGP because the absence of divalent cations leads to a “relaxed” and inactive variant of GS [34, 35]. Nonetheless, we had to remove one Mg2+ ion in the case of GSATP and GSATP+GLU because the additional phosphate group of ATP causes clashes in the starting structure. Protonation states of histidines were assigned according to the protonation that was found to be most likely by visually inspecting the histidine environment.
The generated model systems were prepared for MD simulation with the LEaP program [29] of AmberTools 1.4 [30]. Sodium counter ions were added to the above described structures to neutralize each system. Model systems were placed in a truncated octahedral box of TIP3P water [36], leaving a distance of at least 11 Å between the solute and the border of the box. The finally obtained GS dimer systems comprised ~112,000 atoms. A system of the wild type GS decamer, prepared analogously, comprised ~354,000 atoms. For the polyphosphate chains of ADP and ATP, atomic partial charges and force field parameters were obtained from Meagher et al. [31]. Atomic partial charges for the substrate glutamate and the intermediate GGP were derived according to the restraint electrostatic potential fit (RESP) procedure [37]. Geometry optimizations and subsequent single point calculations were conducted with Gaussian03 [38] using the HF/6-31G* basis set. The resulting electrostatic potentials were fitted using respgen of AmberTools 1.4 [30]. Angle parameters for the phosphate group in GGP were taken from Homeyer et al. [39]. All other parameters were taken from the Amber ff99SB force field [40, 41].
The systems were relaxed by three steps of energy minimization, performed with the sander module of Amber11 [42]. First, harmonic restraints with a force constant of 5 kcal·mol-1·Å-2 were applied to all protein atoms, ligands, and structurally bound ions within the catalytic site while all other atoms were free to move (500 cycles steepest descent (SD) and 2000 cycles conjugate gradient (CG) minimization). Second, we reduced the harmonic restraints and applied a force constant of 1 kcal·mol-1·Å-2 (2000 cycles SD and 8000 cycles CG minimization). Finally, the positional restraints were removed completely, and all atoms were free to move (1000 cycles SD and 4000 cycles CG minimization).
The MD simulation procedure started by heating the respective system from 0 K to 100 K in a canonical (NVT) MD simulation of 50 ps length. During this heating step positional restraints of 1 kcal·mol-1·Å-2 were applied to all protein atoms, ligands, and structurally bound ions within the catalytic site. Afterwards, the temperature was raised from 100 K to ~300 K during 50 ps of isobaric-isothermal (NPT) MD. Subsequently, the density was adjusted to 1 g·cm-3 during 200 ps of NPT-MD. Finally, the harmonic positional restraints were removed by gradually decreasing the force constant from 1 to 0 kcal·mol-1·Å-2 in six NVT-MD runs of 50 ps length each. In the MD simulations, the particle mesh Ewald (PME) method [43–45] was employed to treat long-range electrostatic interactions. The SHAKE algorithm [46] was applied to all bonds involving hydrogens. A time step of 2 fs was used for the integration of the equations of motion. The distance cutoff for short range non-bonded interactions was set to 9 Å. In order to setup three independent MD production simulations, the target temperature was set to 299.9 K, 300.0 K, and 300.1 K in the equilibration, so that we obtained three different starting structures for subsequent MD production runs. Production MD simulations were performed in the NVT ensemble at 300 K for 100 ns. Coordinates were saved in a trajectory file every 20 ps. Using this MD simulation protocol, we generated three independent MD simulations for four different states (see above), for wild type GS and the GS mutants R324C, R324S, and R341C, which resulted in 3 × 4 × 4 = 48 MD simulations and an aggregate simulation time of 4.8 μs.
The 20–100 ns interval of each production run was considered for analysis. The analysis of the MD trajectories was carried out with ptraj [47] of AmberTools 1.4 [30]. The following measures were computed: the root mean-square fluctuation (RMSF) as a measure of mobility, the root mean-square deviation (RMSD) as a measure of structural similarity, the average secondary structure along the MD trajectory, water density grids, and distances. In addition, hydrogen bond interactions were determined using a distance of 2.8 Å between the two donor and acceptor atoms and an angle (donor atom, H, acceptor atom) of 120° as cutoff criteria for strong hydrogen bonds, and a distance of 3.2 Å and an angle of 120° as cutoff criteria for weak hydrogen bonds [48]. We analyzed whether a water-mediated chain of hydrogen bonds exists between ATP and C324 or S324, respectively. A water-mediated interaction was considered present when all hydrogen bonds in the water chain fulfilled the above distance and angle criteria. Results from three independent trajectories of the same system are expressed as means ± standard error of the mean (SEM). Results were analyzed with the R software [49] using the two-sided Student’s t-test. P values < 0.05 were considered significant.
The Constraint Network Analysis (CNA) approach allows linking biomacromolecular structure, flexibility, (thermo-)stability, and function [26]. To analyze the effect of R341 on the structural stability of GS, we extracted an ensemble of 4000 structures from the 20–100 ns interval of the MD simulation of wild type GS in the GSADP+GGP state that was equilibrated at 300.1 K; in this interval, the RMSD of GS relative to the starting structure remained particularly stable on average (~ 2 Å). In addition, we extracted an ensemble of 400 equally distributed structures from the 20–100 ns interval of MD simulations of the decameric wild type GS in the GSADP+ADP state. Coordinates (excluding water molecules, ions, and ligands) were extracted by mm_pbsa.pl [50] of Amber 11 [42]. Coordinates of an R341A GS mutant, used to mimic the loss of interactions of the R341 side chain with its environment, were generated employing the Ala-scan functionality of mm_pbsa.pl. This led to two sets of coordinates for dimeric and decameric GS, respectively, that differed only in residue 341. With CNA, thermal unfolding simulations of wild type GS and R341A GS were then performed to identify differences in the GSs’ structural stability [51]. For this, a hydrogen bond energy cutoff in the range of 0 to 6 kcal·mol-1 with steps of 0.1 kcal·mol-1 was used [26]. Stability maps [51] were then generated, which report when a rigid contact between two amino acids i and j (rcij) vanishes during the thermal unfolding simulation [25]. Finally, a difference stability map was calculated as rcij(wild type GS)–rcij(R341A GS); differences with p < 0.05 according to a Welch test [52] were considered significant.
Effective binding energies, i.e., the sum of gas-phase energies plus solvation free energies [53, 54], for the substrates ATP and glutamate were computed by the molecular mechanics Poisson-Boltzmann surface area (MM-PBSA) approach [55, 56]. The computations were performed with the mm_pbsa.pl script [50] of Amber 12 [57], using the ff99SB force field [40, 41] as in the MD simulations. The polar part of the solvation free energy was computed with the PBSA solver implemented in Amber 12 using dielectric constants of 4 and 80 for the solute and the solvent, respectively, and Parse radii [58] for the solute atoms. A solute dielectric constant of 4 was recommended for highly charged binding sites of proteins [59, 60], as given in the case of GS [22], to adequately account for screening effects of the binding site region.
Effective binding energies were computed according to the 1-trajectory MM-PBSA approach, in which snapshots of complex, receptor, and ligand are obtained from MD simulation of the complex [55]. While this approach neglects energetic effects due to conformational changes upon binding, it generally results in lower statistical uncertainties [55]. Contributions due to changes in the configurational entropy of the ligand or the receptor upon complex formation were neglected, too, in order to avoid introducing additional uncertainty in the computations [53, 59, 61]. Conformational ensembles for the computations were generated by extracting 4000 snapshots from the 20–100 ns interval of the MD trajectories of the GSATP and GSATP+GLU states of wild type GS and all three mutants R324C, R324S, and R341C. In the case of the GSATP+GLU state, effective binding energy calculations were performed considering glutamate as the ligand, whereas ATP was considered part of the receptor. The effective binding energies were averaged over the respective ensembles.
Relative effective binding energies (ΔΔG) were calculated by subtracting the effective binding energy of the wild type (ΔGwild type) from the effective binding energy of the mutant ΔGmutant for trajectory {1, 2, 3} (eq 1).
Results from the three independent MD simulations for a system are expressed as mean over the ΔΔG{1, 2, 3}. The SEM over the three independent MD simulations for a system X (SEMX) was calculated by error propagation according to eq 2.
SEM= SEM12+ SEM22+ SEM32
(2)
where SEM{1,2,3} is the SEM for trajectory {1, 2, 3}. The SEM of the relative effective binding energy (eq 2) was calculated according to eq 3.
A one-sample t-test with ΔΔG = 0 as reference was performed using the R software [49]. P values < 0.05 were considered significant.
L-methionine-S-sulfoximine (MSO) and polyclonal antibodies raised against the C-terminus of glutamine synthetase were from Sigma (Deisenhofen, Germany). The monoclonal antibody directed against GS (clone 6) was from Beckton-Dickinson (Heidelberg, Germany). The monoclonal antibody against GFP (green fluorescent protein), which cross-reacts with the YFP-variant (yellow fluorescent protein), was from Miltenyi-Biotech (Bergisch-Gladbach, Germany). Horseradish peroxidase-coupled goat anti-mouse IgG antibodies were from Bio-Rad International (Munich, Germany). Horseradish peroxidase-coupled goat anti-rabbit IgG antibodies were from Dako (Eching, Germany). The monoclonal antibody against glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was from Biodesign International (Cologne, Germany). Lipofectamine 2000 was from Life Technologies (Darmstadt, Germany).
Human embryonic kidney 293 (HEK293) cells were cultured on Petrie dishes (diameter = 60 mm) in minimal essential medium (MEM) containing Earle´s salt, L-glutamine and 5% fetal bovine serum (PAA, Linz, Austria). HEK293 cells were grown to about 70% confluency before cDNA (2 μg/dish) was introduced by lipofection using Lipofectamin 2000 according to the manufacturer’s instructions. 24 h after transfection, cells were either treated with L-methionine-S-sulfoximine (MSO, 3 mmol/l) or were left untreated for 2 h.
Human glutamine synthetase was cloned using human liver cDNA and the following primers GS-YFP-for: 5’-CGGAATTCATGACCACCTCAGCAAGTTC-3’ and GS-YFP-rev: 5’-CGGGATCCGCGTAATTTTTGTACTGGAAGG-3’. The forward primer contained an EcoRI restriction site, and the stop codon in the reverse primer was replaced by a BamHI site. The PCR product was cloned into the pEYFP-N1 vector (Clontech, Palo Alto, CA). Mutations were introduced into WT-GS-YFP using the QuikChange Multi Site-Directed Mutagenesis Kit (Agilent Technologies, Santa Clara, USA) and the following mutagenesis primers GS-R341A: 5’-GGTTACTTTGAAGATCGTGCCCCCTCTGCCAACTGCG-3’ and GS-SKR: 5’-GGCCATTGAGAAACTAGCCGCGGCGCACCAGTACCACATCC-3’. For the HHY variant the mutations were introduced sequentially using the following primers: GS-H281/84A-for: 5’-GAGAAACTAAGCAAGCGGGCCCAGTACGCCATCCGTGCCTATGATCC-3’ and GS-Y288A: 5’-CAGTACGCCATCCGTGCCGCTGATCCCAAGGGAGGCCTGG-3’. Successful cloning and mutagenesis was verified by sequencing (GenBank accession number: NM_002065).
Western-blot analysis was performed as described recently [62]. In brief, at the end of the experimental procedure proteins were purified from HEK293 cells and protein content was determined by the BioRad protein assay (BioRad, Munich, Germany). After polyacrylamide gel electrophoresis (10%), proteins were transferred onto nitrocellulose membranes using a semi-dry blotting chamber (BioRad, Munich, Germany). Membranes were incubated in bovine serum albumin (BSA, 10%) for 30 min and incubated with antibodies against GS (mAb, 1:5,000; pAb, 1:5,000), green fluorescent protein (GFP mAb 1:5,000), or glyceraldehyde-3-phosphate dehydrogenase (GAPDH, mAb, 1:5,000). Primary antibodies were detected using horseradish peroxidase-coupled anti-mouse or anti-rabbit IgG antibodies (1:10,000, 2 h at room temperature), respectively.
Dot-blot analysis was performed as described recently [62, 63] by spotting 2 μg of protein in a volume of 2 μl protein lysis buffer on a nitrocellulose membrane. After spots were dried for 30 min. at room temperature, immunodetection was performed using the membrane as described above for Western-blot analysis.
Peroxidase activity on the membranes was detected using Western-Lightning chemiluminescence reagent plus (Perkin Elmer, Waltham, USA). Digital images were captured using the Kodak Image Station 4000MM. Signal intensities were measured by densitometric analysis using the Kodak Molecular Imaging Software.
Immunofluorescence analysis was performed by confocal laserscanning microscopy (LSM510-META (Carl Zeiss AG, Oberkochem, Germany). HEK293 cells were seeded on MaTek dishes (MatTek Corporation, Ashland, USA) and transfected with cDNA constructs as described above. At the end of the transfection procedure, cells were incubated with Hoechst34580 (1:10,000; Life Technologies, Darmstadt, Germany) for 10 min at 37°C in an incubator (5% CO2). Cells were washed twice with phosphate-buffered saline before MatTek dishes were mounted on the LSM510-META and analyzed for YFP and Hoechst34580 immunofluorescence.
For statistical analysis, experiments were carried out with three separate HEK293 seedings. Results are expressed as mean values ± SEM and compared using a two-sided Student’s t-test (Excel for Windows; Microsoft, Redmond, USA). P values < 0.05 were considered significant.
In order to perform MD simulations more efficiently and, thus, improve conformational sampling of wild type GS and GS mutants, we established a model system consisting of two adjacent subunits of the GS decamer forming a single catalytic site (Fig 1B and 1C). The dimeric model system and the decameric structure of wild type GS in the GSADP+GGP state were subjected to MD simulations of 100 ns length at T = 300 K to probe the stability of the systems and to validate that the binding site structure does not deteriorate in the model system.
First, the RMSD, a measure of structural deviation along the MD trajectories, of all protein backbone atoms relative to the starting structure was analyzed (Fig 1D). The RMSD values for both systems increase until 20 ns. After this period, the RMSD of the GS decamer remains constant and below 2.0 Å (mean RMSD over 100 ns: 1.65 Å (SEM < 0.1 Å)). Thus, the structure of the GS decamer shows only minor changes during the MD simulation. The dimeric model system yields a mean RMSD of 2.31 Å (SEM < 0.1 Å), with a maximal value of ~3.5 Å after 70 ns (Fig 1D). The overall structural change of the dimeric model system is slightly larger than that of the GS decamer. Still, these values are in the range observed for other protein systems of that size during MD simulations of that length [64, 65]. Regions in the dimeric model that show particularly large conformational variations were identified by computing the RMSF per residue; the RMSF is a measure of the average atomic mobility. The largest differences in the conformational variability between the dimeric model and the GS decamer occur at the N- and C-termini of the subunits (Fig 1E). The lower RMSF values for the GS decamer result from all terminal protein chains interacting with adjacent subunits; such interactions are missing for the termini of the dimeric model. Hence, when considering only those 90% of the residues with the lowest RMSF (“core region”) in the RMSD calculations, the mean RMSD of the dimeric model decreases to 1.58 Å (SEM < 0.1 Å) (Fig 1D, Dimer90), which is comparable to the value for the GS decamer (see above).
Second, we focused our analysis on the binding site region, i.e., all residues that are within 4 Å of the bound ADP and GGP (Fig 1E). These residues show low mean RMSD values relative to the starting structure of 0.89 Å (SEM < 0.1 Å) in the MD simulations of the dimeric model system and for the GS decamer in the range of 0.82–2.29 (SEM < 0.1 Å) (Fig 1F). Regarding the bound ligands themselves, mean RMSD for ADP (GGP) (Fig 1G and 1H) of 0.44–0.69 Å (0.90–1.03 Å) in the case of the decamer and 0.49 Å (0.96 Å) in the case of the dimeric model were found (SEM < 0.1 Å in all cases).
In summary, small structural deviations of similar magnitude are found for the core regions, the binding sites, and the ligands of both the dimeric model system and the GS decamer with respect to the starting structures, demonstrating that both systems remain structurally stable over the simulation time. The dimeric model will thus be used to investigate the effects of the GS mutations.
Residue R324 is located in the catalytic site and forms a salt bridge with the β-phosphate group of ADP (Fig 1C) [22]. Although no crystal structure information for ATP-bound human GS is available, it is likely that R324 is interacting with the β-phosphate group of ATP, too. The substitution of R324 with cysteine reduces GS activity to about 12% of that of the wild type [17]. The substitution of R324 with serine likely partially conserves GS activity [20]. Initially, we investigated whether the R324S or R324C mutations induce structural changes within the catalytic site. For this, we computed the backbone RMSD of the residues of the catalytic site (Fig 1E) for both mutants in the GSAPO state (Fig 2A). In the case of the R324S mutant, the RMSD remains largely constant during three independent MD simulations (mean RMSD: 0.86 Å, 0.85 Å, and 1.26 Å (SEM < 0.1 Å)) (Fig 2A). In the case of the R324C mutant, the RMSD remains largely constant in two MD simulations (mean RMSD: 1.04 Å and 0.84 Å (SEM < 0.1 Å)) (Fig 2A). During one MD simulation, however, the RMSD fluctuates up to 3.5 Å, with a mean RMSD of 1.53 Å (SEM < 0.1 Å) (Fig 2A). Visual inspection of the respective trajectory revealed a highly mobile loop (termed “Glu flap” [21, 24]), formed by residues 304–306, as the cause for these fluctuations; excluding those three residues results in a mean RMSD of 0.95 Å (SEM < 0.1). All mean RMSD values are thus comparable with the mean RMSD obtained for the catalytic site of wild type GS (Fig1F). This demonstrates that neither the R324S nor the R324C mutation changes the catalytic site structure markedly.
Next, we hypothesized that differences in GS activity in the R324C and R324S mutants arise from differences in the interactions between arginine and the mutated residues, respectively, with ATP in the first step of glutamine formation. To investigate this, we subjected wild type GS and the R324C and R324S mutants in the GSATP, GSATP+GLU, and GSADP+GGP states to MD simulations. Distances were measured over the respective structural ensembles between the terminal guanidine nitrogens of R324 and oxygens oriented towards R324 of the β-phosphate group of ADP in the GSADP+GGP state, or ATP in the GSATP and GSATP+GLU states. The distance measurements confirmed the existence of a salt bridge interaction in the wild type for ADP [22] and revealed such an interaction for ATP (mean distances < 3.5 Å (SEM < 0.1 Å)) (Fig 2B), which is lower than the threshold of 4 Å used to define a salt bridge interaction [66]. This interaction is permanently present in eight out of nine trajectories (Figure A in S1 file), and is formed after ~65 ns in one trajectory (Figure A in S1 file) and remains stable thereafter. In contrast, in both R324 mutants, the mean distances between the thiol group of cysteine and the hydroxyl group of serine, respectively, and the β-phosphate group of ADP or ATP are > 7 Å (SEM < 0.1 Å) (Fig 2B). The differences in the mean distances with respect to the wild type are significant (p < 0.05). In addition, time series of the distances over the course of the MD simulations did not show the formation of hydrogen bonds between S324 or C324 and the β-phosphate group of ADP or ATP, respectively, in any conformation (Figure A in S1 file). Thus, neither residue at position 324 forms a direct hydrogen bond with the β-phosphate group of ADP or ATP in the R324C and R324S mutants. The lack of a direct interaction between the residue at position 324 and the β-phosphate group of ATP could lead to an increased mobility of ATP within the catalytic site. This could distort the proper mutual arrangement of the substrates prior to the reaction, leading to a loss in catalytic activity. However, the RMSF of ATP bound to wild type GS or one of the mutants do not differ significantly (Figure B in S1 file), which suggests that the mutation at position 324 may rather influence the affinity of the substrate ATP towards GS (see section “Relative effective binding energies of GS substrates” below).
Subsequently, we investigated why the R324S mutant likely leads to a higher residual GS activity than the R324C mutant [20]. We hypothesized that the direct interaction between the sidechain of R324 in the wild type could be replaced by water-mediated interactions in the R324S mutant but not the R324C mutant. Therefore, we determined the water density between residue 324 and ATP in MD trajectories by counting the presence of water molecules within a cubic grid of 0.33 Å spacing. Isopleth plots of the density distribution that encompass 80% of the maximum occupancy of water molecules in three independent MD simulations of the R324S and R324C mutants, respectively, in the GSATP+GLU state are shown in Fig 2C and 2D. These results qualitatively reveal a much broader region of high water density between R324S and the β-and γ-phosphate groups of ATP than in the R324C case, resulting in regions of high water occupancy close to the R324S side chain. For a quantitative analysis, radial distribution functions (RDF) for oxygen atoms of water molecules around the side chain oxygen (R324S) or sulfur (R324C) were computed (Fig 2E and 2F). The RDF reveals two shells of water molecules around R324S, with the first shell peaking at ~2.8 Å, consistent with previous findings [67], and the second shell at ~4.5 Å (Fig 2E). The distance of the first shell peak is in line with that of a strong hydrogen bond (2.5–3.2 Å) [48]. In the case of R324C, the first shell peaks at ~3.2 Å (Fig 2F). This difference with respect to R324S reflects a similar difference of the van der Waals radii of oxygen and sulfur [68]. More notable, the water density at the position of the first shell is 30% higher in the case of R324S than for R324C, and the second shell is considerably more structured in the former case (Fig 2E), demonstrating stronger hydrogen bonding interactions between serine and water, as expected [69].
Finally, we determined the frequency of occurrence of water-mediated hydrogen bonds between R324S or R324C and the β- and γ-phosphate groups of ATP (Fig 2G and 2H). From the distance between the two side chains and the β-phosphate group of ∼7.0–8.5 Å, respectively, (Fig 2B) and a water diameter of 2.9 Å [70], one can deduce that between two to three water molecules can bridge this gap. In the analysis, we distinguished between weak (distance cutoff between hydrogen bond donor and acceptor of dcut = 3.2 Å) and strong (dcut = 2.8 Å) hydrogen bonds. Only hydrogen bonds with a distance < dcut were considered for analyses. The frequency of occurrence of water-mediated hydrogen bonds between R324S and the β- and γ-phosphate groups of ATP is significantly higher than for R324C for both weak (31.0 ± 4.0% versus 11.8 ± 2.3%; 36.8 ± 7.7% versus 9.9 ± 2.0%; Fig 2G and 2H) and strong (4.8 ± 0.7% versus 0%; 12.2 ± 4.4% versus 0%; Fig 2G and 2H) hydrogen bonds.
In summary, our analyses show the absence of a direct hydrogen bonding interaction between the side chains of R324S and R324C mutants and the β-phosphate group of ATP, in contrast to the wild type GS. In the case of the R324S mutant, the direct interaction is replaced by water-mediated hydrogen bonds; such hydrogen bonds are significantly less frequently observed in the R324C mutant.
In fibroblasts from the father of a patient with an R341C mutation in GS, a 50% drop in specific GS activity was found [17]. As R341 is located at a distance of 10 Å from the catalytic site (Fig 1C) and its side chain points away from this site (Fig 3A), the influence of the R341C mutation on GS activity must arise from a long-range effect that percolates through the GS structure. Analysis of the GS crystal structure [22] revealed R341 as part of an amino acid triplet (Fig 3A; termed triad hereafter) consisting of R341, D339, and R340. While R341 is at the most distant end of the triad with respect to the catalytic site, R340 makes hydrogen bond and salt bridge interactions with the sulfoximine phosphate part of MSO-P (Fig 3A).
During our MD simulations of wild type GS in the GSADP+GGP state, we also observe a hydrogen bond between R340 and GGP (mean distance between the terminal guanidine nitrogens of R340 and the carbonyl oxygen in GGP: 2.63 Å (SEM < 0.1 Å); Fig 3B), in agreement with the crystal structure. This interaction is stable over 100 ns MD simulations (Figure C in S1 file). R340 does not interact with glutamate in the GSATP+GLU state, however (mean distance between the center of the terminal guanidine nitrogens of R340 and the center of the γ-carboxylic function in glutamate: 4.05 Å (SEM < 0.1 Å); Fig 3B, Figure C in S1 file). Rather, R340 interacts with the γ-phosphate group of ATP in the GSATP+GLU state (mean distance between the center of the terminal arginine nitrogens and the center of the oxygens oriented towards R340 of the γ-phosphate group of ATP: 3.06 Å (SEM < 0.1 Å)) as it does in the GSATP state (mean distance between the center of the terminal arginine nitrogens and the center of the oxygens oriented towards R340 of the γ-phosphate group of ATP: 2.88 Å (SEM < 0.1 Å)) (Fig 3B). Again, the interactions with the γ-phosphate group are stable over the course of 100 ns MD simulations (Figure C in S1 file). The observed shift of the interaction of R340 with ATP in the GSATP and GSATP+GLU states to one with GGP in the GSADP+GGP state suggests a prominent involvement of R340 in the first step of the catalytic cycle of GS.
Within the triad, D339 and R341 form a salt bridge in the crystal structure (Fig 3A, [22]). The mean distances SB1 and SB2 (Fig 3A, black dashed lines) between side chains of D339 and R341 are < 3 Å (SEM < 0.1 Å) in all states for wild type GS (Fig 3C). This interaction is present ≥ 98% of the time during MD simulations of wild type GS of all states (Fig 3D, SB1 and SB2 occupancy determined for salt bridges with dcut < 4.0 Å). These findings suggest that D339 and R341 stabilize R340 as flanking residues. This stabilization counteracts the inherent flexibility of the loop region all three residues are located in (Fig 3A). In the R341C mutant, the mean distances SB1 and SB2 between the carboxylate oxygens of D339 and the thiol function of C341 is > 4.5 Å (SEM < 0.1) in all states (Fig 3C). Employing dcut < 3.2 Å and angle < 120°, a hydrogen bond between the carboxylate oxygens of D339 and the thiol function of C341 is present in ≤ 5% of the time during MD simulations of all states (Fig 3D). These results suggest that the stabilizing effect of residue 341 on R340 is largely lost in the R341C mutation. To substantiate this, we computed residue-wise RMSF of R340 for wild type GS and the R341C mutant in all states (Fig 3E). For GSAPO, the RMSF is significantly (p < 0.05) larger in the R341C mutant (1.00 ± 0.03 Å) than the wild type GS (0.74 ± 0.03 Å) (Fig 3E). In contrast, no significant changes were observed for states GSATP, GSATP+GLU, and GSADP+GGP (Fig 3E), likely because R340 is then stabilized by interactions with the substrates (see above; Fig 3B). These results suggest that the R341C mutation indirectly affects the first step of the catalytic cycle, particularly ATP binding, by influencing the R340 mobility in the GSAPO state.
In order to investigate the influence of mutant R341C on GS’ mechanical stability, we applied Constrained Network Analysis (CNA), where biomolecular structures are represented as molecular frameworks [26] and analyzed by means of rigidity theory [71]. That way, regions that are either structurally stable (“rigid”) or flexible are identified. We compared results for wild type GS to those obtained for a perturbed structural ensemble in which interactions by the side chain of R341 are abolished (see “Experimental procedures” for how this ensemble was generated). The result is depicted as a difference stability map Δrcij [51], which shows if a rigid contact between two residues becomes less stable in the perturbed structural ensemble (red colors in Fig 3F; a rigid contact exists if two residues belong to the same rigid region). The loss of side chain interactions of R341 results in a significant (p < 0.05) destabilization of the C-terminus of the GS dimer model (residues 265–365; Fig 3F). Control calculations performed for the GS decamer corroborate this finding (Figure D in S1 file). In this region, helix 8 (H8, residues 266–288), located on the outside of the GS subunit (Fig 3A), shows the most pronounced loss in structural stability (with Δrcij values of -1.4 kcal·mol-1; see ref. [25] for an explanation of the energy values). Visual inspection of the GS crystal structure [22] reveals that the guanidino group of R341 forms hydrogen bonds with three residues on H8 (H281, H284, and Y288) (Fig 3A).
To substantiate these results, we analyzed the structural stability of H8 during MD simulations of wild type GS and the R341C mutant in the GSATP state, as well as of an H281A/H284A/Y288A triple mutant in the same state (hereafter termed HHY mutant). The HHY mutant serves as a mimic of the R341C mutant because here, as in the R341C mutant, no hydrogen bonds between residue 341 and residues 281, 284, and 288, respectively, can be formed. We computed the probability that residues 266–288 form a loop in the course of the MD simulations (Fig 3G). For wild type GS, this loop probability is below 20% for residues 278, 279, and 288, below 10% for residues 280, 286, and 287, and below 5% for the remaining amino acids (Fig 3G). For the R341C mutant, in contrast, a markedly increased loop probability is found for residues 279 to 283 (up to 50%; Fig 3G). In this region, mutant HHY shows the most distinct increases in the loop probability compared to wild type GS, too (up to 22%; Fig 3G). For residues 266–276, there are no differences in loop probability between wild type GS and R341C or HHY mutant, respectively. These results demonstrate that R341 has a stabilizing influence on H8 in wild type GS; this influence is lost in both the R341C and HHY mutants. Considering the above results on R341’s role in the triad, this suggests that R341 can relay information between the catalytic site and H8. Of note in this context, Krajewski et al. observed a shift of H8 in the first step of the catalytic cycle that closes GS’ catalytic site when ATP is bound [22]; this ATP binding-induced shift has been recognized as a prerequisite for glutamate binding [22]. Hence, we hypothesized that the loss of the relaying function in the R341C mutant hampers glutamate binding to GS. For the same reason, glutamate binding should be hampered in the HHY mutant.
YFP-tagged wild type and mutated (R341A; HHY; S278A/K279A/R280A = SKR, the latter mutant was introduced as a negative control to HHY, as S278, K279, and R280 are located on H8 but do not interact with R341) human GS were transiently expressed in HEK293 cells. Expression of human GS-YFP in HEK293 cells was monitored by confocal laser scanning microscopy (Fig 4A) and verified by Western-blot analysis, which showed that YFP-tagged GS-constructs coding for human GS were strongly expressed in HEK293 cells (Fig 4B). Treating GS-YFP transfected HEK293 cells with MSO (3 mmol/l, 2 h) had no effect on GS-YFP expression levels in wild type-, R341A-, or HHY-treated cells compared to the respective control (Fig 4B and 4C). However, anti-YFP-immunoreactivity was slightly elevated in MSO-treated HEK293 cells transfected with SKR-mutated GS-YFP (Fig 4B and 4C).
GS was also detected in GS-YFP transfected cells by Western-blot using a monoclonal antibody formerly shown not to react with GS when arginine at position 341 was mutated to cysteine (R341C) (15). As shown in Fig 4D, the monoclonal anti-GS antibody strongly detected overexpressed wild type-GS as well as HHY- and SKR-mutated GS-YFP but failed to detect GS when arginine 341 is mutated to alanine (R341A). In contrast, when using a polyclonal antibody (Sigma, Deisenhofen, Germany) raised against the C-terminus of GS (amino acids 357–373), GS-YFP mutated on arginine 341 (R341A) was readily detected, as were wildtype- and HHY- or SKR-mutated GS-YFP (Fig 4E). MSO-treatment had no effect on detectability of GS-YFP by Western-blot using the monoclonal antibody raised against GS (Fig 4D).
HEK293 cells also expressed GS endogenously (Fig 4D and 4E). However, overexpression of GS-YFP (Fig 4D and 4E) as well as MSO-treatment in GS-YFP transfected HEK293 cells had no effect on endogeneous GS expression levels (Fig 4D).
The results show that wildtype-, as well as R341A-, HHY- and SKR-mutated GS-YFP is efficiently expressed in HEK293 cells by lipofection and that GS-YFP expression levels are not affected by MSO-treatment compared to the respective control. The results also show that wildtype-, HHY- and SKR-mutated but not R341A-mutated GS is recognized by a monoclonal antibody raised against GS (BD, clone 6) in Western-blot. This is explained by the specificity of a monoclonal antibody, which recognizes only a single epitope, and by the interaction of the paratope of an antibody with an epitope, which is mediated by only about 5 amino acids [72]. Therefore, mutations that are located far away from the presumed recognition site (R341) such as S278 are not expected to impair the binding of the anti-GS antibody.
MSO binds non-covalently to the glutamate-binding pocket of GS [73] thereby masking an epitop which is recognized by a monoclonal antibody (BD, clone 6) [62]. Therefore, loss of anti-GS immunoreactivity after MSO treatment may serve as a surrogate marker for glutamate binding to the catalytic site of GS.
As shown by dot-blot analysis using native protein, anti-GS immunoreactivity was significantly diminished by about 50% after MSO-treatment in HEK293 cells transfected with wild type- or SKR-mutated GS but remained unchanged in HEK293 cells expressing HHY-mutated GS compared to untreated controls (Fig 5A). In contrast, upon heat- and detergent-mediated protein denaturation, anti-GS immunoreactivity was similar in untreated and MSO-treated, GS-transfected HEK293 cells (Fig 5B).
The results suggest that MSO binds to the glutamate-binding pocket of wildtype- and SKR-mutated-GS but not to the pocket of HHY-mutated GS.
In order to determine energetic consequences of the GS mutations on substrate binding, effective binding energies relative to wild type GS (ΔΔG, eq 1) were computed by the MM-PBSA approach for ATP bound to the R342S, R342C, and R341C mutants in the GSATP state and for glutamate bound to R342S, R342C, and R341C in the GSATP+GLU state. The average drift in the effective binding energy ΔGwildtype of glutamate binding to the GSATP+GLU state over the last 80 ns of the MD simulations used for analysis is 0.04 kcal·mol-1·ns-1, as determined by the slope of the least-squares line of best fit from a correlation analysis (Fig 6A). The magnitude of the average drifts in all other effective binding energies ΔGwildtype or ΔGmutant is < 0.15 kcal mol-1 ns-1 (Figure E and F in S1 file). The magnitude of these drifts is comparable to those found for ligands binding to other proteins [74] or ribosomal RNA [64] and indicates converged estimates of ΔGwild type or ΔGmutant.
The computed ΔΔG for ATP in the GSATP state is 4.29 ± 0.19 kcal·mol-1 for the R324S mutant and 4.64 ± 0.21 kcal·mol-1 for the R324C mutant (Fig 6B). While these results demonstrate that ATP binding to the mutants is disfavorable compared to binding to wild type GS, the difference between the two mutants is not significant. The latter finding is unexpected considering that both mutations lead to different clinical outcomes. Likely, the similar ΔΔG result from neglecting explicit water molecules in the MM-PBSA computations, which, consequently, results in missing favorable energetic contributions due to water-mediated hydrogen bonds between S324 and ATP, as observed in the MD simulations (Fig 2G and 2H). Including tightly bound structural water molecules in MM-PBSA calculations may provide a possibility to overcome this shortcoming [55, 75]. However, in the case of R324S GS, this approach does not appear applicable to us as we observed frequent exchanges of water molecules involved in the hydrogen bond formations.
For mutant R341C, ΔΔG is 5.93 ± 0.23 kcal·mol-1 for ATP and 2.24 ± 0.16 kcal·mol-1 for glutamate binding (Fig 6B). Thus, our results show that ATP binding is weakened in all GS mutants; at T = 300 K, the magnitudes of ΔΔG relate to decreases in the binding constants of ATP of 1.6 to 4.3 log units. From a qualitative point of view, the weakening in the R341C mutant is in line with the above findings that the mutation results in a destabilization of R340 (Fig 3E), which, in turn, interacts with ATP in the GSATP state (Fig 3B, Figure C in S1 file). From a quantitative point of view, it is surprising that of all mutants the R341C mutant shows the largest effect on ATP binding, although residue 341 does not directly interact with ATP. Part of this effect may be caused by neglecting energetic contributions due to conformational changes in the solutes upon binding in the 1-trajectory approach pursued here, or by neglecting changes in the configurational entropy of the solutes upon complex formation (see section Materials and Methods).
Glutamate binding is weakened in the R324C (R341C) GS mutant by ΔΔG = 2.09 ± 0.10 kcal mol-1 (ΔΔG = 2.29 ± 0.16 kcal mol-1) but not in R324S GS (ΔΔG = -1.28 ± 0.09 kcal mol-1) (Fig 6B). Considering that, according to Krajewski et al. [22], an ATP binding-induced conformational change in GS’ catalytic site is a prerequisite for glutamate binding, our findings are in line with our above structural and energetic analyses according to which ATP binding is particularly weakened in the R324C and R341C mutants.
The molecular mechanisms of how the three mutations R324C, R324S, and R341C in human GS [17, 19] lead to a glutamine deficiency, resulting in neonatal death in the case of R324C and R341C, have not been understood. Furthermore, it has remained elusive why the R324S mutation, but not the R324C mutation, likely partially conserves GS activity [20]. Here, we show by MD simulations and binding free energy calculations that both R324 mutations lead to a loss of direct interactions with the β-phosphate group of ADP or ATP compared to wild type GS, and weakened ATP binding. In the case of the R324S mutant, the direct interaction is replaced by water-mediated hydrogen bonds, which are significantly less frequently observed in the R324C mutant. MD simulations and binding free energy calculations demonstrate that the R341C mutation indirectly weakens ATP binding. In addition, rigidity analysis reveals that the R341C mutation particularly destabilizes helix H8, which should hamper glutamate binding to GS; in vitro studies provide evidence for this influence.
Initially, we established a model system consisting of two adjacent subunits of the GS decamer forming a single catalytic site for performing MD simulations in a computationally efficient way. During 100 ns of MD simulations, small structural deviations (RMSD < 1.7 Å; Fig 1D) from the starting structure were found for the core regions of both the dimeric model and the decamer reference. Likewise, structural deviations of the binding site region and the bound ligands computed for the dimeric model were at the lower end (RMSD = 0.89 Å (Fig 1F), 0.49 Å (Fig 1G), and 0.96 Å (Fig 1H), respectively) of the range of values found for the decamer (RMSD < 2.3 Å (Fig 1F) and < 0.7 Å (Fig 1G) or < 0.7 Å (Fig 1H), respectively). Thus, no gross conformational changes were observed for the dimeric model despite the lack of interactions to neighboring subunits. These results are in agreement with a crystallographic study that shows no major allosteric changes within or between the pentamers of human GS upon MSO binding as well as only small (RMSD < 0.35 Å) structural differences between canine GS in the apo state and human GS in ligand-bound states [22]. Similarly, only small structural alterations in catalytic site loops during catalysis have been reported for prokaryotic GS of class I-β [21, 24]. Regarding the divalent metal ions required by eukaryotic GS for activity [35], we considered Mg2+ ions in our dimeric model system rather than Mn2+ ions. We did so as only 20–30% of GS subunits from ovine brain tissue have been found trapped with Mn2+ under physiological Mg2+ and Mn2+ concentrations [76], and human GS from brain is 10-fold more active with Mg2+ bound than with Mn2+ [77]. For investigating the effects of the R324S, R324C, and R314C mutations, three independent MD simulations were performed for each of the four states of GS (GSAPO, GSATP, GSATP+GLU, GSADP+GGP) of wild type GS and the three mutants, respectively. The replicate MD simulations allow probing for the influence of the starting conditions and determining the statistical significance of the computed results [78].
Regarding the R324S and R324C mutations, we first tested if they distort the structure of the catalytic site in the GSAPO state. We excluded residues of the Glu flap loop (residues 304–306) from the analysis because this loop was identified to be highly mobile in a previous study [24], which is in line with our results (Fig 1E). For both mutants, the mean backbone RMSD of the residues of the catalytic site is ≤ 1.26 Å with respect to the starting structure (Fig 2A). These conformational changes are only slightly larger than those found in the MD simulations of the dimeric model of wild type GS (RMSD = 0.89 Å (Fig 1F)), which demonstrates that neither mutation changes the catalytic site markedly. In contrast, a significant difference between the wild type residue R324 and serine or cysteine at position 324 is found with respect to interactions with the β-phosphate group of bound ADP or ATP: R324 forms salt bridge interactions with the substrates, respectively (Fig 2B), as also observed in the crystal structure of human GS complexed with ADP [22]; however, neither do S324 nor C324 interact directly with ADP or ATP (Fig 2B). This interaction loss does not result in a significant change of the ATP mobility within the catalytic site of the R324S and R324C mutants compared to wild type GS, likely reflecting the structure-stabilizing influence of the Mg2+ ion at the metal binding site n2, which interacts with the nucleotide [24, 79]. Yet, the mutations result in effective energies for binding of ATP that are disfavorable by about 4 kcal mol-1 with respect to wild type GS (Fig 6B). Neglecting contributions due to changes in the configurational entropy of the solute molecules upon binding [53, 59, 61], this change in the effective binding energies is equivalent to an about 1000-fold lower association constant of ATP. Taken together, these results strongly suggest that the R324S and R324C mutations deteriorate GS catalytic activity by weakening ATP binding in the first step of the catalytic cycle.
The analysis of indirect, water-mediated interactions between R324S or R342C and the β- and γ-phosphate groups of ATP, respectively, suggested that the extent by which the R324S mutation weakens ATP binding is smaller than the effect of the R324C mutation. This suggestion was based on three independent but related analyses. First, the visual inspection of computed water density between residue 324 and the β- and γ-phosphate groups of ATP in MD trajectories of the GSATP+GLU state revealed a much broader region of high water density between R324S and ATP than between R324C and ATP (Fig 2C and 2D). Second, RDFs for oxygen atoms of water molecules around the sidechain oxygen of R324S or sulfur of R324C, respectively, demonstrated a 30% higher water density at the first shell peak for R324S than for R324C, and a considerably more structured second shell in the case of R324S (Fig 2E and 2F). Third, the frequency of occurrence of water-mediated hydrogen bonds between residue 324 and the β- and γ-phosphate groups of ATP is significantly higher for R324S than for R324C for both weak and strong hydrogen bonds (Fig 2G and 2H). Water at the interface of biomolecular complexes can provide increased ligand or substrate affinity [76, 79, 80] due to it mediating the interaction between polar groups via hydrogen bonds and/or filling space providing van der Waals interactions [81]. The higher water density, more ordered water structure, and increased number of hydrogen bonds found for R324S compared to R324C is in line with findings that more hydrophilic groups generally lead to a higher number of water molecules at a binding interface [82] and that hydrogen bonds involving sulfur are more elusive and weaker than those involving oxygen [83–86], resulting in a larger hydrogen bonding potential of serine than of cysteine in proteins [83]. Our finding of significantly more frequent water-mediated hydrogen bonds between R324S and the β- and γ-phosphate groups of ATP can thus provide an explanation for the suggestion that GS residual activity is higher in the R324S mutant than in the R324C mutant [20]. This residual activity has been linked with the survival of the now seven-year old boy [20]. As this boy still suffers from hyperammonemia [19], the finding could also stimulate the development of better ATP binding-enhancing molecules by which the R324S mutant can be “repaired” extrinsically [87].
The wild type residue R341 is part of an amino acid triplet located on a loop region including also D339 and R340 [22]. While the sidechain of R341 points away from the catalytic site, R340 points to it (Fig 3A). In the course of our MD simulations, we observed a persistent hydrogen bond of the guanidino function of R340 to GGP in the GSADP+GGP state (Fig 3B) as also found in the crystal structure [22]. In contrast, for the state GSATP+GLU, we observed a shift of the side chain position of R340 during the MD simulations, leading to a salt-bridge interaction with the γ-phosphate group of ATP (Fig 3B). The loss of salt bridge interactions due to the R341C mutation, observed between R341 and D339 in wild type GS (Fig 3C and 3D), results in a destabilization of R340 in the GSAPO state, as manifested by significantly smaller RMSF of that residue in wild type GS than the R341C mutant (Fig 3E). No such changes in R340 mobility were observed for the other three GS states. Hence, this strongly suggests that the R341C mutation indirectly weakens ATP binding in the first step of the catalytic cycle. This suggestion was corroborated by the computed relative effective energy of binding of ATP, which is about 6 kcal·mol-1 (equivalent to an approximately 10000-fold lower association constant) less favorable for the R341C mutant than for wild type GS (Fig 6B).
An even more indirect, long-range effect of the R341C mutation on the catalytic efficiency of GS was revealed starting from rigidity theory-based analyses [26, 71] of the structural stability of the GSATP state. These analyses demonstrated that the loss of side chain interactions in the R341C mutant, particularly to H281, H284, and Y288 on H8, result in a significant destabilization of the C-terminus of GS compared to wild type (Fig 3F). These results were confirmed by the analysis of secondary structure of H8 during MD simulations of wild type GS, the R341C mutant, and the triple alanine mutant of H281, H284, and Y288 (HHY) considered to act as a mimic of the R341C mutant: For the latter two cases, a marked increase in loop probability was found for the central region of H8 (Fig 3G). Together with the above results on R341’s role in the triad, this suggests that R341 can relay stability information between the catalytic site and H8. Remarkably, these results can provide an explanation at the atomistic level as to why addition of ATP to human GS increases the melting temperature of the enzyme by 11.5°C [22].
From comparative crystal structure analysis, a connection between ATP binding in the first step of the catalytic cycle of GS and a shift of H8 that closes GS’ catalytic site was recognized; this shift is considered a prerequisite for glutamate binding and should be hampered in the R341C or HHY mutants [22], which should weaken glutamate binding as suggested by the computed relative effective energy of binding of about 2.2 kcal·mol-1 (equivalent to an approximately 40-fold lower association constant) (Fig 6B). We therefore tested the role of amino acids H281, H284, Y288 in human GS for substrate binding to the glutamate binding pocket in HEK293 cells overexpressing HHY-mutated YFP-tagged GS using MSO as a substrate (Fig 4C, 4D and 4E).
MSO binds non-covalently to the glutamate binding pocket of GS and irreversibly inactivates the enzyme [73] thereby masking an epitope which was recently shown to be recognized by a monoclonal antibody raised against GS [62]. In line with this, MSO-treatment strongly reduced anti-GS immunoreactivity in wildtype-YFP-GS transfected HEK293 cells compared to untreated controls (Fig 5A). In contrast, anti-GS immunoreactivity was unchanged in MSO-treated HEK293-cells transfected with HHY-mutated GS, indicating impaired substrate binding to the catalytic center and suggesting loss of enzymatic activity. Mutation of neighboring amino acids S278, K279, and R280, which do not interact with R341 and were suggested to not play a role for maintenance of the tertiary structure of the catalytic center, did not prevent MSO-binding to GS as indicated by reduced anti-GS immunoreactivity (Fig 5A). Reduced anti-GS immunoreactivity in MSO-treated HEK293 cells was not due to MSO-mediated downregulation of YFP-GS in the respective transfected HEK293 cells (Fig 4C).
In order to validate that MSO masks an epitope recognized by the monoclonal anti-GS antibody, protein samples were denaturated, which was shown to release MSO from GS [73] and to recover recognition of GS by the anti-GS antibody [62]. Indeed, this treatment completely restored anti-GS immunoreactivity in MSO-treated HEK293 cells transfected with wild type- and SKR-mutated GS as shown by dot-blot analysis (Fig 5B). In line with this, anti-GS immunoreactivity was unchanged in MSO-treated GS-transfected HEK293 cells compared to untreated controls when analysed by Western-blot using heat- and detergent-treated protein samples (Fig 4D).
Binding of ATP to GS is a prerequisite for accessibility of the active center for glutamate [23, 24]. Thus, impaired glutamate binding as indicated by unchanged anti-GS immunoreactivity in MSO-treated HEK293 cells transfected with HHY-mutated GS (Fig 5A) may be a consequence of impaired binding of ATP. This hypothesis was verified by precipitating ATP-binding proteins from HEK293 cells transfected with wildtype- or HHY-mutated GS by N6-(6-aminohexyl)-ATP-agarose and anti-GS Western-blot analysis. Both endogenous and overexpressed wildtype- or HHY-mutated YFP-GS were detected in precipitates of ATP-binding proteins from HEK293 cell lysates (Figure G in S1 file). Almost no anti-GS immunoreactivity was detected when ATP-binding proteins were precipitated in the presence of excess ATP (10 mmol/l; Figure G in S1 file). These data suggest that mutating amino acids H281, H284, and Y288 of GS does not affect the binding of ATP to GS but that of glutamate.
The present findings also confirm earlier results [17] showing that arginine 341 is critical for recognition by a monoclonal antibody (BD, clone6) raised against GS (Fig 4D). In line with this, mutating amino acids H281, H284 and Y288 or S278, K279 and R280 had no impact on GS-recognition by the monoclonal antibody (Fig 4D).
These results suggest that amino acids H281, H284 and Y288 on helix H8 in human GS may stabilize the tertiary structure of the glutamate-binding site through interaction with R341, that way enabling glutamate binding to the catalytic center.
In summary, we identified the molecular mechanisms of the GS mutations R324S, R324C, and R341C that lead to clinically relevant glutamine deficiency. All three mutants are suggested to influence the first step of GS’ catalytic cycle, namely ATP or glutamate binding. Our analyses reveal a complex set of effects including the loss of direct interactions to substrates, the involvement of water-mediated interactions that alleviate part of the mutation effect, and long-range effects between the catalytic site and structural parts distant from it. As to the latter, dot-blot analysis of HEK293 cells overexpressing GS is in line with our prediction of a significant destabilization of helix H8 in the R341C mutant, which should negatively affect glutamate binding.
|
10.1371/journal.pbio.1000495 | Genomic Fossils Calibrate the Long-Term Evolution of Hepadnaviruses | Because most extant viruses mutate rapidly and lack a true fossil record, their deep evolution and long-term substitution rates remain poorly understood. In addition to retroviruses, which rely on chromosomal integration for their replication, many other viruses replicate in the nucleus of their host's cells and are therefore prone to endogenization, a process that involves integration of viral DNA into the host's germline genome followed by long-term vertical inheritance. Such endogenous viruses are highly valuable as they provide a molecular fossil record of past viral invasions, which may be used to decipher the origins and long-term evolutionary characteristics of modern pathogenic viruses. Hepadnaviruses (Hepadnaviridae) are a family of small, partially double-stranded DNA viruses that include hepatitis B viruses. Here we report the discovery of endogenous hepadnaviruses in the genome of the zebra finch. We used a combination of cross-species analysis of orthologous insertions, molecular dating, and phylogenetic analyses to demonstrate that hepadnaviruses infiltrated repeatedly the germline genome of passerine birds. We provide evidence that some of the avian hepadnavirus integration events are at least 19 My old, which reveals a much deeper ancestry of Hepadnaviridae than could be inferred based on the coalescence times of modern hepadnaviruses. Furthermore, the remarkable sequence similarity between endogenous and extant avian hepadnaviruses (up to 75% identity) suggests that long-term substitution rates for these viruses are on the order of 10−8 substitutions per site per year, which is a 1,000-fold slower than short-term rates estimated based on the sequences of circulating hepadnaviruses. Together, these results imply a drastic shift in our understanding of the time scale of hepadnavirus evolution, and suggest that the rapid evolutionary dynamics characterizing modern avian hepadnaviruses do not reflect their mode of evolution on a deep time scale.
| Paleovirology is the study of ancient viruses and the way they have shaped the innate immune system of their hosts over millions of years. One way to reconstruct the deep evolution of viruses is to search for viral sequences “fossilized” at different evolutionary time points in the genome of their hosts. Besides retroviruses, few virus families are known to have deposited molecular relics in their host's genomes. Here we report on the discovery of multiple fragments of viruses belonging to the Hepadnaviridae family (which includes the human hepatitis B viruses) fossilized in the genome of the zebra finch. We show that some of these fragments infiltrated the germline genome of passerine birds more than 19 million years ago, which implies that hepadnaviruses are much older than previously thought. Based on this age, we can infer a long-term avian hepadnavirus substitution rate, which is a 1,000-fold slower than all short-term substitution rates calculated based on extant hepadnavirus sequences. These results call for a reevaluation of the long-term evolution of Hepadnaviridae, and indicate that some exogenous hepadnaviruses may still be circulating today in various passerine birds.
| Most viruses are characterized by high substitution rates, which generally prevent reconstruction of their long-term evolutionary history [1]. Consequently, the origins and age of most extant viruses remain elusive [2]. One solution to this conundrum lies in the advent of paleovirology, the study of paleoviruses and the way they have shaped the antiviral genes of their hosts over millions of years [3]. Although viruses lack a true geological fossil record, some have left footprints of their evolution in their hosts' genome. For example, vertebrate retroviruses are RNA viruses that normally integrate into the genome of their host's somatic cells as part of their replication cycle. On occasion, these viruses may integrate into the germline genome of their host, and become inactive and vertically inherited over millions of years. Their molecular relics, called endogenous retroviruses, now make up a substantial fraction of vertebrate genomes (∼8% in human; [4]).
While retroviruses account for the major fraction of known viral genomic fossils, various other viruses that do not normally integrate into the genome but replicate in the nucleus of the host cell are susceptible to fortuitous chromosomal integration. For example, pararetroviruses (double-stranded DNA) have deposited numerous endogenous copies in the genome of several plant species [5], and singular integration events have been reported for gemini-like viruses (single-stranded DNA) in tobacco [6], and for non-retroviral RNA viruses such as totovirus-like and M2-killer-like viruses in fungi (double-stranded RNA; [7],[8]) and flaviviruses in mosquitoes [9],[10].
Genomic fossils closely related to modern viral groups are of particular interest as they have the potential to unveil otherwise inaccessible features of the long-term evolution of viruses. A handful of such precious paleoviruses have recently been unearthed from mammalian genomes. Among these, two ancient lentiviruses (RELIK in rabbit [11] and pSIV in primates [12],[13]) and one foamy virus (SloEFV in xenarthrans [14]) revealed that the history of these two retroviral genera can be rooted on a deep time scale, challenging earlier views on retroviral evolution based on comparisons of extant viral genomes. Likewise, the recent discovery of multiple endogenous bornaviruses and filoviruses in diverse mammals showed that these single-stranded RNA viruses were able to infiltrate repeatedly the germline of distant mammalian species over at least the past 40 My [15]–[17].
Hepadnaviridae (including hepatitis B viruses [HBVs]) are compact (∼3,000 bp), partially double-stranded circular DNA viruses infecting various mammal and bird species and responsible for ∼600,000 human deaths of acute or chronic liver disease per year [18]. While replication of these viruses does not rely on integration into the host genome, a relatively large number of chromosomal integration events have been characterized in mammalian liver cells sampled from chronically infected individuals [19]. In this study, we show that hepadnaviruses have also infiltrated the germline genome of some of their vertebrate hosts in the distant past. The viral sequences fossilized since these endogenization events offer an unprecedented opportunity to reevaluate the mode and tempo of Hepadnaviridae evolution.
TBLASTN searches using the duck HBV (DHBV) proteins on all available genomes in GenBank yielded 15 hepadnavirus-like fragments (collectively called endogenous zebra finch HBVs [eZHBVs]). These sequences are interspersed into ten different chromosomes of the zebra finch (Taeniopygia guttata, Estrildidae) and show between 55% and 75% nucleotide similarity to the DHBV genome (Figure 1; Table 1; Dataset S1). Most of these fragments contain one or more mutations compromising their coding capacity, which suggests that they have evolved under no functional constraint since integration. Together, the 15 eZHBV segments cover ∼70% of the DHBV genome, which is structurally representative of all hepadnaviruses [20] (Figure 1). eZHBVs tend to map within two loosely defined regions of DHBV, one encompassing the core and polymerase N-terminal domains (eZHBVc–eZHBVi; group 1), and one overlapping with the preS/S and polymerase C-terminal domains (eZHBVj–eZHBVn; group 2). In addition, two eZHBVs (eZHBVa and eZHBVb) map to other regions of the core domain (Figure 1). eZHBVl and eZHBVl* (both located on Chromosome 20) map to the same region of the DHBV genome and are highly similar (97% over 537 bp). Similar levels of identity are observed between their flanking genomic regions: 96.7% identity over 637 bp in the 5′ flanking region and 97% identity over 534 bp in the 3′ flanking region. These observations suggest that one insertion most likely derives from the other through intrachromosomal duplication of a genomic fragment including the initial eZHBV insertion along with its flanking regions.
In order to assess the phylogenetic relationship among eZHBVs and hepadnaviruses, we conducted phylogenetic analyses of amino acid alignments including extant hepadnaviruses and group 1 (106 amino acids) and group 2 (293 amino acids) eZHBVs. The results show that in both phylogenies (Figure 2A and 2B) hepadnaviruses can be divided into two clusters, one grouping eZHBVs and extant avian hepadnaviruses and the other including all mammalian hepadnaviruses. Within the former cluster, eZHBVs are consistently more distant from extant avian hepadnaviruses than these are from each other. While group 1 eZHBVs form a monophyletic group (Figure 2A), there is no statistically supported clustering of group 2 eZHBVs with each other (Figure 2B). The only exception is the close clustering of eZHBVl and eZHBVl*, which likely reflects their relatively recent origin by duplication rather than as independent insertions (see above).
A first minimal estimate of the age of eZHBVs can be derived indirectly from the time at which the duplication yielding eZHBVl and eZHBVl* occurred, which must postdate the chromosomal integration of the ancestral eZHBVl element. The distance between these duplicates is 0.03 (Table 2). To our knowledge, the most comprehensive estimate of neutral nuclear substitution rates available for birds, calculated based on a comparison of multiple intron sequences between chicken and turkey, was found to range between 2×10−9 and 3.9×10−9 substitutions per site per year (subs/site/year) [21], values similar to the range of those estimated for mammals (2.2×10−9 to 4.5×10−9 subs/site/year; [22],[23]). The avian rates are based on a fossil calibration of the split between Anatidae and Anhimidae at 55 My [21],[24],[25]. Dividing half of the distance between eZHBVl duplicates (0.015) by the bird neutral substitution rates yields a duplication time ranging between 3.8 My (with 3.9×10−9 subs/site/year) and 7.5 My (with 2×10−9 subs/site/year). The timing of this duplication provides a minimal estimate for the integration of the ancestral eZHBV fragment.
A more direct way to estimate the age of eZHBVs is to use a phylogenetic approach, reasoning that if an insertion is shared by two species at the same (orthologous) locus, the integration event must be at least as old as the last common ancestor of the two species. It is important to note that the analysis of a large number of chromosomal integrants of HBV in somatic mammalian cells has revealed no preference for insertion in a specific sequence motif (e.g., [19],[26]). Thus, the possibility that two identical viral fragments would integrate at the exact same genomic position (i.e., between the same two nucleotides) independently in multiple species is extremely unlikely. Using PCR primers designed on the genomic regions flanking three eZHBVs, we were able to amplify two orthologous insertions (eZHBVa and eZHBVl) in three other species of estrildid finches (black throated finch [Poephila cincta], scaly breasted munia [Lonchura punctulata], and gouldian finch [Chloebia gouldiae]) and in the dark-eyed junco (Junco hyemalis), a non-estrildid passerine bird belonging to the Emberizidae family (Figure 3). We also obtained a positive PCR product for eZHBVj in the three estrildid finches, and were able to amplify the empty site orthologous to eZHBVa in the olive sunbird (Cyanomitra olivaceus, Nectariniidae) (Figures 3 and 4). The identity of all the eZHBV fragments amplified by PCR was confirmed by DNA sequencing (Datasets S4, S5, S6). This revealed that each orthologous eZHBV is present at the same chromosomal position in all species where it could be amplified. Furthermore, in all three cases, the phylogenetic relationships between orthologous eZHBVs reflect the phylogenetic relationships of the bird species (Figure 3). Together, these data strongly suggest that each of these three insertions descend from an ancestral integration event that occurred prior to the split of the different bird species.
The most recent molecular phylogenetic analyses divide finches and their allies into two major monophyletic clades, one consisting of African and Australasian estrildid finches and weavers, and the other grouping American emberizid sparrows (including the dark-eyed junco) together with fringillid finches and Old World sparrows [27]. Within Estrildidae, the gouldian finch is sister to a clade grouping the scaly breasted munia and finches of the genera Poephila (black throated finch) and Taeniopygia (zebra finch) (Figures 3A and 4; [28]). The congruence between these relationships and the phylogenies of orthologous eZHBVa and eZHBVl (Figure 3) indicates that the two eZHBVs result from two independent germline integration events of hepadnavirus-like sequences in a common ancestor of Estrildidae and Emberizidae, and that eZHBVa was inserted after the divergence of the Nectariniidae lineage. The divergence time between Estrildidae and Emberizidae has been estimated at 25 My based on relaxed molecular clock analyses of rag1 and rag2 nuclear genes using a paleobiogeographical calibration of 82 My for the split between Acanthisittidae and other passerine birds [29],[30]. The same analysis yielded an age of 35 My for the most recent common ancestor of Nectariniidae and Estrildidae. These dates would place the origin of eZHBVl prior to 25 My, and that of eZHBVa between 35 and 25 My ago.
Our last estimate of the age of eZHBVs relies on the level of sequence divergence between orthologous eZHBV sequences. The corrected distances inferred for orthologous eZHBVa (222 bp) and eZHBVl (238 bp) are 0.15 and 0.16 respectively between the zebra finch and the dark-eyed junco (see Materials and Methods). Selection analyses on these two fragments did not reveal any sign of positive or purifying selection (see Materials and Methods), suggesting that eZHBVa and eZHBVl have evolved under no functional constraint since their chromosomal integration in the common ancestor of these two birds, thereby accumulating substitutions at the neutral rate of these species. Applying the above-mentioned bird neutral substitution rates to half of the zebra finch/junco distances for eZHBVa and eZHBVl yielded integration times ranging between 40 My (with the eZHBVl distance of 0.08 and a rate of 2×10−9 subs/site/year) and 19.2 My (with the eZHBVa distance of 0.075 and a rate of 3.9×10−9 subs/site/year).
While our estimates of the age of eZHBVs are based on two different calibration points located at distant phylogenetic positions within the avian tree (55 My for the split between Anatidae and Anhimidae, or 82 My for the split between Acanthisittidae and other passerine birds), both approaches yield dates that largely overlap (40–19.2 My and 35–25 My). This suggests that eZHBVa and eZHBVl are at least 19 My old (and may be as much as 40 My old), which implies that the origin of avian hepadnaviruses as a whole (including extant and extinct viral lineages) is much deeper than the origin of currently circulating avian hepadnaviruses (time to most recent common ancestor <6,000 y; [31]).
Because eZHBVa and eZHBVl are at least 19 My old, the total genetic distance between these fragments and extant bird hepadnaviruses is expected to correspond to the sum of (i) the distance accumulated over the past 19 My at the bird neutral substitution rate (A in Figure 4), which can be approximated as half the distance between orthologous junco/zebra finch eZHBVa (0.075) or eZHBVl (0.08), (ii) the distance accumulated at the viral rate during the same period (D in Figure 4), and (iii) the distance accumulated at the viral rate between the time at which extant avian hepadnaviruses and eZHBVs diverged and the time of eZHBV endogenization (e.g., C+B for eZHBVl in Figure 4). The average corrected distance between eZHBVl and extant avian hepadnaviruses after subtracting the distance accumulated during 19 My at the bird rate (0.08) is 0.41 (range = 0.39–0.45). For eZHBVa, this distance is 1.3 (range = 1.15–1.5). Dividing these distances by 19 My yields average estimates of long-term substitution rates of 2.15×10−8 subs/site/year for eZHBVl and 6.8×10−8 subs/site/year for eZHBVa. Note that these values are likely to be overestimates as the distance between the time at which extant avian hepadnaviruses and eZHBVs diverged and the time of eZHBV endogenization is unknown (C+B for eZHBVl and F+E for eZHBVa, Figure 4), and therefore could not be subtracted from the total extant avian hepadnaviruses/eZHBV distance.
In this study we provide evidence that the germline genome of passerine birds has been infiltrated by several ancient and diverse hepadnaviruses that still show surprisingly high levels of similarity to extant avian hepadnaviruses. Although eZHBVs represent, to our knowledge, the first instance of endogenous DNA viruses reported in animals, several characteristics of hepadnaviruses suggest that endogenization of these viruses may be likely. Hepadnaviruses replicate in the nucleus of their host's cells via a reverse-transcribed RNA intermediate [32],[33]. Part of their life cycle is therefore spent in close proximity to the host DNA, which may facilitate chromosomal integration via various host- or transposable-element-mediated mechanisms that use either DNA or RNA templates (e.g., [15]). Indeed, although integration into the host genome is not required for the replication of the virus, integrated HBV genomic fragments are commonly observed in liver cells of individuals persistently infected, where they tend to be associated with hepatocarcinoma [19]. In addition, while hepadnavirus replication is thought to occur mainly in hepatocytes, its tropism may extend to other tissue and cell types, including germ cells. For example, avian hepadnavirus replication has been shown to occur in the yolk sac of developing duck embryos [34]. Typically, large quantities of viral particles circulate in the blood during HBV infection [35]. These particles have the capacity to tightly bind to many different cell types [35], and there is evidence supporting the presence of HBV DNA in spermatozoa and ovaries as well as the chromosomal integration of HBV in spermatozoa [36]–[38]. Based on these data, infiltration of the germline genome by hepadnaviruses followed by long-term vertical inheritance appears largely plausible. Thus, it is likely that other endogenous hepadnaviruses await discovery in other birds and perhaps also in mammalian genomes.
The precise mechanisms underlying the chromosomal integration of HBV remain unclear [19]. One model supported by experimental evidence posits that viral linear double-stranded DNA resulting from aberrant replication can be integrated during repair of double strand breaks via non-homologous end joining [39]. As the 3′ extremity of eZHBVj (position 2521) and eZHBVk (position 2512) map to a region of the DHBV genome that corresponds to the predicted end of a typical linear HBV precursor [40], the structure of these two fragments is potentially consistent with integration via non-homologous end joining. We also note that the extremities of several other fragments map to fairly narrow regions of the viral genome (e.g., same 5′ position for eZHBVd and eZHBVe; Figure 1), which may reflect the presence of breakpoint hotspots in the viral genomes that gave rise to eZHBVs. Finally, while the zebra finch genome contains several families of long terminal repeat (LTR) and non-LTR retrotransposons [41] whose enzymatic machinery could have potentially promoted the chromosomal integration of eZHBVs, none of the insertions examined were terminated by a poly-A tail or flanked by direct repeats, as would be expected if they had occurred through retrotransposition [15].
An intriguing question is whether the multiple eZHBVs result from endogenization events that took place during a short period of time or whether they were assimilated at widely different times over (at least) the past 19 My. Hepadnaviruses do not encode an integrase, and chromosomal integrants generally correspond to truncated genomes (as observed here). Thus, unlike retroviruses, integrated HBV fragments cannot in principle replicate further through intragenomic transposition or reinfection, and as such they can be considered essentially “dead on arrival.” With this in mind, we contend that eZHBVs are likely to result from multiple independent episodes of germline infiltrations that took place on a deep time scale, possibly spanning several millions of years, and involving distantly related hepadnaviruses. This inference is supported by the large distances observed between eZHBVs (Tables 2 and 3). Specifically, all pairwise distances involving eZHBVi and those between eZHBVl, eZHBVj, eZHBVk, and eZHBVn are more than 2-fold higher than the average distance separating extant avian HBVs, even when subtracting an approximate distance accumulated at the bird genome rate since integration (distances in bold in Tables 2 and 3). Together with the long branches leading to eZHBVs in the hepadnavirus tree (Figure 2), these data strongly suggest that diverse hepadnaviruses (at least five based on the distance threshold described above) have been circulating in birds for several million years. More specifically, we believe that the large inter-eZHBV distances likely reflect the fact that eZHBVs stem from viruses that were already deeply divergent at the time of integration, and/or that eZHBVs were integrated at time points separated by several million years over at least 19 My. A third, non-mutually exclusive explanation for these large distances is that the evolution of the hepadnavirus genome may be subject to strong mutational saturation (see also below). Considering that these viruses have crossed species boundaries repeatedly over the past 6,000 y [20],[31],[42], we speculate that a wide range of bird species may have been, and may still be, infected by hepadnaviruses. It would be interesting to explore whether hepadnaviruses are still circulating in extant estrildid finches such as the zebra finch. Such a discovery would provide a powerful system to study the virus and its potential association with hepatocarcinoma in a model bird species with a complete genome sequence [41].
Various calculations of HBV substitution rates based on comparison of extant viruses have produced broadly similar estimates, ranging from 7.72×10−4 to 7.9×10−5 subs/site/year [31],[43]–[47]. Surprisingly, we infer long-term substitution rates that are more than three orders of magnitude slower than these short-term rates. It is important to note that while eZHBVs evolved at the bird genome rate since their integration, this cannot explain the slowdown in long-term rates inferred in this study as the distance accumulated at the bird rate (A in Figure 4) was removed from our calculation of long-term hepadnavirus rates. Our estimates (2.15×10−8 to 6.8×10−8 subs/site/year) therefore represent a range of rates under which avian hepadnaviruses have evolved from the time just preceding the integration of eZHBVa and eZHBVl in the bird genome (∼19 My ago) to the time at which circulating avian hepadnavirus genomes were sequenced (the last two decades).
Gibbs et al. [48] recently suggested that viral evolutionary rates may vary dramatically depending on the time scale on which they are measured. The main line of evidence supporting this view was that rates inferred from serially or heterochronously sampled sequences are invariably more than two orders of magnitude higher than those calculated when assuming viruses have co-diverged with, and are therefore as old as, their hosts. In most cases, however, the hypothesis of host/virus co-divergence is only indirectly supported by the seemingly strong host specificity of the virus, and/or the apparent topological congruence (often not formally tested) between host and virus phylogenies. A major pitfall in this reasoning is that processes other than co-divergence may explain congruent phylogenies between hosts and viruses [49]–[51]. Given the potential caveats associated with the hypothesis of host/virus co-divergence, it is important to emphasize that our results do not rely on this assumption. Rather, they are based on a direct measure of the distance separating extant hepadnaviruses from extinct ones that are at least 19 My old.
How can we explain the apparent major disparity between short- and long-term substitution rates of hepadnaviruses? The rate of nucleotide substitution in any system depends on the background mutation rate, the rate of replication, and the rate of fixation. Hepadnaviruses replicate their genome via an RNA intermediate using a reverse transcriptase (RT). While to our knowledge there is no precise measure of the fidelity of the hepadnavirus RT, this enzyme lacks a proofreading activity and is known to be highly error prone in all retroviruses and other retroelements for which an error rate has been estimated [52],[53]. Up to 20-fold variations in RT error rates have been reported between different families of retroviruses [52]. It is therefore conceivable that variations in the fidelity of the enzyme (i.e., background mutation rate) over time might explain some of the difference between short- and long-term hepadnavirus substitution rates. However, slow long-term substitution rates similar to those reported here have been inferred for mammalian foamy viruses (1.7×10−8 subs/site/year) and human T cell lymphotropic virus type II (1.091×10−7 to 7.118×10−7 subs/site/year), two mammalian retroviruses that yet replicate via a highly error-prone RT [54],[55]. In those cases, it is thought that both viruses evolve slowly because they are non-pathogenic and replicate mainly as integrated proviruses, using the high-fidelity DNA polymerases of their hosts [56],[57]. These two examples therefore suggest that even in the presence of a high background mutation rate, viruses can evolve slowly if their replication rate is reduced. By analogy, it could be that hepadnaviruses have been characterized by low levels of pathogenicity and by low rates of replication for most of their evolutionary history. In this context, the high substitution rates and epidemiological dynamics currently associated with circulating hepadnaviruses might reflect recent drastic alterations in the biology of these viruses and of the selective pressures acting on them.
Another major process that may be responsible for the time dependency of substitution rates suggested by this study is purifying selection, as proposed for cellular organisms (e.g., [58]–[60]; see [61] for discussion). About 60% of the HBV genome codes for at least two overlapping open reading frames and therefore contains very few synonymous sites. Consistent with this, it was shown that nonoverlapping regions of the HBV genome evolve faster than overlapping regions [31],[62]. This tightly constrained genetic organization, combined with the intrinsically low fidelity of the RT, suggests that the effect of purifying selection on long-term rates may be more pronounced for hepadnaviruses than for other viruses and for cellular organisms. Lastly, the high background mutation rates of hepadnaviruses may also result in strong mutational saturation (homoplasy and back mutations), which could also explain part of the difference between short- and long-term hepadnavirus substitution rates (see also above). While it is possible that saturation may in part hinder our ability to accurately infer the long-term hepadnavirus substitution rates, we believe that this phenomenon alone cannot explain the 1,000-fold difference between short- and long-term substitution rates. Because our knowledge on the deep evolution of extant viruses remains fragmentary and because many factors may influence substitution rates and their variation over time [1],[63], it would be necessary to revisit these questions when more fossil and modern hepadnavirus sequences become available.
In order to screen for the presence or absence of orthologous eZHBVs in several species of passerine birds (Table S1), we designed PCR primers on the flanking regions of three insertions. The sequences produced using these primers were aligned and are provided, together with the sequence of the primers, in Datasets S5 (eZHBVl), S6 (eZHBVj), and S7 (eZHBVa). For eZHBVl, we used a forward primer (1978F) anchored in the 5′ flanking region (86 bp upstream of the insertion) in combination with a reverse primer (hfr1) anchored within eZHBVl, at position 239–257. For eZHBVj, we used a forward primer (8718F) anchored within eZHBVj at position 712–734 in combination with a reverse primer anchored in the 3′ flanking region (86 bp downstream of the insertion). For eZHBVa, we used a forward primer (Scn3b-F) anchored in the 5′ flanking region (768 bp upstream of the insertion in T. guttata) that corresponds to the fourth exon of a predicted gene homologous to human SCN3B. The reverse primer (Scn3b-R) was anchored in the 3′ flanking region (47 bp downstream of the insertion in T. guttata), corresponding to the third exon of the predicted scn3b gene.
The identity of the different bird species used in this study was verified by sequencing a 420-bp fragment of the mitochondrial NADH dehydrogenase subunit 2 (NADH2) gene (Figure S1) using the following primers: Fwd 5′–AGT CAT TTW GGS AGG AAT CCT G; Rev 5′–TTC CAY TTC TGA TTY CCA GAA G. Standard PCR conditions were as follows: 2 min at 94°C; 30 cycles of 1 min at 94°C, 30 s at 48–62°C, and 30 s to 2 min at 72°C. PCR mix was buffer (5×), 5 µl; MgCl2 (25 mM), 2 µl; dNTP (10 mM), 0.5 µl; primer 1 (10 µM), 1 µl; primer 2 (10 µM), 1 µl; Taq (GoTaq, Promega), 1.25 U; DNA, 30–100 ng; and H2O up to 25 µl. PCR products were directly sequenced on an ABI 3130XL sequencer (Applied Biosystems). All sequences produced in this study were submitted to GenBank (accession numbers HQ116564–HQ116583).
Analyses of selection were carried out on alignments of each set of orthologous insertions amplified in the various passerine birds (eZHBVl, eZHBVj, and eZHBVa; provided in Datasets S4, S5, and S6, respectively) using HyPhy [64]. We used the trees corresponding to each alignment as inferred in Figure 3. The nucleotide substitution model accomplishing the most accurate fit to the data was determined using the NucModelCompare.bf procedure: HKY85 for each of the three alignments. The MG94xHKY85_3x4 codon substitution model was then fitted to each alignment with global parameters and partition-based equilibrium frequencies. This yielded a global ω (non-synonymous substitutions/synonymous substitutions) ratio of 0.98 (confidence interval: 0.642323, 1.327), 0.66 (confidence interval: 0.44, 0.88), and 0.93 (confidence interval: 0.62, 1.24) for eZHBVl, eZHBVa, and eZHBVj respectively. Using a likelihood ratio test, the likelihood function states for each alignment were then compared to likelihood function states obtained using the same model/alignment/tree but enforcing ω = 1 (neutral evolution). This revealed no significant difference (p = 0.95 for eZHBVl, 0.16 for eZHBVa, and 0.81 for eZHBVj), suggesting that eZHBVl, eZHBVj, and eZHBVa are evolving neutrally. We further tested this by re-optimizing the likelihood function with local parameters (where each branch of the tree has its own parameters) and comparing the likelihood function state obtained when the non-synonymous substitution rate and the synonymous substitution rate can have their own value on each branch with the likelihood function state obtained when the non-synonymous substitution rate is forced to be equal to the synonymous substitution rate on each branch. Again, the likelihood ratio test revealed no significant difference (p = 0.61 for eZHBVl, 0.29 for eZHBVa, and 0.85 for eZHBVj), suggesting neutral evolution in all branches.
All distances were calculated under maximum likelihood settings in PAUP 4.0 [65], using models of nucleotide substitution chosen based on the Akaike Information Criterion in jModeltest [66]: TPM2uf+G for group 1 eZHBVs, TVM+G for group 2 eZHBVs and for the distance between eZHBVa and extant avian hepadnaviruses, TPM1 for the distances between passerine eZHBVa orthologs, and HKY for the distance between passerine eZHBVl orthologs.
In order to estimate whether eZHBVs result from multiple integrations of a few very similar viral strains during a narrow time frame or whether more divergent strains were endogenized at widely different times during the last 19 My, we compared inter-eZHBV distances to the average distances between extant avian hepadnaviruses. In this context, it is important to keep in mind that each pairwise inter-eZHBV distance as we observe them today results from (i) the distance accumulated at the viral rate during the time separating the endogenization of each two sequences being compared (corresponding to B+C+E+F if eZHBVl and eZHBVa are compared, for example; Figure 4) and (ii) the distance accumulated on each sequence at the bird neutral rate after endogenization (2×A in Figure 4). Several inter-eZHBV distances are more than 2-fold higher than the average distances between extant hepadnaviruses, i.e., more than 2×0.27 = 0.54 for the region corresponding to group 1 eZHBVs, and more than 2×0.19 = 0.38 for the region corresponding to group 2 eZHBVs (Tables 2 and 3). Notably, most of these high inter-eZHBV distances remain more than 2-fold higher than distances between extant hepadnaviruses even when subtracting a 0.16 distance, which corresponds to a conservatively high estimate of the distance accumulated at the bird genome rate assuming the two eZHBVs being compared were both integrated 19 My ago. The 0.16 estimate is based on the highest of the distances between dark-eyed junco and zebra finch orthologs (eZHBVl), i.e., 2×A in Figure 4.
Sequences were aligned by hand using BioEdit 7.0.5.3 [67], and ambiguous regions were removed. Bayesian and maximum likelihood phylogenetic analyses were carried out using MrBayes 3.1.2 [68] and PHYML 3.0 [69], respectively. Nucleotide and amino acid substitution models were chosen based on the Akaike Information Criterion in jModelTest 0.1 [66], MrModeltest 2.3 [70], and ProtTest 2.4 [71]. eZHBVs were aligned at the amino acid level with representative members of extant avian and mammalian hepadnaviruses and analyzed using the rtREV (group 1 eZHBVs) and LG+G+F (group 2 eZHBVs) models in PHYML and with a prior setting allowing model jumping between fixed-rate amino acid models in MrBayes. eZHBVa, eZHBVj, and eZHBVl orthologs were analyzed with the TPM2uf+G, TPM2uf+G, and TIM3+G models of nucleotide substitution, respectively, in PHYML and with the GTR+G, HKY+G, and GTR+G models, respectively, in MrBayes. In order to verify the identity of the bird specimens included in this study, we also analyzed an alignment of a fragment of NADH2 nucleotide sequence produced in this study, as well as GenBank NADH2 sequences available for these species and for representatives of the families Paridae, Corvidae, Pycnonotidae, Turdidae, and Phasianidae (Figure S1). This alignment was analyzed with the TPM2uf+G model in PHYML and with the HKY+I+G model in MrBayes. For maximum likelihood analyses, the robustness of the branches was evaluated by non-parametric bootstrap analyses involving 1,000 pseudoreplicates of the original matrix. Bayesian analyses were run for at least one million generations, or until the standard deviation of split frequencies between the two parallel runs dropped below 0.01. Then, 25% of the sampled trees were discarded before summarizing the trees. The sequences used for the phylogenetic analyses are provided in Datasets S2, S3, S4, S5, S6, S7.
|
10.1371/journal.pgen.1003359 | Ataxin1L Is a Regulator of HSC Function Highlighting the Utility of Cross-Tissue Comparisons for Gene Discovery | Hematopoietic stem cells (HSCs) are rare quiescent cells that continuously replenish the cellular components of the peripheral blood. Observing that the ataxia-associated gene Ataxin-1-like (Atxn1L) was highly expressed in HSCs, we examined its role in HSC function through in vitro and in vivo assays. Mice lacking Atxn1L had greater numbers of HSCs that regenerated the blood more quickly than their wild-type counterparts. Molecular analyses indicated Atxn1L null HSCs had gene expression changes that regulate a program consistent with their higher level of proliferation, suggesting that Atxn1L is a novel regulator of HSC quiescence. To determine if additional brain-associated genes were candidates for hematologic regulation, we examined genes encoding proteins from autism- and ataxia-associated protein–protein interaction networks for their representation in hematopoietic cell populations. The interactomes were found to be highly enriched for proteins encoded by genes specifically expressed in HSCs relative to their differentiated progeny. Our data suggest a heretofore unappreciated similarity between regulatory modules in the brain and HSCs, offering a new strategy for novel gene discovery in both systems.
| Our labs, working separately on brain function and blood stem cells, noticed that a particular gene involved in movement disorders was also expressed in the blood system. We discovered through bone marrow transplantation experiments that this gene, called Ataxin-1-like, normally plays a role in restricting the number of blood-forming stem cells; stem cells lacking this gene were more numerous and more active. We wondered if this brain-blood similarity would hold for a larger number of genes, so we used bioinformatics approaches to compare large datasets our labs had generated from each system. We found that a surprising number of genes implicated in autism and ataxia by molecular studies were also highly expressed in blood-forming stem cells. We suggest that such cross-system comparisons could be used more widely to discover genes with important functions in brain and blood, but also perhaps other systems.
| Lifelong blood production is sustained by a quiescent reserve of hematopoietic stem cells (HSCs), which have the capacity to generate both additional stem cells (self-renewal) and differentiated blood cells. The balance between self-renewal and differentiation is tightly regulated and also flexible, ensuring adequate blood production under a variety of conditions while also maintaining a stem cell pool. While knock-out (KO) mice have allowed the identification of a number of genes that influence this balance, the relative scarcity of HSCs in the bone marrow limits the application of some genome-wide technologies that would uncover additional critical players and the basic biology of their regulation.
In contrast to the active turnover of the hematopoietic system, the brain is relatively static; it is primarily composed of terminally differentiated neurons and glia, but also contains rare self-renewing stem cells. We knew from the literature that a number of genes that exhibit roles in neurogenesis and neuronal function also play a key role in hematopoiesis. For example, Gfi1 is critical for Purkinje cell function in the brain [1], as well as maintenance of hematopoietic stem cell function and myeloid development [2]. In addition, Scl/Tal1 is critical for HSC development and function [3] and also for normal brain development [4].
With these examples in mind, when Ataxin-1L, which has been implicated to play a role in neurological disease [5], [6], but has no known hematopoietic function, was highly expressed in a microarray from HSCs [7], we wanted to test whether it too played a role in both tissues. We thus tested its function in the hematopoietic system through in vitro and in vivo assays using Atxn-1L null mice. We discovered that Atxn-1L is a strong negative regulator of hematopoietic stem cells, as knock-out mice exhibit greater numbers of more active stem cells. These data, together with the literature examples above, led us to examine the brain-blood relationship in a systematic way using bioinformatics strategies. Here, we show that genes and proteins identified functionally or by computational approaches as relevant in the brain are also implicated in hematopoiesis by multiple criteria, supporting the value of cross-tissue comparisons for gene discovery.
Atxn1L is a paralog of ATXN1 (aka ATAXIN1) [5], [8], originally identified in humans as the gene mutated in Spinocerebellar ataxia type 1 (SCA1) [9], [10]. ATXN1 has a triplet repeat sequence that becomes expanded and pathogenic in SCA1 patients, resulting in progressive ataxia with age. Atxn1L expression is highly overlapping with that of Ataxin1, and the two genes are at least partially functionally redundant [5]. Double Atxn1/Atxn1L knockout (KO) mice have a set of severe phenotypes absent in either of the single KO mice and die shortly after birth [11]. The phenotypes of the double KO mice include hydrocephalus, omphalocele, and a lung alveolarization defect. Hematopoietic system phenotypes of the single or double KO mice have not been previously reported.
Here we focused on Atxn1L because of its high expression in the hematopoietic system. Having a mouse with a null allele for Ataxn1L in our lab [11] and little prior information about any potential function for this gene in the hematopoietic system, we proceeded to study its function in HSCs.
Atxn1L is expressed in multiple hematopoietic lineages, but most highly in the stem cells, with an expression level comparable to that of other key hematopoietic regulators such as Gfi1 and Scl/Tal1 (Figure 1). To determine whether Atxn1L plays a role in HSC function, we first examined complete blood counts of adult Atxn1L−/− mice and the proportions of myeloid and lymphoid cells in the peripheral blood. There were no significant differences from the numbers in their wild-type counterparts (data not shown). Similarly, bone marrow progenitor populations were present at normal frequency, however, there was a slight increase in the proportion of long-term HSCs (P = 0.047) (Figure 1).
Because Atxn1L is most highly expressed in HSCs, we next examined HSC function via bone marrow transplantation studies. We first carried out competitive whole bone marrow transplantation assays in which KO bone marrow was competed against WT bone marrow from syngeneic strains of mice that are distinguishable using the CD45.1 and CD45.2 allelic system (Figure 2A). Equal numbers (250×103) of Atxn1L−/− donor and WT competitor whole BM cells were transplanted into lethally irradiated recipients and their contribution to peripheral blood production was assessed at 4-week intervals. The contribution of Atxn1L−/− BM to peripheral blood regeneration four weeks after the transplant was equivalent to WT; however, over time, the contribution of Atxn1L−/− increased significantly such that 70–80% of the blood was derived from the bone marrow of Atxn1L KO mice, whereas the controls, in which WT bone marrow was competed with WT bone marrow, remained around 50% (Figure 2B). The difference in repopulation activity between WT and KO bone marrow at 16 weeks, when the majority of the blood cells are considered derived from stem cells within the transplanted bone marrow was highly significant (P<0.01).
Because the bone marrow of Atxn1L KO mice harbored a slightly higher proportion of phenotypically-defined HSCs (Figure 1), we wanted to determine whether the higher peripheral blood reconstitution activity of Atxn1L−/− bone marrow was simply due to a higher number of HSCs, or a higher inherent repopulating activity of mutant HSCs. To test this, we examined the repopulation ability of HSCs purified from KO and WT bone marrow. Twenty HSCs (side population (SP), c-Kit+ Sca1+ Lineage− (KLS) CD150+) were mixed with 250,000 WT whole bone marrow cells and transplanted into lethally irradiated recipient mice. We found that Atxn1L−/− HSCs were superior in regeneration of the hematopoietic system, providing nearly double the blood contribution at 16 weeks after transplant (P<0.05) (Figure 2C). When the bone marrow of these recipient mice was examined 16 weeks after transplantation, almost 80% of the HSCs were donor-derived in mice transplanted with Atxn1L null HSCs, compared to 50% in the WT control group (P<0.05) (Figure 2D). This significantly better reconstituting activity of purified Atxn1L−/− HSCs indicates that Atxn1L−/− null mice harbor more HSCs that are inherently more active.
Bone marrow or stem cell transplantation assesses the ability of HSCs to repopulate the bone marrow and differentiate, but the ability of stem cells to self-renew is most rigorously assessed by secondary transplantation. To examine whether the self-renewal capacity of Atxn1L−/− HSCs was also enhanced, we re-isolated HSCs from the bone marrow of primary transplant recipients and re-transplanted them with fresh competitor bone marrow into secondary recipients (Figure 2A). Again, the performance of the Atxn1L−/− HSCs was superior to WT HSCs (P<0.05), with nearly double the activity in this rigorous assay (Figure 2E).
Finally, since both primary and secondary HSC transplants rely on the phenotypic definition of HSCs, we carried out a limiting dilution assay to assess the presence of functional repopulating units in WT compared with Atxn1L−/− bone marrow. This assay does not rely on competition, and is considered the most exacting to observe relative functional HSC activity [12]. In mice transplanted with the lowest doses of WT bone marrow (10,000 cells), less than half of the mice were engrafted. In contrast, 8/9 animals transplanted with the same amount of bone marrow from KO mice were engrafted (Figure 2F, 2G). These data indicate that the frequency of repopulating units in Atxn1L−/− marrow is ∼1/4500 cells, while the frequency in WT marrow is ∼1/15,000 cells (the latter being in line with standard estimates).
Together, these data establish that Atxn1L−/− mice have enhanced HSC activity. The slightly higher frequencies of phenotypically defined HSCs in bone marrow can not account for the significantly higher functional HSC activity observed by bone marrow and stem cell transplantation experiments. Thus, Atxn1L appears to be a negative regulator of HSC function.
One known factor that contributes to repopulation efficiency after bone marrow transplantation is the ability of the HSCs and their progenitors to home to the bone marrow niche. Thus, we tested whether the superior activity of Atxn1L−/− HSCs could be attributed to enhanced bone marrow homing. We intravenously transplanted 30,000 purified hematopoietic progenitors (c-Kit+, Sca-1+ Lineage− (KSL) cells into lethally irradiated recipient mice, and sacrificed the recipients 18 hours later to examine the proportion of progenitors reaching the femurs and tibias using flow cytometry (Figure 3A). We found no difference between the number of WT and Atxn1L−/−cells that were able to home within 18 hours (Figure 3B, 3C). In addition, we plated a portion of the bone marrow extracted from these recipients into methylcellulose media to assess colony formation activity from donor cells, a further indication of homing efficiency. In contrast to the direct homing assay, we found a significant increase (P<0.01) in the number of colonies generated from the Atxn1L−/− donor cells compared to WT (Figure 3D). This finding suggests that although similar numbers of progenitors are reaching the bone marrow, the Atxn1L−/− progenitors are significantly more proliferative. Since long-term HSCs are only a small portion (∼10%) of the KSL fraction, the above result largely reflects the properties of short-term HSCs and committed progenitors. Together, these data indicate that increased homing of hematopoietic progenitors is not likely to be a major factor contributing to the enhanced repopulating potential of the Atxn1L−/− HSCs, and underscores the observation of augmented activity from the mutant HSCs.
To test the hypothesis that Atxn1L−/− HSCs are more proliferative, we performed both in vivo and in vitro assays. We sorted single HSCs from unperturbed Atxn1L−/− mice into hematopoietic colony-promoting methylcellulose media in 96-well plates, counted the total number of colonies at multiple times points, and analyzed colony morphology to determine their myeloid differentiation potential. We found a higher number of colonies derived from the Atxn1L−/− vs. WT HSCs when counted after 7 days (P<0.01), although this difference was not as significant after 14 days, suggesting that the rate of growth of the colonies was faster in the Atxn1L−/− group than the WT, consistent with a higher proliferation rate. There was no difference in the colony types produced (Figure 4B).
To determine whether overexpression of Atxn1L would cause the opposite phenotype, we cloned Atxn1L into a retroviral vector that also expresses GFP downstream of an IRES that is cloned in tandem with the Atxn1L coding sequence. Hematopoietic progenitor cells were transduced with the virus, cultured for 24 hours, and GFP+ cells were sorted into wells containing methylcellulose medium. Stem and progenitor cells over-expressing Atxn1L generated fewer colonies compared to cells transduced with an empty vector control (P<0.01) (Figure 4C), consistent with a role for Atxn1L in negatively regulating HSC proliferation.
To examine whether Atxn1L−/− cells exhibited altered proliferative activity directly, we purified HSCs and immunostained them for Ki67, a marker of cycling cells. As expected, only a small portion of WT HSCs were Ki67-positive, whereas the proportion of Ki67-positive cells in Atxn1L−/− KO HSCs was significantly higher (P<0.01) (Figure 4D). This was also true in a less purified progenitor population (c-kit+, Sca-1+, lineage− (KSL), P<0.01) (Figure 4E, 4F).
If Atxn1L−/− HSCs have enhanced proliferation in vivo, this may be manifested in a differential response to agents that are toxic to proliferating cells. We thus tested the response of WT and Atxn1L−/− mice to a single injection of 5-FU by measuring their CBCs over an extended period of time. Overall recovery time was unchanged; however, recovery of white blood cells reached a higher peak at days 15–17 (P = 0.0007) (Figure 4G), consistent with a greater proliferative response from mutant stem and progenitor cells (Figure 4D–4F). Recovery of platelets and RBCs showed no significant differences (not shown). To exclude the possibility that the phenotype is accounted for by greater resistance to apoptosis we analyzed stem and progenitor cells in WT and Atxn1L−/− mice, before and after 5-FU treatment, but there was no significant difference in the number of apoptotic cells (data not shown). While one might expect a higher stem cell cycling rate to be accompanied by slower HSC recovery time because 5FU would also kill the cycling KO HSCs, the faster recovery suggests the KO HSCs can become quickly activated. Furthermore, because the progenitors are also more in cycle, their rapid expansion may compensate for the effect of 5FU on the HSCs leading to faster recovery.
Atxn1L−/− HSCs are more proliferative than WT HSCs and engraft better, but do not appear to cause leukemia, at least within the time frame of our analysis. To begin to gain insight into possible regulatory mechanisms that contribute to their high proliferative capacity, we determined the gene expression differences in Atxn1L−/− vs WT HSCs using expression microarrays (Figure 5). Consistent with the phenotype, where the KO HSCs are relatively normal in terms of their differentiation capacity but show an improvement of HSC activity, the gene expression differences were relatively modest. We found a total of 1013 genes that were different (484 up and 529 down; P-value≤0.05 and fold-change ≥1.5). Because the Atxn1L−/− HSCs resemble super-HSCs, we considered the possibility that their expression profile would reflect enhanced expression of HSC-critical genes. To test this, we compared the gene expression differences to the list of HSC-specific “fingerprint” genes [7]. We found a significant overlap of genes that are differentially expressed in Atxn1L−/− mice compared to WT and genes that are HSC fingerprint genes (Odds Ratio = 1.685, 95% CI 1.14–2.43, P = 0.00598). Some of the genes down-regulated in the KO are those identified functionally in other studies to have a key role in HSC maintenance and/or quiescence. For example both HoxA7 and HoxA9 are lower in the KO, and their loss has been associated with increased HSC proliferation [13], [14]. Similarly, Pbx1 is lower in the KO, and also has been linked to maintaining HSC quiescence [15]. Some of the genes upregulated in the KO stem cells are associated with high stem cell quality. For example, Tgf-beta-induced (Tgfbi) is upregulated. TGFb signaling is associated with the most long-term HSCs, which have the greatest self-renewal capacity [16]. Similarly, upregulated in the KO HSCs, is Plagl1, which is a member of a group of imprinted genes that are associated with somatic stem cells of multiple tissues [17].
To investigate the link between Atxn1L and HSC proliferation, we compared the transcriptional changes in Atxn1L−/− HSCs with signatures of HSCs during proliferation vs quiescence [18]. We found that there is significant overlap between the genes that are down-regulated in Atxn1L−/− HSCs and the quiescence signature (Qsig) genes (Odds Ratio = 1.914, 95% CI 1.42–2.54, P = 2.044×10−5) and a nearly-significant overlap between the up-regulated genes with the proliferation signature (Psig) gene sets (Odds ratio = 1.376, 95% CI 0.96–1.93, P = 0.066) (Figure 5C). These data are consistent with an impact of loss of Atxn1L on maintenance of quiescence, as shown in Figure 2 and Figure 4. Aside from a few genes, we did not see differences at the pathway level in TGFβ, Wnt or Pten/AKT pathways. The finding that the gene expression changes in Atxn1L null HSCs are concordant with decreased quiescence and increased HSC proliferation provides insight into the potential molecules mediating the HSC phenotypes, although the precise molecular event(s) leading to these phenotypes remains elusive.
To determine whether additional brain-associated genes could be used to identify other hematopoietic regulators, we systematically examined genome-wide data sets available in our laboratories and in public repositories in order to test for correlations between genes important in the brain and hematopoietic system. We previously identified a set of genes uniquely expressed in HSCs relative to their differentiated counterparts (HSC “fingerprint” genes) [7], as well as genes expressed in differentiated hematopoietic lineages but excluded from HSCs. Similarly, we have developed brain protein-protein interaction networks for several proteins that are abundant in the brain and are known to cause either ataxia or autism [19]–[21]. We used the genes from these data sets to probe the Mouse Genome Informatics (MGI) repository of knock-out mouse phenotypes in order to link genes within these datasets with neurological phenotypes in existing mutant mice (Figure 6A). As expected, the genes from proteins in the ataxia and autism interactomes are enriched for genes annotated to nervous system and behavioral phenotypes by the MGI. In other words, a significant number of those genes have already been reported to result in nervous system or behavioral phenotypes after genetic manipulation (most frequently knock-out). Surprisingly, the HSC fingerprint set (319 genes) was also highly enriched for genes reported to cause a nervous system or behavioral phenotypes after genetic manipulation (Fisher's Test P = 3.39×10−9), with a Z-score similar to the enrichment of genes derived from the autism interactome (Table S1). In contrast, lists containing genes specifically expressed in multiple other hematopoietic cell types, including multiple lymphoid and myeloid cell types did not show a similar enrichment for phenotypic annotations to nervous system/behavior phenotypes. The proliferation rate of normal HSCs is low, as for most cells in the brain. To determine whether enrichment for genes reflecting the proliferation state could account for the parallels between the HSC-specific genes and the nervous system/behavior phenotypes, we performed parallel analysis with a list of genes derived from quiescent vs proliferating progenitors [18]. However, these did not show enrichment with nervous system/behavior phenotypes, indicating that the enrichment in the HSC-specific gene list is not due solely to their proliferative state.
To examine this relationship using an independent list of genes, we took a compendium of genes shown to have an impact on HSC function when ablated in mice [22]. Among genes with a reported HSC phenotype, we found a significant enrichment in those likely to have a nervous system/behavioral phenotype (Z = 4.06; p = 5.72×10−5; Table S2).
To ensure that this relationship between HSC-specific genes and the likelihood of a reported nervous-system-related phenotype was not simply due to a testing artifact based on properties of the MGI data, we examined the MGI database more systematically. We examined reported phenotypes for all genes in the MGI repository to determine the inherent correlation between genes tested for, and reported to have, KO phenotypes associated with the hematopoietic system, the immune system, the nervous system, and neurological/behavioral defects. In other words, we examined overlap, or similarity, between these phenotypic categories, determining the frequency with which a gene KO is reported to have both hematopoietic system defects as well as neurological phenotypes. As expected, genes that were reported to have an immune system phenotype were likely to also be reported as having a hematopoietic phenotype. Similarly, genes that were annotated as resulting in a nervous system phenotype after KO were likely to also have a behavioral phenotype (Figure 6B). However, there was very low overlap in genes listed as having a neuronal phenotype with those having a hematologic phenotype. These classes are distinctly lacking in commonly annotated genes. This contrast may arise because investigators often focus on one tissue, and therefore phenotypes in other tissues may be missed or under-reported. Nevertheless, these findings stand in sharp contrast to the significant overlap we empirically observed between our experimentally-derived HSC-specific gene sets and genes reported in MGI to have neurological phenotypes (Figure 6A). This strongly supports the concept of a unique relationship between the HSC-specific genes and the nervous system, and argues for the need of more cross-system phenotyping, and comprehensive reporting of phenotypes.
We therefore hypothesized that we may find direct enrichment of genes expressed in HSCs in the neuronal disease protein interactomes as opposed to the indirect analyses through the MGI phenotypes. To examine this, we tested whether genes expressed in HSCs were enriched in the ataxia and autism interactome, and in gene expression data from various brain regions including the hypothalamus, cerebellum and amygdala in wild-type mice [23]–[25]. We found that HSC-expressed genes were highly enriched in all brain regions as well as in gene sets encoding proteins in the ataxia and autism interactomes (Figure 6C; Table S1). To determine the specificity of the relationship to the brain, we also compared the HSC-expressed genes to those found in data sets from skeletal muscle, bone, and liver. There was no significant enrichment in these tissues, indicating that the HSC-brain relationship is relatively unique.
The ataxia protein interactome was then independently used to generate a subnetwork comprised of HSC-expressed genes and their interacting partners (Figure 6D) and we plotted the protein interactions of this sub-network based on evidence from the ataxia interactome. The dense interconnected network that emerges suggests that at least some of the protein-protein interactions observed in the brain are likely to also take place in the HSCs based on their gene expression. The overlap with the interactome data suggests that the relationship between HSC and neuronal genes appears to be rooted in molecular interactions, and is not simply a feature of the similarity of the phenotypes reported for the relevant genes.
Finally, we examined the identity of the genes that overlap in the narrowly defined data sets. When HSC-specific genes (274 that are mapped to the human genome) are overlapped with genes encoding proteins in the Autism interactome (2437 human genes), we identify 36 genes (Table S3). Similarly, the overlap between HSC fingerprint genes and those from the Ataxia interactome (3436 genes) is 45 genes (Table S4). The genes included in both interactomes that are also HSC-specific number only 17, of which three are known to have a function in the hematopoietic system (e.g. Gata2). Several are implicated in the neurologic system (e.g. Col4a2), but few have a precisely-defined role (Table 1). Of the remainder, most have functions described outside the hematopoietic and nervous systems (e. g. Tle1, a Groucho-like repressor). Intriguingly, three of these (Grb10, Peg3, and Ndn) are monoallelically expressed depending on their parent-of-origin (imprinted), a gene group known to be enriched in multiple types of somatic stem cells [17]. Three others are members of the Wnt signaling pathway (Mdfi, Tcf7l1, Enah). All of these 17 genes would be intriguing candidates for further study in the hematopoietic or nervous systems.
We have identified surprising overlap between genes expressed in HSCs and genes that are expressed in the brain and encode proteins in protein interaction networks for neurological diseases such as ataxia and autism. We have shown that this relationship is not recapitulated by genes expressed in differentiated hematopoietic cells, but is specific to genes expressed in HSCs. Thus, these data reveal a previously underappreciated functional relationship and raise the possibility that additional genes critical for normal brain function might be candidates for regulating HSCs, and vice versa.
This finding is surprising given the generally low overlap between genes annotated to neuronal and hematologic phenotypes (Figure 6C). The fact that there is little correlation between mice with a nervous system phenotype and those with a reported hematopoietic stem cell phenotype as described in MGI, most likely reflects the limited set of phenotypes most investigators consider when studying their genes of interest. Our finding that Atxn1L, a gene identified from the Ataxia interactome, has a hematopoietic phenotype when ablated supports this concept. These data argue that this approach of highly focused phenotyping may obscure unexpected correlations that may have functional relevance. Cross-system analyses, particularly when functions might be predicted from computational approaches derived by mining available biological and in silico data, may be of significant value.
Our data clearly establish Atxn1L as a negative regulator of HSC function. By employing multiple functional assays we show that Atxn1L−/− HSCs are super-HSCs. They regenerate the blood of recipient mice to higher levels than WT HSCs, they recover more quickly from myeloablative treatment, and they exhibit better engraftment even after secondary transplantation, a rigorous measure of HSC self-renewal capacity. Although, there are now a number of genes that when ablated result in decreased stem cell function, there are relatively few that result in enhanced HSC activity [22]. Genes that act normally to restrain HSC activity, resulting in higher performance after KO, include Cbl [26], Slug [27], Cdkn2c (p18) [28] and Gli1 [29]. Importantly, none of these genes showed significant down-regulation in Atxn1L−/− HSCs, suggesting that the mechanism of enhanced stem cell function in the Atxn1L−/− mice is distinct.
Many of the genes that affect HSC function impact the proliferation rate of HSCs [22]. Paradoxically, higher proliferation of HSCs is usually linked to lower HSC activity. For example, HSCs from Irgm KO mice show excessive proliferation and poor engraftment properties, owing to hyper interferon signaling [30], [31], and Gfi1 mutant HSCs are also hyperproliferative and similarly defective [2]. While not well understood, this link between high HSC proliferation and poor engraftment probably relates to differentiation-associated HSC proliferation that ultimately depletes the stem cell pool. Consistent with this, some mutants that decrease HSC proliferation, for example Gli1, augment HSC function [29]. On the other hand, increased HSC proliferation can also be associated with enhanced HSC function: KO of Slug or Gfi1b, both putative transcriptional repressors, results in improved HSC bone marrow engraftment activity along with slightly increased HSC proliferation [27], [32], similar to our observations in Atxn1L−/− mice. Again, while not fully understood, moderately higher proliferation may enable more rapid engraftment after transplantation (similar to accelerated recovery of blood counts after 5FU) that, if not excessive, may also preserve stem cell function. These findings underscore the critical balance that is maintained to optimize the competing roles of stem cells in self-renewal and differentiation.
The molecular mechanism of the Atxn1L−/− HSC phenotype is not easy to establish at this time as no major pathways were altered in the gene expression analysis to suggest particular avenues for further study. Atxn1 and Atxn1L have both been shown to interact with the transcriptional repressor Capicua (Cic), which mediates a number of their downstream effects. In the lung, loss of the Atxn1/Atxn1L destabilizes Cic complexes leading to de-repression of activators of matrix metalloproteinases that in turn contribute to the lung alveolarization defects [11]. We have not examined protein levels of Cic in HSCs, but we do detect high expression of Cic in HSCs [7], which leaves open the possibility of a role for Cic in the hematopoietic phenotype as well. Ultimately, better understanding of the mechanisms that lead to enhanced stem cell function could lead to strategies to expand HSCs for bone marrow transplantation which, despite much effort, has still not been achieved.
More broadly, our work suggests the existence of molecular networks that are utilized in both brain and hematopoietic stem cells, but not their differentiated counterparts. Whether these networks are also used in other adult stem cells, as was recently suggested for imprinted genes [17] is an open question. Our work also suggests a paradigm for using cross-tissue bioinformatic analyses to identify new key regulators in blood or brain. While other genes are anecdotally linked in both systems, we expect many others could be probed. With the advent of the large-scale mouse phenotyping efforts stimulated by the knock-out mouse consortia, these types of analyses offer a parsimonious use of resources to efficiently identify important phenotypes and cross-tissue phenotype comparisons.
It is interesting to consider why this apparent relationship exists. The fact that neither HSCs nor most brain cells actively divide does not seem to be the cause, as our quiescence signature genes do not show the same enrichment as the HSC fingerprint (Figure 6). We speculate that there is either a relationship rooted in ontology or evolution that has not been previously noted, or that there is some underlying functional origin. For example, HSCs have a close relationship with other cells in their niche- perhaps they utilize a “synapse” to communicate with other key bone marrow components. Along these lines, a link has previously been noted between some genes with an impact on endothelial cell function and those involved in brain function. For example, classical axon-guidance cues also help guide blood vessel formation [33]. HSCs and endothelial cells have a close relationship that originates in their development. HSCs arise from specialized endothelial cells [34], [35] and co-express a number of key genes such as Runx1, Sca1, and Scl/Tal1. Thus, it is possible that underlying relationship between HSCs and the brain is also linked to their commonalities with endothelial cells. Systematic analyses with endothelial-specific genes of the type we have performed here would be required to probe this possibility further.
It is also possible that our observations of common brain-HSC networks may hold for humans. Some well known genetic syndromes have been recognized to exhibit both neurologic and hematologic components. For example, Alpha-Thalassemia mental Retardation X-linked syndrome (ATRX) is named for its involvement in both alpha thalassemia and mental retardation (OMIM 301040). Similarly, Ataxia-Telangiectasia (OMIM 208900) and Nijmegen Breakage syndrome, (OMIM 251260) have both hematologic and neurological manifestations. Furthermore, Autism patients may have higher frequencies of infections [36], which could suggest shared genetic etiology. Further studies to explore this intriguing link between the neurologic and hematopoietic system defects are clearly warranted.
All mice were backcrossed to the C57Bl/6 background and were housed in a specific-pathogen-free animal facility, AALAC-accredited, at Baylor College of Medicine (Houston, TX).
For peripheral blood analysis, transplant recipient mice (n = 8/genotype) were bled at 4, 8,12 and 16 weeks post transplantation. Red blood cells were lysed and samples were stained with CD45.1-APC, CD45.2-FITC, CD4-pacific blue, CD8-pacific blue, B220-pacific blue, B220-PE-cy7, Mac1-PE-cy7 and Gr-1-PE-cy-7 antibodies (BD Pharmingen, eBiosciences). FacsARIA, LSRII and FACS-Scan flow cytometers were used for analysis and cell sorting. Hematopoietic committed progenitors were analyzed based on expression of cell surface markers that can be identified using flow cytometry as described [37]. For complete blood counts, peripheral blood was collected from the retro-orbital plexus into tubes containing potassium EDTA (Sarstedt, Nümbrecht, Germany) from 8–10 week old WT and Atxn1L−/− mice (n = 10 mice/genotype) and analyzed with a Hemavet analyzer (Drew Scientific, TX, USA). LT-HSCs defined in text and legends. ST-HSCs: Sca-1+, c-kit+, CD34+, Flt3−; MPP: Sca-1+, c-kit+, CD34+, Flt3+; CLPs: Lineage−, IL7rα+, Sca-1+, c-kit+; CMP: Lineage−, IL7rα−, Sca-1−, c-kit+, CD34+, CD16/32−; GMP: Lineage−, IL7rα−, Sca-1−, c-kit+, CD34+, CD16/32+; MEP: : Lineage−, IL7rα−, Sca-1−, c-kit+, CD34−, CD16/32−.
Competitive bone marrow transplantation assays were performed by intravenous injection of admixed CD45.2 donor whole bone marrow cells with CD45.1 competitor bone marrow. Recipient C57Bl/6 mice had been lethally irradiated with a split dose of 10.5 Gy, 3 hours apart. Sex- and age-matched C57Bl/6 mice were used as competitors for every experiment. Eight recipient mice were used in each experiment (n = 8 mice/genotype), and each experiment was repeated at least twice. The competitor cell dose was kept constant at 250,000 cells in all transplants.
For the limiting dilution assays to determine repopulating units, we used 1.0×103, 3.0×104 and 1.0×105 WT or Atxn1L−/− whole bone marrow cells pooled from (n = 3/genotype) CD45.2 sex- and age-matched mice (for the number of recipient animals in each dilution group, see Figure 2), mixed with 250,000 CD45.1 cells. Positive engraftment was scored based on multilineage repopulation of higher than 0.1%. The percentage of non-responders was calculated using the L-Calc software (StemCell Technologies).
HSC transplants were carried out as described above, but instead of whole bone marrow donor cells, purified HSCs were transplanted. Unless specified otherwise, HSCs were isolated using the side population (SP) method for Hoechst dye efflux [38], followed by KSL (c-Kit+, Sca1+, lineage−) and CD150+ staining. Recipient mice (n = 8/genotype) received 20 sorted HSCs and 250,000 WT competitor cells. Staining and isolation of HSCs were carried out as previously described [37].
For the secondary transplants, primary recipient mice were sacrificed 16 weeks post transplant and CD45.2 donor-derived HSCs were isolated as described above. Fifty HSCs were transplanted into secondary CD45.1 recipients, along with 250,000 CD45.1 competitor whole bone marrow.
Homing efficiency of donor cells into the recipient bone marrow was characterized in two ways. First, 30,000 KSL cells from pooled WT and Atxn1L−/− mice (CD45.2) were isolated and transplanted into lethally irradiated CD45.1 recipient mice (n = 5/genotype). Eighteen hours after the transplant, the recipient mice were sacrificed and their bone marrow was analyzed for the presence of CD45.2 positive cells using flow cytometry. A fraction of the whole bone marrow was also plated on methylcellulose medium in 32 mm dishes (n = 5 dishes/genotype). Controls included irradiated recipient mice that received no CD45.2 marrow. The resulting colonies are derived from the CD45.2 donor cells that homed into the recipient mice bone marrow. Thus, 12 days after plating, colonies were counted in each well to assess the homing efficiency of donor cells.
HSCs were identified using the Hoechst dye efflux method along with positive staining for c-kit, Sca-1, CD150 and excluding lineage positive cells. Single HSCs were sorted in 96-well plates containing methylcellulose medium (StemCell Technologies). The number of colonies was counted at days 4, 7 and 14 and scored based on morphology on day 10.
For the in vitro colony proliferation assay, HSCs were sorted and plated into 6-well plates containing methylcellulose, 100 HSCs per well. Seven days later, single colonies were picked and resuspended in HBSS medium contain FBS (Gibco). The cell suspension was washed twice, stained with PI in sodium citrate and analyzed by flow cytometry.
MSCV-Atxn1L-IRES-GFP and MSCV-IRES-GFP vectors were packaged using HEK293T cells by co-transfecting with pCL-Eco [39]. Mice were treated with 5-fluorouracil (150 mg/Kg body weight, American Pharmaceutical Partners) 6 days before harvesting the whole bone marrow. The bone marrow was enriched for Sca-1 expressing cells using magnetic selection (AutoMACS, Miltenyi), transduced with the retrovirus as previously described [7], and grown in culture. After 48 hours, cells were collected, stained for Sca-1, c-kit and lineage markers, and GFP+ KSL cells were sorted and plated into 6-well plates containing methylcellulose medium. Ten days later, the number of colonies in each well was counted.
Whole bone marrow from WT and Atxn1L−/− mice (n = 5/genotype) was isolated and stained for different hematopoietic progenitor populations. Cells were then fixed and stained for either BrdU or Ki-67 according to the BrdU staining protocol supplied by the manufacturer (BD-Pharmingen).
To determine how WT and Atxn1L−/− mice respond to stress, mice were treated with the chemotherapeutic drug, 5-FU. To determine the survival rate, mice (n = 5/genotype) received one injection of 5-FU (150 mg/Kg body weight) and analyzed for proliferation or apoptosis 3, 5 and 7 days later by flow cytometry.
In order to assess hematopoietic recovery after stress, WT and Atxn1L−/− mice (n = 10/genotype) were treated once with 5-FU and their peripheral blood counts were monitored every three days for 28 days using the Hemavet analyzer (Drew Scientific, TX, USA).
Annexin-V staining was used to assess cell death and apoptosis. Briefly, cells were harvested and stained with the markers of interest according to the staining protocol described above. Cells were washed twice with cold PBS and incubated at room temperature in 1× binding buffer (10 mM HEPES, 140 mM NaCl, 2.5 mM CaCl2) containing Annexin V-APC (BD-Pharmingen). Cells were analyzed by flow cytometry within one hour of staining.
HSCs were purified as described above. Cells were purified from 8-week-old mice. Approximately 30,000 HSCs from WT and Atxn1L−/− mice were purified for RNA isolation. RNA was isolated using the RNAqueous kit (Ambion, Austin, TX, USA), and treated with DNase I. The RNA was linearly amplified using two rounds of T-7 based in vitro transcription using the MessageAmp kit (Ambion). The RNA was subsequently labeled with biotin-conjugated UTP and CTP (Enzo Biotech). The amplified RNA was hybridized to MOE430.2 chips according to standard protocol at the BCM Microarray core (Houston, TX). Data were analyzed by GCRMA with correction for false-discovery [40]. Data can be found in GEO, with accession number: GSE44285.
In all cases of overlap analysis between gene sets, a Fisher's exact test was performed to determine p-values and statistical significance. We also generated Z-scores to measure the deviation between the observed overlap (number of genes in common between two sets) and what would be expected if one set were fixed and a random set was generated to overlap with it (eg. array overlap with P-sig, Q-sig and the HSC fingerprint). The expected overlap size was determined as the product between frequency of the fixed set and the size of the comparator set. The frequency was determined as the number of genes in the fixed set divided by the number of all genes that could be sampled; the size of the comparator set was limited to the size of the comparator overlapped with the sample universe (i.e. when doing cross platform comparison, the comparator size was limited to the subset of the comparator represented among the universe of the fixed set. To generate the network in Figure 6D, we used homologene to map mouse gene symbols to human orthologs. We then identified all interactions in the interactome where one of the partners was expressed in HSC according to the HSC fingerprint dataset [7]. We used Cytoscape to generate the network image. We identified the gene products that were additionally identified as being HSC fingerprint genes by coloring them red. HSC expressed genes are colored pink.
For microarray data we used the R Bioconductor package GCRMA to process the low-level intensity data. We used the limma package to generate T-statistics and moderated p-values. We used the Bioconductor package mouse4302 v2.2 to determine the gene symbols for the probe sets on the array, and we used the same release to compare both the previously published HSC gene lists and our new array results.
All animal work has been conducted according to national and international guidelines. The institutional animal care and use committee (IACUC) at Baylor College of Medicine approved the animal protocols for the work described herein. No human or primate samples were used for this work (data mining only).
|
10.1371/journal.pgen.1006443 | Loss of C9orf72 Enhances Autophagic Activity via Deregulated mTOR and TFEB Signaling | The most common cause of the neurodegenerative diseases amyotrophic lateral sclerosis and frontotemporal dementia is a hexanucleotide repeat expansion in C9orf72. Here we report a study of the C9orf72 protein by examining the consequences of loss of C9orf72 functions. Deletion of one or both alleles of the C9orf72 gene in mice causes age-dependent lethality phenotypes. We demonstrate that C9orf72 regulates nutrient sensing as the loss of C9orf72 decreases phosphorylation of the mTOR substrate S6K1. The transcription factor EB (TFEB), a master regulator of lysosomal and autophagy genes, which is negatively regulated by mTOR, is substantially up-regulated in C9orf72 loss-of-function animal and cellular models. Consistent with reduced mTOR activity and increased TFEB levels, loss of C9orf72 enhances autophagic flux, suggesting that C9orf72 is a negative regulator of autophagy. We identified a protein complex consisting of C9orf72 and SMCR8, both of which are homologous to DENN-like proteins. The depletion of C9orf72 or SMCR8 leads to significant down-regulation of each other’s protein level. Loss of SMCR8 alters mTOR signaling and autophagy. These results demonstrate that the C9orf72-SMCR8 protein complex functions in the regulation of metabolism and provide evidence that loss of C9orf72 function may contribute to the pathogenesis of relevant diseases.
| C9orf72 is one of many uncharacterized genes in the human genome. The presence of repeated nucleotides in the non-coding region of the C9orf72 gene (GGGGCC) has been linked to the neurodegenerative diseases Amyotrophic Lateral Sclerosis (ALS) and Frontotemporal dementia (FTD). However, how the presence of these repeats in the gene leads to neurodegeneration is unknown. One possible explanation is that the repeats lead to a reduced expression of the C9orf72 gene and loss of function of the C9orf72 protein. Although C9orf72 is well-conserved among multi-cellular organisms, its protein function remains to be determined. In this study, we demonstrated that loss of C9orf72 reduces mTOR signaling and enhances autophagy. mTOR signaling and autophagy are important for the cellular maintenance of metabolic balances, especially under stress conditions. C9orf72 protein exists in a complex with another DENN-like protein, SMCR8, which also regulates mTOR signaling and autophagy. We generated mice lacking C9orf72, which died prematurely and showed dramatic upregulation of TFEB, a crucial transcriptional regulator of autophagy and lysosomal genes, that integrates mTOR activity state and autophagic capacity. We propose that C9orf72 function is important for metabolic control and its deficiency can contribute to the development of neurodegenerative diseases.
| Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease characterized by the progressive degeneration of motor neurons. Frontotemporal dementia (FTD) is the second most common type of dementia in people younger than 65 and is characterized by degeneration of the frontal and temporal lobes of the brain. A hexanucleotide repeat expansion (HRE), (GGGGCC)n, in the promoter or intron of the uncharacterized gene, chromosome 9 open reading frame 72 (C9orf72), has been found to be the most common cause of both ALS and FTD [1, 2] and has been linked to a number of other neurological disorders. How the C9orf72 HRE leads to neurodegeneration remains to be determined, although both gain-of-toxicity and loss-of-function mechanisms have been proposed. The gain-of-toxicity mechanisms involve both RNA and protein products generated from the expanded hexanucleotide repeats. For example, RNAs containing the expanded repeats can interfere with the functions of specific RNA-binding proteins [3–5], and toxic repeat polypeptides can be generated through repeat-associated non-ATG-dependent translation [6–10]. However, the HRE could be pathogenic through loss-of-function mechanisms when the expression of the C9orf72 gene is disrupted. Multiple studies have demonstrated that C9orf72 RNA and protein levels are reduced in patient cells and brains [11–15]. Although partial knockdown of C9orf72 in the brain or its neural-specific deletion does not affect survival in mice [16, 17], loss of C9orf72 orthologs in zebrafish and C. elegans has deleterious effects [18, 19].
Studies of these loss-of-function mechanisms are hampered by a lack of knowledge about the physiological function of the C9orf72 protein. Bioinformatic analysis suggested that C9orf72 is a DENN-like protein [20, 21], which is a family of proteins that regulate small GTPases and membrane trafficking. DENN domain-containing proteins have also been implicated in autophagy and in the mammalian target of rapamycin (mTOR) signaling pathways [22]. Although a recent study has reported that C9orf72 regulates autophagy and endosomal trafficking [23], the function of the C9orf72 protein remains largely unknown.
Here we report the findings in mice and human cells that loss of C9orf72 inhibits mTOR signaling and leads to a profound upregulation of transcription factor EB (TFEB) and enhanced autophagy flux. We further show that C9orf72 interacts with another DENN-like protein Smith-Magenis syndrome chromosome region candidate 8 (SMCR8), which also regulates mTOR signaling and autophagy. The results suggest that a deficiency in the function of C9orf72 may contribute to the pathogenesis of relevant neurodegenerative diseases.
To study the physiological functions of C9orf72 in mammals, we generated a knockout (KO) mouse model lacking the protein. Human C9orf72 has one orthologous gene in the mouse, 3110043O21Rik, which is located on chromosome 4. For convenience, we refer to the mouse gene as C9orf72 hereafter. The mouse C9orf72 gene is predicted to produce seven transcripts, three of which are protein-coding, as compared to the human C9orf72 gene, which produces three transcripts and two protein isoforms. The mouse C9orf72 proteins share 98% identity with their human C9orf72 counterparts (S1 Fig). We generated C9orf72 KO mice by using a mouse embryonic stem (ES) cell line that contains a heterozygous allele of a 7754 base pair deletion in the C9orf72 gene. This deletion results in the removal of exons 2–6 and is predicted to produce nonfunctional truncated protein products from all three protein-coding transcripts of the mouse C9orf72 gene (Fig 1A). We further removed the neomycin cassette by crossing the C9orf72 KO male mice carrying the original targeted allele with SOX2-Cre transgenic females (Fig 1A). Western blotting of brain homogenates from C9orf72 wild-type and KO littermates, using an antibody predicted to detect all mouse C9orf72 isoforms, showed a protein band at 55 kDa (corresponding to mouse isoform 1), not present in the C9orf72-/- samples (Fig 1B), confirming that our KO mice lack C9orf72 in brain. We were unable to detect the other two mouse C9orf72 isoforms, suggesting that mouse isoform 1 is the major isoform in the mouse brain. The homozygous C9orf72 KO mice showed a decrease in survival compared with littermates, with more than 50% dead in 600 days (Fig 1C). This decrease in survival was also observed in heterozygous C9orf72+/- animals to a lesser degree with only about 20% dead in 600 days. Both C9orf72 homozygous and heterozygous knockout mice developed normally before exhibiting rapidly progressive lethargy before death. The stage of lethargy could last for days up to a month. At the end stage, the animals showed a lack of excitability or response to external stimuli (S1 Movie). In post-mortem examination, consistent with recent reports of immune dysregulation in C9orf72 knockout mice [24–27], we observed splenomegaly in the C9orf72-/- mice. The spleen was generally increased in length from ~3/4 inches to 1–1.25 inches. In addition, we frequently observed potential tumors in the thymus or in the regions of the abdomen. There was no obvious neuronal cell death in brain or spinal cord, but functional deficits of the nervous system could not be excluded. The exact cause of death for these C9orf72 knockout mice remains to be determined.
Although we observed no obvious neuronal defects in C9orf72 KO mice, it is possible that C9orf72 has functions in the nervous system in response to stresses. Thus, we asked if mTOR signaling is altered when C9orf72 is absent, since mTOR signaling is a central signaling pathway that senses the stresses related to nutrient availability, oxygen, and energy levels [28]. Also, DENN-like proteins have been implicated in mTOR signaling and nutrient sensing [29–31] and C9orf72 contains DENN domains. We monitored mTOR activity by assessing the phosphorylation of its downstream target ribosomal protein S6 kinase B1 (S6K1). Cells were starved for amino acids for 50 minutes before amino acids were added back to induce the phosphorylation of S6K1. Interestingly, knockdown of C9orf72 in HEK293T cells resulted in a decrease in the phosphorylation of S6K1 within 10 to 20 minutes after addition of amino acids, as compared with control cells transfected with scrambled control shRNAs (Fig 2A and 2B). These results suggest that the loss of C9orf72 decreases mTOR activation after amino acid stimulation. To study the molecular defect in the complete absence of C9orf72 protein, we generated mouse embryonic fibroblasts (MEFs) from C9orf72 wild-type and KO littermates. And we assessed the phosphorylation of S6K1 in the C9orf72-/- MEF lines. Phosphorylation of S6K1 was decreased in C9orf72-/- MEF lines compared with lines derived from wild-type littermates (Fig 2C), suggesting that mTOR activation after amino acid stimulation is diminished in the absence of C9orf72.
Subsequently, we asked whether the observed reduction of mTOR activation in the absence of C9orf72 impacts the function of TFEB, a transcription factor that is a master regulator of lysosome biogenesis and autophagy-related genes, and a substrate of mTOR [32]. In an autoregulatory loop, nuclear translocation of TFEB leads to increased expression of itself. Phosphorylation of TFEB by mTOR prevents its translocation to the nucleus and causes down-regulation of TFEB. We transfected GFP-TFEB into HEK293T cells and observed that knockdown of C9orf72 resulted in a significant increase in GFP-TFEB levels (Fig 3A and 3B), consistent with the decrease in mTOR activity. Moreover, imaging analysis indicated that nuclear localization of GFP-TFEB was significantly increased upon knockdown of C9orf72 as compared with cells treated with control shRNAs (Fig 3C and 3D). Western blotting of the nuclear and cytoplasmic fractions further confirmed that GFP-TFEB was enriched in the nucleus upon knockdown of C9orf72 (Fig 3E). Next, we validated these results in C9orf72-/- and wild-type MEF cells. Consistently, the complete absence of C9orf72 led to a significant increase in the nucleus to cytoplasm ratio of GFP-TFEB signals (Fig 3H and 3F). Furthermore, consistent with the notion that TFEB promotes the biogenesis and activity of lysosomes [32], we observed a significant increase in the number of LysoTracker-stained acidic vesicles in the C9orf72-/- MEF cells, confirming functional consequences on lysosomes of enhanced nuclear TFEB (Fig 3H and 3G).
We then questioned whether our results held true in vivo. Analysis of brain homogenates by western blotting from all examined C9orf72 KO mice showed a dramatic increase in endogenous TFEB levels compared with wild-type controls (Fig 3I), consistent with our results in tissue culture. We next asked whether downstream targets of TFEB were also increased by loss of C9orf72. Indeed, western blot analysis of lysosome-associated membrane glycoprotein 1 (LAMP1), which is a transcriptional target of TFEB [33], indicated that LAMP1 was profoundly increased in the C9orf72 KO mouse brains (Fig 3I). Another related lysosomal protein LAMP2 was also markedly increased in the absence of C9orf72 in the KO mouse brains. Taken together, these results suggest that, consistent with the inhibition of mTOR signaling, loss of C9orf72 increases TFEB activity.
Since we observed a function of C9orf72 in mTOR signaling and mTOR is known to negatively regulate autophagy, we assessed the levels of the autophagy marker LC3 by immunoblotting in these cells. During autophagy, LC3I is processed to LC3II via lipidation, which allows for insertion of the LC3 protein into the autophagosome membrane. Our results show a significant increase of LC3I in C9orf72-/- MEFs when compared with wild-type MEFs, indicating that basal autophagy is altered in these cells (S2A and S2B Fig). To test the role of C9orf72 in neurally differentiated cells, we generated embryonic stem cells from C9orf72 KO mice and littermate controls and differentiated them into the motor neuron precursors that further grew into mature motor neurons (~40% of the culture) plus astrocytes and oligodendrocytes. We assessed the level of LC3 by immunoblotting and found that the C9orf72-/- cells enriched with motor neurons showed a substantial accumulation in LC3I (S2C Fig), in line with what was observed in C9orf72-/- MEFs.
The decreases in LC3II/LC3I ratio observed in our western blots can indicate a defect in lipidation or an increase in degradation via the lysosome. To distinguish between these two possibilities, we assessed LC3 levels after nutrient deprivation-induced autophagy in the absence and presence of the lysosomal inhibitor Bafilomycin in C9orf72-/- and wild-type MEF cells. We found that the Bafilomycin-induced accumulation of LC3II was significantly enhanced in C9orf72-/- MEFs compared with wild-type MEFs (Fig 4A and 4B), indicative of an enhanced autophagic flux in C9orf72-depleted cells. To further examine the status of autophagic flux, we analyzed the numbers of LC3-positive autophagic vesicles and the colocalization between LC3 vesicles and Rab7, a late endosome-/lysosome-associated GTPase that marks mature autophagolysosomes [34–36] (Fig 4C–4F). Quantification of LC3-positive vesicles in the absence and presence of Bafilomycin, demonstrates that, consistent with western blot results, the number of LC3-positive autophagic vesicles was significantly increased in C9orf72-/- MEFs (Fig 4D). An autophagic flux index, defined as the difference in the volumes of LC3-positive vesicles before and after Bafilomycin treatment, was quantified, further confirming the increased autophagic flux capacity in C9orf72-/- MEFs (Fig 4E). Similarly, we observed an increase in the colocalization of LC3-positive autophagic vesicles with Rab7-positive vesicles (Fig 4F), confirming enhanced autophagolysosome formation. In addition to induced autophagy, we also examined basal autophagic flux under fully supplemented nutrient conditions in the absence of C9orf72 (S3 Fig). Despite of relatively low level of signals, the LC3/Rab7 vesicle colocalization assay indicated a trend that there were more LC3-positive vesicles and more colocalized LC3/Rab7 vesicles in Bafilomycin-treated C9orf72-/- MEFs than in wild-type control cells (S3A Fig). This result is consistent with the western analysis of LC3 protein levels, in which LC3II accumulated robustly in Bafilomycin-treated C9orf72-/- MEFs (S3B Fig).
In addition to MEF cells, we observed a similar result in HEK293T cells for autophagic flux after knockdown of C9orf72. Under nutrient deprivation, in cells treated with C9orf72 shRNA, despite a decrease in LC3II/LC3I ratio before Bafilomycin treatment, lysosomal inhibition induced a robust accumulation of LC3II (S3C Fig), suggesting that the total autophagic flux was increased. Taken together, the observed increases in autophagic activity are consistent with the impairment of mTOR signaling and the profound increase of TFEB as results of loss of C9orf72.
Of note, our results do not rule out the possibility that C9orf72 functions in other aspects of autophagy. For example, we examined the activity of ATG4B, which catalyzes the cleavage of proLC3 to produce LC3I and also removes LC3II from the autophagosome membrane after it fuses with the lysosome (S4A Fig). Knockdown of C9orf72 in HEK293T cells resulted in a significant decrease in the signal of an ATG4B activity luciferase reporter when compared with control cells (S4B Fig), suggesting that ATG4B activity is impaired under basal conditions. However, no change was detected in ATG4B protein levels by western blotting upon knockdown of C9orf72 (S4C Fig and S4D Fig), indicating that the reduction in ATG4B activity was not due to a decrease in its protein level. The unchanged level of ATG4B protein could presumably make it readily available to support the enhanced autophagic flux observed in nutrient deprivation-treated C9orf72 deficient cells.
We next investigated whether the absence of C9orf72 alters the markers of autophagy in vivo. Since mTOR signaling senses nutrient stresses and autophagy induction is a natural response to nutrient stresses through mTOR, we asked whether the absence of C9orf72 affects the autophagic response under these stress conditions. We applied amino acid withdrawal by feeding mice a low-protein diet that is well-tolerated in young animals [37]. Beginning at 4 months of age, gender-matched wild-type and C9orf72 KO littermates were fed either normal or amino acid-deficient chow for four weeks before tissues were harvested for analysis (Fig 5A). We first examined autophagy in the brain of C9orf72 KO mice. Because of the low levels of LC3 conversion during starvation in the brain [38], we assessed the levels of the autophagy marker protein p62. Western blotting of brain homogenates showed a slight decrease in p62 levels in C9orf72 KO mice when compared with wild-type littermates, a defect that became more pronounced when the mice were on the low-protein diet (Fig 5C and 5D). The decrease of p62 was not due to change in its solubility since no insoluble p62 was detected in western analysis or histological examinations (Fig 5B). The lack of accumulation of p62 in the brains of C9orf72 KO mice suggests an increased autophagy activity. Consistent with the results in the mouse brain, we also observed a decrease in p62 levels in C9orf72-/- MEF lines compared with wild-type cells (S5A Fig and S5B Fig). We next examined the liver, a common tissue type used to study autophagy, harvested from the wild-type and C9orf72 KO littermates, for changes in LC3. We observed a decrease in the level of LC3II protein or relative increase of LC3I protein in C9orf72-/- livers relative to the wild-type controls under the low protein diet condition (S5C Fig and S5D Fig).
To gain molecular insight into the function of C9orf72, we performed a quantitative proteomic screen for protein interactors of the C9orf72 protein using stable isotope labeling by amino acids in cell culture (SILAC) mass spectrometry (Fig 6A). Human C9orf72 Isoform A with a C-terminal Flag tag was expressed in HEK293T cells metabolically labeled with 13C,15N L-Arginine and L-Lysine and immunoprecipitated using Flag-tag beads. A parallel immunoprecipitation was performed using unlabeled mock-transfected cells as a control to identify proteins that bound to the Flag-tagged beads alone. The resulting immunoprecipitates were pooled and analyzed via mass spectrometry to identify proteins that were enriched by the C9orf72 bait. We identified SMCR8 as the top C9orf72 interactor since it had the highest SILAC ratio or enrichment (S6A and S6B Fig and S1 Table). Notably, SMCR8, although uncharacterized, is also a DENN-like protein [20, 21]. We validated this interaction by co-immunoprecipitation, with Flag-tagged C9orf72 pulling down endogenous SMCR8 in HEK293T cells (Fig 6B). Conversely, reciprocal immunoprecipitation experiments demonstrated that an anti-SMCR8 antibody pulled down Flag-tagged C9orf72, confirming their interaction (Fig 6C). The interaction was further validated by co-immunoprecipitation of co-expressed Flag-SMCR8 and C9orf72-V5 proteins (S6C Fig). Consistently, GFP-tagged C9orf72 and mCherry-tagged SMCR8 both localized to the nucleus and the cytoplasm in HEK293T cells (S6D Fig).
Since we identified SMCR8 as the most abundant protein interactor of C9orf72, we asked whether C9orf72 influences the level of SMCR8 protein. While examining the brain lysates from the C9orf72 KO mice, we observed a dramatic reduction in the level of SMCR8 protein. Although present in wild-type brains, SMCR8 was not detected in C9orf72-/- brain homogenates by western blotting (Fig 6D). Examination of SMCR8 transcripts by qPCR showed no reduction in its mRNA levels, supporting that C9orf72 influences SMCR8 at the protein level (S7A Fig). Notably, WD repeat-containing protein 41 (WDR41), another protein identified in our proteomic screen (S1 Table) and recently confirmed to be an interactor of the C9orf72/SMCR8 complex [28, 29], was not decreased in C9orf72-/- brain samples (S7C Fig). In addition, overexpression of C9orf72 in HEK293T cells increases SMCR8, suggesting that C9orf72 regulates SMCR8 protein levels (Fig 6F and 6H).
To further study the function of SMCR8, we obtained a CRISPR/Cas-9 generated SMCR8 KO cell line. This cell line contains a frameshift mutation in the first exon of SMCR8 resulting in the loss of the full-length protein product (S8A–S8C Fig). Since we observed that C9orf72 regulates SMCR8 protein levels, we asked whether SMCR8 reciprocally influences the levels of C9orf72. By examining the lysates from the SMCR8 KO cells by western blotting, we observed a dramatic reduction in the level of C9orf72 protein (Fig 6E). We observed the same effect on the C9orf72 protein when we treated HEK293T cells with validated SMCR8 shRNA compared with control cells transfected with a scrambled shRNA control (S8D and S8E Fig). Examination of C9orf72 transcripts in SMCR8 KO cells by qPCR showed an increase in its mRNA levels (S7B Fig), suggesting that the loss of SMCR8 decreased the C9orf72 protein level not by reducing its RNAs.
Next we studied how SMCR8 regulates C9ORF72 protein levels. We first asked whether the regulation occurs due to changes in protein stability or turnover. Since the level of C9ORF72 was too low to allow for chase experiments to probe their turnover in the SMCR8 KO cells, we overexpressed C-terminal-V5 tagged C9orf72 and N-terminal-mCherry tagged SMCR8, or an mCherry only control, into HEK293T cells and studied their protein levels. Compared with the mCherry control, the expression of mCherry-SMCR8 substantially increased the level of C9orf72-V5 (Fig 6G). Importantly, under a 12-hr chase condition after treatment of the cells with the translation inhibitor cycloheximide, mCherry-SMCR8 dramatically stabilized the co-expressed C9orf72-V5 as compared with the mCherry control (Fig 6G). We also confirmed that the C9orf72-V5 protein was degraded through both proteosomal and lysosomal pathways, since inhibition of proteasomal degradation by MG132 treatment or inhibition of lysosomal degradation by Bafilomycin treatment stabilized the C9orf72-V5 protein (S7E Fig). These data indicate that C9orf72 and SMCR8 form a stable cognate protein complex that protects C9orf72 from degradation.
Given the connections between C9orf72 and SMCR8, we asked whether loss of SMCR8 plays a role in mTOR signaling similar to that of C9orf72. In accordance with the results from C9orf72-/- MEF cells, knockout of SMCR8 led to a similar defect. In SMCR8 KO HAP1 cells, the phosphorylation of S6K1 after amino acid treatment was significantly decreased when compared with control cells (Fig 7A and 7B).
Next, we investigated if loss of SMCR8 also affected autophagy. First, we examined LC3 levels after shRNA-mediated knockdown of SMCR8 in HEK293T cells by immunoblotting. As observed in C9orf72-/- MEFs and C9orf72 shRNA treated HEK293T cells, knockdown of SMCR8 led to a decrease in the ratio of LC3II to LC3I, when compared with cells treated with scrambled shRNA (Fig 7C and 7D). Additionally, Bafilomycin treatment of the cells under starvation showed a similar accumulation of LC3II with the SMCR8 knockdown as that of the control cells (Fig 7E). Thus, the autophagic flux appears to be intact in the absence of SMCR8 in this cell line.
In the present study, we have identified a function of C9orf72 in regulating mTOR signaling and autophagy. Loss of C9orf72 leads to deficiency in the phosphorylation of S6K1 and increase of TFEB protein levels and nuclear activity, demonstrating a regulatory role of C9orf72 in the mTOR signaling pathway upstream of autophagy.
We identified the major interacting partner of C9orf72 protein as SMCR8. The most structurally homologous proteins to SMCR8 and C9orf72 in the human proteome are folliculin (FLCN) and folliculin-interacting proteins (FNIP1 or 2), respectively [20, 21]. Like SMCR8 and C9orf72, FNIP and FLCN are DENN domain-containing proteins [20, 21] that interact with each other in a protein complex [29], that have also been shown to regulate autophagy and mTOR signaling [30, 31]. Since the FNIP and FLCN complex was shown to function as either GAP or GEF for the Rag GTPases in the mTORC1 pathway, we speculate that the C9orf72-SMCR8 complex may function in a similar fashion in autophagy and mTOR signaling.
Our results demonstrate that loss of C9orf72 can alter the dynamics of autophagy. We observed a relative increase in LC3I levels upon loss of C9orf72 (S2 Fig), in consistence with a recent report for LC3 levels in C9orf72 KO mouse liver and spleen tissues [39], which we interpret as an increase in autophagosome turnover instead of a decrease in LC3II formation. In support of this model, we did not observe a decrease in LC3II levels after Bafilomycin treatment under full nutrient conditions, suggesting that the formation of LC3II is intact (S3 Fig). Moreover, we observed increased autophagic flux in response to nutrient deprivation in C9orf72-/- cells (Fig 4). Consistent with our model of increased autophagic flux, we observed a loss of mTOR activity after loss of C9orf72, which is classically associated with increases in the autophagic pathway. In support of our finding, a recent study showed decreased mTOR signaling in C9orf72-depleted HeLa cells [40]. Importantly, we observed a substantial increase of TFEB and its lysosomal targets in C9orf72 knockout mice (Fig 3). As a master regulator of lysosome biogenesis, TFEB is known to promote cellular lysosomal capacity and autophagy [32]. Consistent with our findings, we also observed a decrease in levels of the autophagy receptor p62 in brain tissues from C9orf72 KO mice and observed a similar decrease in the C9orf72 KO MEFs. Interestingly, it was recently reported that loss of the SMCR8 homologue folliculin similarly results in decreased mTOR signaling and a TFEB-mediated enhancement of the lysosomal compartment [31].
There have been recent reports describing C9orf72’s functions in autophagy [39, 41–43], including a decrease in autophagy initiation as a result of knockdown of C9orf72 [41, 42]. These observations are not necessarily mutually exclusive to our present study. C9orf72 might play a multifunctional role in different steps of the autophagic pathways. While C9orf72 may influence the function of the FIP200/ULK1 autophagy initiation complex [41, 42], it could also regulate mTOR signaling and TFEB and thus promote autophagic flux, as observed in the present study. Furthermore, the manifestation of the phenotypes could be influenced by the dynamic nature and condition-dependent activity levels of autophagy pathways. Due to the reduced state of mTOR signaling in C9orf72-depleted cells, the increased autophagic flux of these cells could be more readily revealed under nutrient deprivation conditions, as employed in the present study. Notably, the autophagy receptor p62 is both a substrate of autophagy and a transcriptional target of TFEB [44], therefore it is subject to opposing regulation by upregulation of TFEB. Taken together, our study provides evidence that long-term loss of C9orf72 leads to physiological changes that are characterized by reduced mTOR activity, in consistence with increased TFEB signaling leading to enhanced cellular lysosomal capacity and autophagic flux.
Since multiple studies have reported that the hexanucleotide repeat expansion led to reduced expression of C9orf72 mRNAs and proteins in patient cells and brains [11–15], the defects associated with loss of C9orf72 protein function could contribute to the pathogenesis of relevant neurodegenerative diseases. Several studies have reported that neither mice lacking C9orf72 protein nor those expressing the human C9orf72 gene containing the HRE mutation exhibited major neuronal loss [17, 45–47], with the exception of one study reporting neurodegeneration in transgenic mice expressing HRE-containing C9orf72 [48]. Our observation that C9orf72 ablation changes LC3 levels in motor neuron cultures suggests that loss of C9orf72 might affect neuronal functions. Autophagy and nutrient sensing are essential for neuronal health and their alteration is an increasingly recognized feature in aging-related neurodegenerative diseases [49, 50]. Of note, several autophagy-related genes, including p62, optineurin, and TBK1, have been linked to ALS [51–53]. Proteinaceous inclusions positive for p62 are a pathologic feature in brains from patients carrying the C9orf72 HRE mutation [54]. Taken together, our findings suggest that C9orf72 protein has a function in the metabolic processes of the cell and reduction in its function may contribute to related age-dependent neurodegenerative diseases.
The animal protocol (MO15M165) was approved by the Johns Hopkins Animal Care and Use Committee following the National Research Council’s guide to the Care and Use of Laboratory Animals.
C9orf72 cDNA (HsCD00398737) was obtained from Arizona State University and SMCR8 cDNA (HsCD00347993) from Harvard Plasmid Repositories. The C9orf72 constructs were generated using the Gateway cloning system (ThermoFisher, Waltham, MA) with a C-terminal 3xFlag or V5 tag. The SMCR8 constructs were generated with an N-terminal Flag or mCherry tag using Gateway or classical cloning methods, respectively. All shRNAs were cloned into the pRFP-C-RS vector (Origene), which was modified to remove the RFP coding sequence via digestion with MluI and BglII followed by blunting and religation. The following shRNA sequences were used: 5’ctgtgttacctcctgaccagtcagattga 3’ (SMCR8); 5’cttccacagacagaacttagtttctacct 3’ (C9orf72). The autophagy luciferase assay plasmids were kindly provided by Brian Seed (Harvard) and the normalization plasmid pCMV-SEAP was from Addgene (24595, Alan Cochrane, University of Toronto). GFP-TFEB was obtained from Addgene (38119, Shawn Ferguson, Yale University). GFP-TFEB used for MEF experiments was described before [55]. For GFP-LC3, human LC3 was cloned into pEGFP-C1. RFP-Rab7 was generated from EGFP-Rab7 (a kind gift from Bo van Deurs at University of Copenhagen) by exchanging EGFP into RFP.
Mouse ES cell lines containing a heterozygous allele of 3110043O21Riktm1.1(KOMP)Mbp were obtained from the KOMP repository. The ES cells with a strain background of C57BL/6N-Atm1Brd were microinjected into blastocysts, and the germline-transmitted allele was maintained on the C57BL/6 background. Male mice bearing the original targeting allele were crossed with SOX2-Cre recombinase transgenic female mice (Jackson Laboratory, 008454) to remove the LoxP-flanked neomycin selection cassette. The resulting allele was bred to heterozygotes and homozygotes that were used in this study. The genotyping primers were the following: gaatggagatcggagcacttatgg (wild-type, forward), gccttagtaactaagcttgctgccc (wild-type, reverse), gcacaagctatgttcatttgg (KO, forward), gactaacagaagaacccgttgtg (KO, reverse).
For the low-protein diet assay, 16 week old, gender-matched littermates were fed a low-protein diet (Test Diet 5767, 5% protein) or standard chow for 4 weeks prior to tissue collection. Mouse tissue was lysed in modified RIPA buffer (50 mM Tris pH 6.8, 150 mM NaCl, 0.5% SDS, 0.5% Sarkosyl, 0.5% NP40, 20 mM EDTA, Roche protease inhibitors) using a Dounce homogenizer, sonicated, and used for further analysis. For the survival analysis, Kaplan-Meyer curves were generated using GraphPad Prism software.
All cells were maintained in DMEM supplemented with 10% FBS unless otherwise noted. The SMCR8 knockout HAP1 cells (HZGHC003606c011) were created at Horizon Genomics (Vienna, Austria) by using CRISPR/Cas9 and maintained in IMDM supplemented with 10% FBS. All cell lines were cultured in 95% O2/5% CO2. Cell lines were transfected using Lipofectamine 2000 (ThermoFisher) according to the manufacturer’s instructions.
Mouse embryonic fibroblasts were isolated from Day 13 embryos by trypsin digestion and their genotypes confirmed by PCR. The lines were immortalized by transfecting cells with the SV40-T antigen-expressing plasmid pSG5 Large T using Lipofectamine 2000. The cells were passaged at least 5x to ensure the homogeneity of the cell population before use in experiments. To isolate embryonic stem cells, 14-week old C9orf72 heterozygous females were treated with Pregnant Mare Serum Gonadotropin via intraperitoneal injection followed by injection 24 hours later with human chorionic gonadotrophin to induce superovulation prior to mating with C9orf72 heterozygous males. Embryos were collected 48 hours after the second injection at the transgenic core facility at Johns Hopkins University and the genotypes confirmed by PCR. Wild type and C9ORF72-/- ES cells were cultured on 0.1% gelatin coated plates in 2i media consisting of half of DMEM/F12 and half of Neurobasal media containing N2-supplement (ThermoFisher Scientific 17502048), B-27 supplement (ThermoFisher Scientific 17504044), 0.05% BSA (ThermoFisher Scientific 15260037), 50 units Penicillin-Streptomycin, 1 μM PD03259010 (Stemgent 04–0006), 3 μM CHIR99021 (Stemgent 04–0004), 2 mM Glutamine, 150 μM Monothioglycerol (Sigma M6145) and 1,000 U/ml LIF.
Motorneuron differentiation protocol was modified from a previously reported induction protocol using retinoic acid and Smoothened agonist (SAG, Millipore) [56]. Briefly, 1 X 106 ES cells were harvested by dissociating with 0.05% trypsin-EDTA (ThermoFisher) and cultured in suspension condition in DFK5 media (DMEM/F12 based media containing 5% knockout serum replacement, 1 x insulin transferrin selenium (ThermoFisher), 50 μM nonessential amino acids, 100 μM β-mercaptoehanol, 5 μM thymidine, 15 μM adenosine, 15 μM cytosine, 15 μM guanosine and 15 μM uridine) for 48 hours. After two days, the resulting embryonic bodies were treated with 2 μM retinoic acid and 600 nM of SAG in fresh DFK5 media and cultured another 4 days. Media was replaced every two days. For experiments, 1.5 x 106 cells were plated on each well of laminin-coated 6 well plates in DFK5 media containing 5 ng/mL glial-derived neurotrophic factor (GDNF; Peprotech), 5 ng/mL brain-derived neurotrophic factor (BDNF; Peprotech), 5 ng/mL neurotrophin-3 (NT-3; Peprotech) for 24 hours. After 24 hours, media were changed with DFKNB media consisting of half of DFK5 media and half of Neurobasal media with B27, 5 ng/mL GDNF, 5 ng/mL of BDNF and 5 ng/mL of NT-3.
All cells were starved using Earles’s balanced salt solution (EBSS; Sigma) for 50 min. Amino acid stimulation was applied by treating cells with essential amino acids (Gibco) and non-essential amino acids (Quality Biologicals) in EBSS. Amino acids were diluted to match DMEM concentrations. Cells were treated in EBSS plus amino acids for 10–20 min prior to lysate collection. Lysates were processed as described above except that Phospho-stop inhibitor tablets (Roche) were added to the lysis buffer.
Wild type or C9orf72-/- MEF cells were transfected with GFP-LC3 and RFP-Rab7, and cells were treated with DMEM containing 10% FBS (full medium; FM) or EBSS (nutrient deprivation; ND), in the presence or absence of the lysosomal inhibitor Bafilomycin A1 (Baf) for 3 hours, before being fixed with 4% paraformaldehyde. High resolution images were acquired using a Z sweep function, permitting acquisition of total cellular fluorescence using a DeltaVision Elite microscope (GE Healthcare) with 60× PlanApo NA 1.4 Oil objective lens (Olympus) and images were deconvolvd using SoftWoRx software. Subsequently, individual cells were manually segmented; LC3-positve vesicles in the green channel and Rab7-positive vesicles in the red channel. Using a boolean function, the overlap between these segmented images was used to generate a third mask corresponding to co-localized LC3 and Rab7 vesicles (Yellow image) which are autolysosomes. Single vesicle areas were calculated from LC3, Rab7 and co-localization masks, and mean values for each experiment were normalized to ND or FM, as indicated.
HEK293T cells were grown on class coverslips and transfected with the indicated constructs as described above. Images were captured using an SP8 confocal microscope (Leica) and processed using ImageJ software. For GFP-TFEB imaging in HEK293T cells, live cells were imaged while maintained in phenol red free DMEM containing 10% FBS. For LysoTrackerBlue staining, 50nM of LysoTrackerBlue was added into the media for an hour and the media was changed before imaging.
Cells were collected in modified RIPA lysis buffer (50 mM Tris pH 6.8, 150 mM NaCl, 0.5% SDS, 0.5% Sarkosyl, 0.5% NP40, 20 mM EDTA, Roche protease inhibitors) and sonicated using a Diagenode Bioruptor for 15 min (high setting, 30 sec pulse, 3x 5 min) and the resulting lysates were centrifuged at 16,000 x g for 10 min at 4°C. For mTOR assays, cells were collected in HEPES lysis buffer (40 mM HEPES pH 7.4, 2 mM EDTA, 1% Triton, Roche protease and PhosphoStop inhibitors) and centrifuged at 16,000 x g for 10 min at 4°C. Protein concentrations were determined using the bicinchonic acid assay (ThermoFisher). For GFP-TFEB nuclear import analysis, cells were fractionated using Subcellular Fractionation Kit for Cultured Cells (ThermoFisher) following the manufacturer’s protocols. Then the cytoplasmic and membrane proteins were combined as the cytosolic fraction and the nuclear soluble and chromatin bound proteins were combined as the nuclear fraction. PARP was used as a nuclear marker and Caspase 3 as a cytoplasmic marker. For autophagic flux determination, wild type and C9orf72-/- MEFs were subjected to fresh fully supplemented medium or nutrient deprivation (EBSS) for 3 hours. Antibodies used were: mouse anti-Flag, rabbit anti-SMCR8 (Sigma), mouse anti-GFP, rabbit anti-GAPDH (ThermoFisher), rabbit anti-C9orf72, mouse anti-actin (Santa Cruz), rabbit anti-p62, rabbit anti p-70S6K, rabbit anti p-p70S6K, rabbit anti-PARP, rabbit anti-Caspase 3, rabbit anti-DYKDDDDK (Cell signaling), rabbit anti-V5 (Novus), mouse anti-V5 (Invitrogen), mouse anti-TFEB (Mybiosources), mouse anti-Lamp1 (Hybridoma Bank; #H4A3-s), mouse anti-Lamp2 (Hybridoma Bank, #H4B4-s), and rabbit anti-LC3B (Abcam).
HEK293T cells were incubated in heavy (13C6,15N4 L-Arginine, 13C6,15N2 L-Lysine) or light (12C6,14N4 L-Arginine, 12C6,14N2 L-Lysine) DMEM and verified for near-completion of labeling by mass spectrometry. Heavy isotope-labeled cells were transfected with C9orf72-Flag and light isotope-labeled cells were mock transfected with Lipofectamine. After immunoprecipitation with Flag-tag beads, the resulting immunoprecipitates were pooled, concentrated, separated via SDS-PAGE, and subjected to trypsin in-gel digestion. The digested samples were subjected to LC-MS/MS analysis on an Orbitrap Elite mass spectrometer coupled with Easy nLC II liquid chromatography system. The mass spectrometry data were analyzed using the Proteome Discoverer 1.4 software suite against human Refseq 59 protein database. A 1% peptide-spectrum-match and peptide-level false discovery rate was applied for data analysis.
Cells were lysed in IP buffer (50 mM Tris pH 7.4, 150 mM NaCl, 1% Triton, Roche protease inhibitors), incubated for 30 min on ice and centrifuged at 16,000 x g prior to immunoprecipitation. For Flag immunoprecipitations, the resulting supernatants were added to Flag-conjugated beads (Sigma, St. Louis, MO) and incubated for 2 h at 4°C with gentle rotation. The beads were washed 5x with IP buffer and the immunoprecipitates eluted by incubating the beads with SDS-PAGE loading dye for 5 min at 95° C. For SILAC analysis, immunoprecipitates were eluted using Flag peptide (Sigma) at 5μg/μl. For SMCR8 immunoprecipitation, anti-SMCR8 antibody (Abcam) was incubated with protein A Sepharose beads (BioRad, Hercules, CA) and incubated at room temperature for 2 h and the beads treated as described above.
The ATG4B-dependent processing of LC3 in autophagy was quantified with a Gaussia luciferase release assay [57, 58]. ATG4B-induced proteolytic cleavage of an actin-anchored LC3-luciferase fusion protein (Act-LC3-Gluc) releases the Gluc fragment and enables its secretion into the cell medium. Cells were transfected with shRNA or scrambled control and Act-LC3-Gluc or control Act-Gluc plasmid together with the secreted alkaline phosphatase normalization control, CMV-SEAP. Cell medium (150 μl) was withdrawn 48–72 h after transfection and the luciferase and SEAP in the medium were analyzed by using Gaussia luciferase assay kit (New Englabnd Biolabs) and the Phospha-light SEAP reporter system (ThermoFisher) using a microplate reader (Synergy H1, Bio-Tek).
Gender matched four month old mice were intracardially perfused with ice-cold 4% paraformaldehyde. Brains were removed and post-fixed and equilibrated with 30% sucrose. Sections were prepared using Cryostar NX70 (ThermoScientific). Sections were washed with PBST three times to permeabilize cells, pre-incubated with 10% anti-goat serum for an hour at RT, incubated with an anti-p62 antibody (Cell Signaling) for overnight at 4°C, and then incubated with Alexa488-congugated secondary antibody after washing three times with PBS. Images were obtained using an SP8 confocal microscope (Leica) after samples were washed three times with PBS and mounted with Vectashield.
Total RNA was isolated from cells with the RNeasy Plus Mini kit and cDNAs were synthesized with the QuantiTect reverse transcription kit (Qiagen). Primers for quantitative RT-qPCR were from PrimerBank unless otherwise noted (S2 Table). RT-qPCRs were performed on a BioRad thermal cycler with iQ SYBER Green PCR mix (BioRad).
All quantitation and statistical tests were performed using ImageJ and GraphPad Prism software (Version 6.0). The p-values for all experiments were obtained using Student’s t tests unless indicated otherwise.
|
10.1371/journal.pgen.1000187 | The Genomic Distribution and Function of Histone Variant HTZ-1 during C. elegans Embryogenesis | In all eukaryotes, histone variants are incorporated into a subset of nucleosomes to create functionally specialized regions of chromatin. One such variant, H2A.Z, replaces histone H2A and is required for development and viability in all animals tested to date. However, the function of H2A.Z in development remains unclear. Here, we use ChIP-chip, genetic mutation, RNAi, and immunofluorescence microscopy to interrogate the function of H2A.Z (HTZ-1) during embryogenesis in Caenorhabditis elegans, a key model of metazoan development. We find that HTZ-1 is expressed in every cell of the developing embryo and is essential for normal development. The sites of HTZ-1 incorporation during embryogenesis reveal a genome wrought by developmental processes. HTZ-1 is incorporated upstream of 23% of C. elegans genes. While these genes tend to be required for development and occupied by RNA polymerase II, HTZ-1 incorporation does not specify a stereotypic transcription program. The data also provide evidence for unexpectedly widespread independent regulation of genes within operons during development; in 37% of operons, HTZ-1 is incorporated upstream of internally encoded genes. Fewer sites of HTZ-1 incorporation occur on the X chromosome relative to autosomes, which our data suggest is due to a paucity of developmentally important genes on X, rather than a direct function for HTZ-1 in dosage compensation. Our experiments indicate that HTZ-1 functions in establishing or maintaining an essential chromatin state at promoters regulated dynamically during C. elegans embryogenesis.
| To fit within a cell's nucleus, DNA is wrapped around protein spools composed of the histones H3, H4, H2A, and H2B. One spool and the DNA wrapped around it are called a nucleosome, and all of the packaged DNA in a cell's nucleus is collectively called “chromatin.” Chromatin is important because it modulates access to information encoded in the underlying DNA. Spools with specialized functions can be created by replacing a typical histone component with a variant version of the histone protein. Here, we examine the distribution and function of the C. elegans histone H2A variant H2A.Z (called HTZ-1) during development. We demonstrate that HTZ-1 is required for proper development, and that embryos are dependent on a contribution of HTZ-1 from their mothers for survival. We mapped the location of HTZ-1 incorporation genome-wide and found that HTZ-1 binds upstream of 23% of genes, which tend to be genes that are essential for development and occupied by RNA polymerase. Fewer sites of HTZ-1 incorporation were found on the X chromosome, probably due to an under-representation of essential genes on X rather than a direct role for HTZ-1 in X-chromosome dosage compensation. Our study reveals how the genome is remodeled by HTZ-1 to allow the proper regulation of genes critical for development.
| In genomes ranging from protozoa to humans, specialized regions of chromatin are created by the local incorporation of variant histones into nucleosomes. The histone H2A variant H2A.Z is one such highly conserved variant, though the biophysical and biological function of H2A.Z incorporation into chromatin remains unresolved.
Evidence from Tetrahymena suggests a function for H2A.Z in transcriptional activation due to its localization to the transcriptionally active macronucleus [1]–[3]. This function is consistent with genome-wide studies of Htz1 occupancy in S. cerevisiae (hereafter “yeast”), which revealed Htz1 incorporation flanking a nucleosome-free region upstream of most genes. It has been hypothesized that H2A.Z-containing nucleosomes may contribute to transcriptional activation by being less stable than H2A-containing nucleosomes [4]–[6]. However, others have reported that H2A.Z-containing nucleosomes are in fact slightly more stable than canonical nucleosomes [7]–[9]. This seeming contradiction may have been resolved by studies examining H2A.Z in combination with the histone H3 variant H3.3. In combination with histone H3, H2A.Z nucleosomes were at least as stable as H2A nucleosomes, but the combination of H2A.Z and H3.3 results in highly unstable nucleosomes [4]. This instability in conjunction with H3.3 could facilitate timely and efficient gene activation. Indeed, in yeast cells lacking H2A.Z, the activation of genes in response to heat shock or galactose is delayed, and recruitment of RNA polymerase II and TATA-binding protein to responsive promoters is diminished [10],[11]. H2A.Z is also required for a form of “transcriptional memory” in yeast, in which recently transcribed chromatin is retained at the nuclear membrane to allow rapid re-activation of the gene [12]. Recent high-resolution mapping of H2A.Z in human cells also revealed a positive correlation between H2A.Z occupancy and transcription, providing additional support for an H2A.Z function in transcriptional activation [13].
Despite the wealth of evidence for a positive association between H2A.Z and transcription, other genetic and cytological evidence suggests that H2A.Z also functions in gene silencing. The functional homolog of H2A.Z in Drosophila, H2Avd, is localized to both euchromatin and heterochromatin on polytene chromosomes, including the heterochromatic chromocenter [14],[15]. By genetic criteria, H2Avd is considered to have a repressive function. H2Avd mutations are enhancers of Polycomb mutant phenotypes, suppressors of Trithorax group mutant phenotypes, and suppressors of position-effect variegation [16]. Further evidence for a repressive function is found in mice, where H2A.Z promotes heterochromatin protein HP1α binding and co-localizes with HP1 at pericentric heterochromatin [17],[18]. In mammalian cells, mono-ubiquitylation of the H2A.Z C-terminus may distinguish “repressive H2A.Z” from “activating H2A.Z”, particularly on the silent X chromosome [19]. Even within the yeast literature, there are conflicting conclusions regarding correlation with transcriptional activity and RNA Polymerase II. One study found no correlation between Htz1 occupancy and transcription rate of the downstream gene [11], while others reported an inverse correlation with transcription rate [20]–[22].
The resolution of these apparently contradictory activating and silencing functions could be explained by a requirement for H2A.Z in regulating the precise timing and kinetics transcription, rather than simply promoting an “on” or “off” transcriptional state. This potentiation of transcription would be especially critical during periods of dynamic transcriptional regulation, such as occurs in development and environmental responses. There is growing evidence for this hypothesis. In C. elegans, knockdown of htz-1 by RNAi caused expression of genes dependent on the FoxA transcription factor PHA-4 to be delayed [23]. Furthermore, HTZ-1 and components of the C. elegans Swr1 complex (SSL-1) required for HTZ-1 deposition have been identified in genetic screens for suppressors of vulval induction, a process highly dependent on precise timing of transcriptional cascades and tightly coordinated with cell divisions [24],[25].
Another clue to the function of H2A.Z may lie in the fact that it is required for viability in all metazoans tested [26]–[29], but is not required for viability in single-celled yeast. A lack of H2A.Z during metazoan development typically causes defects that lead to late embryonic lethality [23],[26],[27],[29],[30]. This is consistent with expression of H2A.Z in mice, where the undifferentiated cells of the inner cell mass have low H2A.Z protein levels, with H2A.Z protein levels increasing as the cells differentiate into extraembryonic endoderm [17].
Whether H2A.Z has been associated with gene activation or repression in one study versus another may not represent a universal regulatory function for H2A.Z, but may instead be a reflection of the specific biological conditions under which the function of H2A.Z was observed in a given experiment, and the temporal resolution of the particular assays employed. In this light and with a focus on development, we used Chromatin ImmunoPrecipitation on DNA microarrays (ChIP-chip), genetic mutation, and RNAi to interrogate the function of HTZ-1 during embryogenesis in C. elegans.
HTZ-1 knockdown by RNAi has been previously shown to cause embryonic lethality [23]. To further characterize the function of HTZ-1 (R08C7.3) in C. elegans development, we analyzed animals harboring a deletion in the C. elegans htz-1 gene. The mutant htz-1(tm2469) contains a deletion of 345 bp of the htz-1 gene, thereby eliminating 97 of the 140 predicted amino acids and making it a likely genetic null. The majority of homozygous htz-1(tm2469) offspring from htz-1(tm2469)/+ heterozygotes (denoted as maternal +; zygotic −, or M+Z−) animals are rescued from embryonic lethality by a maternal contribution of HTZ-1. These rescued animals develop into worms exhibiting grossly normal morphology and germ cell proliferation until late adulthood (Figure 1A–B). Of the M+Z− animals that reach adulthood, 80% are sterile and do not generate any embryos, instead producing unfertilized oocytes that eventually fill the uterus (Figure 1B). In 20% of the rescued animals, M−Z− embryos are observed in the uterus (Figure 1C). None of the embryos produced by M+Z− mothers were expelled from the uterus onto plates, indicating that the M+Z− mothers have an egg-laying defect (Egl). Somewhat unexpectedly, 28% of the M−Z− embryos (n = 32) progressed through embryogenesis to produce a few hatched larvae. All of these M−Z− escapers arrest at the first larval stage (Figure 1D–E). The M−Z− embryos that hatched tended to arise from the first few eggs produced by M+Z− mothers, suggesting that in these animals HTZ-1 were still maternally provided at very low levels, but subsequent divisions of the germ cell precursors diluted HTZ-1 such that later embryos received a level below that required for viability.
The viability and semi-fertility of the htz-1(tm2469) M+Z− offspring suggested that the maternal load of HTZ-1 received by an embryo is sufficient to allow it to reach adulthood with defects limited to germ cells and specification of cells in post-embryonic lineages, for example vulval development. To test this, we targeted the maternal complement of htz-1 mRNA using RNAi. Direct injection of dsRNA into the gonad of adult wild-type animals produced a more severe phenotype than was observed in M+Z− offspring. Instead, the RNAi phenotypes are consistent with those observed in htz-1(tm2469) M−Z− embryos. Specifically, embryonic lethality was observed for 70% of the embryos, with the remaining animals dying as larvae (Figure 1F–G; Text S1). We verified that the htz-1 dsRNA injections did not cross-react with H2A mRNA by showing that expression of a GFP-tagged version of H2A was not affected (Figure 2A–C). We interpret the progression of phenotypes resulting from either RNAi treatment or genetic mutation to indicate that HTZ-1 is required for both embryogenesis and for post-embryonic development. We propose that the occasional escape from lethality occurs due to perdurance of maternal HTZ-1 protein or RNA for as long as two generations, or in the case of RNAi, a failure to completely eliminate HTZ-1 protein or message in the offspring of injected mothers (Discussion).
HTZ-1 RNA is abundant in the form of a maternal contribution, and remains abundant throughout the majority of embryogenesis, suggesting that the function of HTZ-1 in development is widespread [31]. To investigate the distribution of HTZ-1 protein, we generated polyclonal antisera specific to a unique peptide sequence in the C-terminal region of HTZ-1 (Methods). The antibody recognized a single band of 15 kD on western blots of C. elegans protein extract, corresponding to the predicted molecular weight of HTZ-1 (Figure S1). Using these antibodies, we stained whole embryos and adults and found that HTZ-1 protein is present in all cell types throughout all stages of development. HTZ-1 protein levels are low in early embryos (1–12 cell), but increase as development progresses (Figure 2D–F).
HTZ-1 protein becomes detectably incorporated into chromosomes by the four-cell stage, coincident with the onset of zygotic transcription. This occurs in both wild-type and M+Z− embryos, demonstrating that zygotic transcription of htz-1 itself is not required for incorporation of HTZ-1 protein into chromatin. In wild-type adults, HTZ-1 protein is observed in both somatic and germline precursor cells (data not shown). No HTZ-1 protein was observed by immunofluorescence in M+Z− adult gonads or their M−Z− embryos (Figures G–O). In addition, no protein staining was observed in the offspring of animals injected with HTZ-1 RNAi (Figure 2P–R).
The low levels of HTZ-1 protein in young embryos, despite abundant htz-1 mRNA, suggests that much of the maternal contribution is RNA-based, with HTZ-1 protein levels controlled post-transcriptionally (Figure 2D–F). Another case in which HTZ-1 protein levels do not depend on zygotic transcription can be inferred from the presence of HTZ-1 protein in the germline precursors (P lineage). In these cells, HTZ-1 protein is present in chromatin at levels comparable to the surrounding somatic blastomeres, despite the repression of zygotic mRNA production in the P lineage (Figure 2D–F) [32]. HTZ-1 protein is also observed in the chromatin of the primordial germ cells Z2 and Z3 (data not shown), which undergoes a dramatic erasure of histone H3 modifications during development [33],[34].
To determine the genomic locations at which HTZ-1 functions, we performed ChIP-chip of HTZ-1 from extracts of wildtype N2 C. elegans embryos (Methods). For detection of ChIP-enriched loci, we used DNA microarrays consisting of 50-bp oligonucleotide probes that tile across the entire genome with 86-bp start-to-start spacing (Methods). Peaks of HTZ-1 binding were identified using ChIPOTle [35]. Throughout the genome, 5163 sites of HTZ-1 incorporation were found, with 85% of the peaks occurring within intergenic regions. Intergenic regions are defined as those that occur outside the boundaries defined by the translation start and stop sites of annotated transcripts or predicted genes. Under this definition, intergenic regions comprise 58% of the bases in the genome. Of the peaks within an intergenic region, 71% were within the 2-kb upstream of an annotated translation start site, 25% were within 2-kb of the translation stop, and only 4% were greater than 2-kb upstream of a translation start site. Among the 15% of peaks found to occur within an annotated transcription unit, most occurred near the 5′ end (median +545 bp downstream of the annotated translation start site). Therefore, like yeast Htz1, C. elegans HTZ-1 is preferentially incorporated into intergenic regions, specifically at promoters (Figure 3A).
We next investigated whether HTZ-1 was incorporated specifically at sites of transcriptional initiation. The majority of transcription initiation sites are not well-annotated in C. elegans, due in part to the prevalence of trans-splicing [36]. Therefore as a proxy for transcription initiation sites, we plotted HTZ-1 binding relative to annotated translation start codons. On average, the peak of HTZ-1 incorporation occurs just upstream of the translation start codon (Figure 3B), which we interpreted to indicate incorporation at or near sites of transcription initiation. To further test whether the observed signal represents sites of transcription initiation, we took advantage of a unique feature of the C. elegans genome. Approximately 15% of C. elegans genes are predicted to reside in operons that are transcribed as a large polycistronic pre-mRNA, which is then trans-spliced into mRNAs for the individual genes [37]. We plotted HTZ-1 incorporation relative to the first gene in operons, where transcription is expected to initiate, and also plotted incorporation relative to internal genes, where transcription is not expected to initiate. Indeed, HTZ-1 incorporation is generally observed upstream of the first gene in an operon, and does not generally occur upstream of internal genes (Figure 3C), indicating that C. elegans HTZ-1 is incorporated primarily at or near sites of transcription initiation. We also observed some important exceptions to this general rule, which are discussed below.
Currently most C. elegans operons are identified primarily by two criteria: the appearance of two or more genes in close proximity that are transcribed on the same strand, and the isolation of a downstream RNA transcript with an SL2 trans-spliced leader [38],[39]. In this way, a total of 1118 putative operons have been identified (genome release ws170). However, these criteria are imperfect, and do not provide information about genes that may be regulated both as part of an operon and by their own independent promoter. Independent transcription events within operons have been difficult to detect because the 5′ ends of mRNAs produced by either trans-splicing of a poly-cistronic mRNA or an independent transcription event are not readily distinguishable.
To identify genes that are likely to be regulated both as part of an operon and individually, we examined incorporation of HTZ-1 at internally encoded genes of annotated operons. Overall, 75% of operons contained at least one HTZ-1 peak. A gene within an operon was more than twice as likely as a non-operon gene to have an HTZ-1 peak at its promoter (Figure 4A–B). Of operons containing at least one site of HTZ-1 incorporation, 85% contained a peak upstream of the first gene, as one might expect. However, 49% of operons with HTZ-1 incorporation at the first gene also exhibited an internal peak of HTZ-1 incorporation. This strongly suggests internal transcription start sites at 416 (37%) of the currently annotated operons (Figure 4B, Table S2). Because some operons contain multiple internal HTZ-1 peaks, this represents a total of 455 putative independently regulated genes within annotated operons. This is likely to be an underestimate, since the HTZ-1 localization data is derived only from embryonic extracts, meaning that genes and operons regulated specifically in adults or germ cells are not represented.
The unexpectedly high number of individually regulated genes within operons may to some extent reflect a mis-annotation of operons based on traditional criteria. To show that internal HTZ-1 incorporation can occur at verified operons, we examined CEOP1456, one of the first characterized operons, supported by cistronic RNA evidence [37],[40]. In this well-characterized operon, both HTZ-1 and RNA Polymerase II occupy the chromatin immediately upstream of the internal kin-10 gene, strongly suggesting independent regulation (Figure 4C). Recently, differential regulation of genes driven by internal operon promoters was demonstrated using a GFP reporter assay [41]. We find that one-third of these internal promoters are occupied by HTZ-1 in embryos (Table S2). A time-course of the early embryonic transcription [31] provides evidence that genes within operons that contain multiple sites of HTZ-1 incorporation exhibit differential expression (Figure S2).
In contrast to yeast, in which Htz1 is incorporated into nearly every promoter [42], our ChIP-chip data indicate that HTZ-1 is incorporated into the promoters of only 23% of C. elegans genes (Methods). To determine what might be held in common among the particular subset of genes that were occupied by HTZ-1, peaks were annotated to gene promoters, assigned Gene Ontology (GO) terms according to the nearest downstream gene, and evaluated with GO::TermFinder [43]. To avoid ambiguous gene assignments, only peaks annotated to unidirectional promoters or within coding regions were used in the input set. We found that GO terms associated with metazoan development and positive regulation of growth were strongly over-represented among HTZ-1 bound genes, while no overrepresented GO term was associated with the non-HTZ-1 bound genes (Table 1, Table S1). This finding suggests that HTZ-1 functions preferentially at the promoters of genes essential for growth and development.
We next sought to examine the relationship between HTZ-1 occupancy at promoters and transcriptional activity during embryogenesis. We found that, in general, transcript levels [31] were positively correlated with HTZ-1 promoter occupancy (Figure 5A; Spearman rank-order correlation = 0.35). A positive correlation was also observed between RNA levels reported by a completely independent study [44] and HTZ-1 occupancy (Figure S3). Despite the positive overall correlation between occupancy and transcript levels, the relationship becomes negative at promoters of genes with very high transcript abundance (Figure 5A). This observation is consistent with a general loss of nucleosomes upstream of highly transcribed genes [45],[46].
We sought to establish a more direct link between HTZ-1 occupancy and transcription, so we determined the genome-wide occupancy of RNA polymerase II by ChIP-chip using an antibody specific to the C-terminal domain heptapeptide (8WG16, Methods). At gene promoters, HTZ-1 occupancy was strongly correlated with RNA Polymerase II occupancy (Figure 5B). In fact, the correlation was stronger than that observed between HTZ-1 occupancy and transcript levels (Spearman rank-order correlation = 0.57). Consistent with the correlation with transcript levels, at the promoters most highly occupied by RNA Polymerase II, the correlation with HTZ-1 occupancy was negative. Again, this observation is likely due to general nucleosome loss at the promoters of highly transcribed genes, for example those that encode the histone and ribosomal proteins [45],[46]. Temporal regulation gene expression during embryogenesis may also affect this correlation and is considered in the Discussion.
To further illustrate the relationship between HTZ-1 localization and polymerase occupancy, the 4650 genes with HTZ-1 incorporated into their promoters were aligned according to their translation start site, and average RNA polymerase II occupancy relative to the start site was plotted (Figure 5C). HTZ-1-occupied promoters were on average occupied by RNA Polymerase II, whereas genes lacking HTZ-1 were not (Figure 5D). At promoters occupied by HTZ-1, the average peak of HTZ-1 occupancy was at negative 12 bp relative to the translation start, while the average peak of RNA Polymerase II occupancy was slightly upstream at negative 98 bp (Discussion).
An important consideration in interpreting these relationships is that our experiments were performed using extract derived from a mixed population of embryos composed of many cell types. Therefore, our results are a projection of HTZ-1 occupancy throughout embryogenesis and represent a temporal and spatial average of the relationship between HTZ-1, RNA Polymerase II, and transcription (Discussion and Text S1).
To examine if HTZ-1 occupied promoters direct a stereotypic pattern of gene expression, we compared HTZ-1 occupancy, RNA Polymerase II occupancy, and transcription at each gene using a published time-course of transcript abundance during embryonic development [31]. Promoters occupied by HTZ-1 were clustered according to our RNA Polymerase II promoter occupancy data and the change in transcript abundance relative to the onset of zygotic transcription. To avoid ambiguity, transcripts that were highly maternally loaded (>100 parts per million (ppm)) were removed from analysis. Consistent with the aggregate analysis, RNA Polymerase II is abundant at most HTZ-1 occupied genes (Figure 6A), while promoters at which HTZ-1 is not incorporated generally lack RNA Polymerase II (Figure 6B). However, a large proportion of genes downstream of promoters occupied by both HTZ-1 and RNA polymerase II produce low transcript levels (Figure 6A), and conversely some genes produce high transcript levels despite low levels of HTZ-1 and RNA polymerase II at their promoters (Figure 6B; Discussion). Therefore, while HTZ-1 is strongly linked to RNA Polymerase II occupancy in aggregate, HTZ-1 bound promoters do not specify a stereotypic pattern of transcriptional regulation during development, suggesting that RNA polymerase occupancy and transcript levels are decoupled at some promoters.
The sex chromosomes are often sites of specialized chromatin, harboring unique histone variants and chromatin modifications. To determine whether HTZ-1 was differentially localized to X, we co-stained embryos with anti-HTZ-1 in combination with either anti-DPY-27, which marks the X chromosomes in embryos of greater than about 30 cells (Figure 7A–D), or anti-MES-4, which marks the autosomes but not X chromosomes in early embryos (Figure 7E–H). In embryos that had initiated somatic dosage compensation, HTZ-1 incorporation was noticeably reduced on the X chromosomes, which was marked by DPY-27 staining (Figure 7A–D). However, co-staining with MES-4 revealed HTZ-1 under-representation on X even before the onset of somatic dosage compensation (Figure 7E–H). These results indicate that in both early-stage embryos before the onset of dosage compensation and late-stage C. elegans embryos after dosage compensation is established, there is significantly less HTZ-1 associated with the X than with autosomes.
We next aimed to ensure that reduction of HTZ-1 we observed on the X chromosome by immunofluorescence was not due to epitope exclusion. This concern was prompted by reports that mammalian H2A.Z on the inactive X chromosome is ubiquitylated, and that this modification can interfere with recognition by antibodies raised against a C-terminal peptide epitope [19]. Our antisera were also raised against a C-terminal peptide. To address this concern, we co-stained embryos expressing a HTZ-1:YFP transgene with anti-DPY-27 and anti-YFP antibodies. We observed a similarly reduced YFP signal coincident with regions of DPY-27 signal. This serves as independent evidence that within in the same nucleus, X chromatin has less HTZ-1 incorporation than autosomes (Figure S4).
Two explanations for the under-incorporation of HTZ-1 on X immediately come to mind. One is that less HTZ-1 is incorporated on X as part of the C. elegans dosage compensation mechanism. A second explanation, which we favor for the reasons presented below, is that genes important for development, whose promoters tend to be occupied by HTZ-1, are under-represented on the X chromosome [47]–[50]. To distinguish these possibilities, we examined at high resolution the sites of HTZ-1 incorporation on the X chromosomes relative to the autosomes (Figure 7I).
One possible variation of the “dosage compensation” hypothesis predicts that sites of HTZ-1 incorporation are excluded or diminished on X as a consequence of the transcriptional repression imposed by the DCC. In this case, one would expect HTZ-1 occupancy on X to be excluded from sites occupied by the dosage compensation machinery [51]. Contrary to this prediction, we found strong co-localization of HTZ-1 incorporation and DCC binding, such that over 62% of HTZ-1 peaks are coincident with a DPY-27 peak (Figure 7J–K, Figure S7). The highly concordant binding pattern of HTZ-1 and the DCC on X would appear to rule out a function for HTZ-1 as a direct negative regulator of DCC binding to autosomes (Discussion).
We then considered the possibility that HTZ-1 incorporation is in fact a requirement for the loading of the DCC onto X. However, there are far more sites of DCC localization on X than HTZ-1 incorporation, meaning that most DCC-bound loci are not sites of HTZ-1 localization. For example, while both HTZ-1 and DPY-27 are incorporated at the X-linked dpy-23 promoter, HTZ-1 is not incorporated at the well-characterized DCC recruitment site rex-1, which occurs just 5 kb downstream of dpy-23 (Figure 7L) [51],[52]. We also examined in more detail apl-1 and lin-15, two of the few genes known with some certainty to be dosage compensated [53],[54]. Although the DCC and RNA Polymerase II are present at both loci, HTZ-1 is present at lin-15, but not at apl-1 (Figure S5), reinforcing the interpretation that HTZ-1 is not required for dosage compensation. Conversely, the under-representation of HTZ-1 on X is not dependent on the dosage compensation process, because it is evident in the germline and before the onset of somatic dosage compensation (Figure 7H).
The alternative “developmental gene” hypothesis for the under-incorporation of HTZ-1 on X is based on the observation that only about half as many essential genes occur on X as would be expected to occur on an autosome of the same size (201 vs. 562 expected, wormbase release ws170) [47]–[50]. This hypothesis predicts that there would be fewer sites of HTZ-1 incorporation on X, but that those that do occur on X behave like those on autosomes. The X harbored 495 HTZ-1 peaks, about half the number expected from a hypothetical autosome with the size and gene density of X (p-value = 2.05×10−43 and 8.09×10−93 respectively, Figure 7I). There was no significant difference between the median height and width of HTZ-1 peaks on X (z-score = 2.28 and 774 bp, respectively) as compared to the median height and width of HTZ-1 peaks on autosomes (z-score = 2.20 and 860 bp, respectively) (Figure S8). This indicates that while HTZ-1 incorporation occurs at fewer loci on X, where it does occur the degree of incorporation is the same as the autosomes.
The most parsimonious explanation for the under-representation of HTZ-1 on the X is that the types of genes that require HTZ-1 for proper regulation are themselves under-represented on the X chromosome. Nonetheless, HTZ-1 is likely to have an indirect function in the dosage compensation process by affecting the regulation of genes required for dosage compensation. Strong HTZ-1 incorporation is observed at the promoters of sdc-1, sdc-2, sdc-3, dpy-27, mix-1, and dpy-30, all of which are required for dosage compensation. Although any number of complex scenarios involving a direct relationship between HTZ-1 and the canonical dosage compensation process remain possible, we interpret the under-representation of sites of HTZ-1 localization on X to be a simple consequence of the under-representation of germline and developmentally important genes on the X chromosome (Discussion).
Using a combination of genetic mutation, RNAi, microscopy, and ChIP-chip, we have characterized the function and genomic distribution of the histone variant HTZ-1 in C. elegans. Our study examines several unresolved issues surrounding H2A.Z function during development, including its relationship to the process of dosage compensation and the function of H2A.Z at genes essential for embryogenesis. In addition, our study reveals unexpected properties of C. elegans genome organization and regulation.
The C. elegans genome has been shaped by the developmental programs it must coordinately execute. The general requirement of H2A.Z for development in metazoans suggests a function for H2A.Z in establishing or maintaining a specialized chromatin state at developmentally regulated promoters [27]–[30],[55]. In this study, we have established that HTZ-1 is incorporated upstream of genes critical for development, and that maternally provided HTZ-1 is sufficient for C. elegans embryogenesis. We infer by the progressively deteriorating phenotype suffered by offspring lacking HTZ-1 that HTZ-1 is required for both embryogenesis and post-embryonic development.
The function of HTZ-1 in pharyngeal organogenesis may provide a model for the mechanism by which HTZ-1 is generally required for C. elegans development. The development of the pharynx relies on precise temporal regulation of transcription activation, mediated in part by PHA-4, a FoxA transcription factor [56],[57]. HTZ-1 depletion enhances defects in pharyngeal organogenesis associated with loss of PHA-4, and activation of PHA-4-dependent promoters is delayed in the absence of HTZ-1 [23]. This is reminiscent of the delay of yeast GAL gene activation in the absence of Htz1 [10], and indicates a conserved role for H2A.Z in facilitating timely gene expression.
Previous genome-wide studies in yeast and other organisms have reached differing conclusions regarding the relationship between H2A.Z, RNA Polymerase II, and transcription [11], [12], [15], [16], [20]–[22],[42],[58],[59]. Functional divergence between yeast Htz1 and metazoan homologs are a possible source of the discrepancy. Consistent with this, C. elegans HTZ-1 is only 61% identical to yeast Htz1, but 77% identical to Drosophila H2Avd, and 83% identical to mouse or human H2A.Z (Figure S6).
In C. elegans, we found that HTZ-1 is incorporated specifically at promoters, where its occupancy is strongly correlated with RNA polymerase II occupancy, and to a lesser degree with transcript levels (see Text S1). This suggests that RNA polymerase II is present at some HTZ-1 occupied promoters without being linked to a corresponding increase in transcripts. One possible explanation is pausing of RNA polymerase II near initiation sites. This phenomenon is common in human and Drosophila cells [15], [60]–[62] but has not yet been established to occur in C. elegans. The 8WG16 RNA polymerase II antibody we used is probably not the appropriate choice for making conclusions about RNA Pol II pausing, because the antibody recognizes primarily the unphosphorylated RNA Pol II CTD, but it is known to have some cross-reactivity with both CTD-Ser5P and CTD-Ser2P. RNA Polymerase II pausing would be more appropriately examined with an independent, non-C-terminal domain RNA Pol II antibody or a CTD-Ser5P specific antibody. Nonetheless, using the 8WG16 antibody, we observed a small number of genes (about 300, or ∼1.5%) with promoter-restricted RNA Polymerase II.
A recent genome-wide study of the Drosophila H2A.Z homolog at mononucleosome resolution reported that an H2A.Z-containing nucleosome was often positioned just downstream of a paused RNA polymerase II [15]. Although we did not observe any relationship, positive or negative, between HTZ-1 occupancy and this putative paused state, peak HTZ-1 occupancy occurs about 80 bp downstream of peak RNA Pol II occupancy at promoters (Figure 5C). Thus, the putative poised state may in some cases be facilitated by HTZ-1, and could contribute to the efficient and timely activation of developmental promoters. Indeed, our data does not formally exclude the possibility that H2A.Z functions to dampen transcription [63]. In Drosophila and mammalian cells, H2A.Z plays a role in gene silencing by participating in the assembly of heterochromatin [64],[65]. While another study of C. elegans HTZ-1 argues against a repressive role [23], and we observe high levels of expression from many genes that contain HTZ-1 at their promoters, we cannot exclude the possibility that transcription at these loci would be even higher in the absence of HTZ-1.
One key question for future studies concerns how H2A.Z is directed to developmental promoters. Sequence-specific transcription factor binding at promoters is likely an important driver of Swr1-mediated H2A.Z incorporation [23],[66]. At the human p21 promoter, sites of p53 binding are occupied by H2A.Z and p400 (a human Swr1 homolog) and this enrichment is dependent on p53 binding [13]. In C. elegans, association of HTZ-1 with pharyngeal promoters is dependent upon the presence of promoter PHA-4 motifs [23]. This requirement of PHA-4 for HTZ-1 association may be one specific example of the general mechanism underlying the specificity of HTZ-1 for developmental promoters.
Studies in yeast implicate histone tail acetylation as another important factor. Histone H4K16 acetylation is a prerequisite for Htz1 association near yeast telomeres [16],[67]. Yeast Htz1 recruitment is reduced in the absence of Bdf1, a bromodomain containing protein that binds acetylated histone tails, and GCN5, a histone acetyltransferase that acetylates Histone H3 tails [11]. The NuA4 histone acetyltransferase complex, which interacts with Bdf1 and is responsible for bulk H4 acetylation and acetylation of Htz1 itself, shares multiple non-catalytic components with the Swr1 complex [11], [68]–[70]. Nucleosome free regions (NFRs) at promoters may also play a role. Htz1 was deposited at sites flanking NFRs, which often harbor 22-nt motif that contained a Reb1 transcription factor binding site [20]. Insertion of this motif at an ectopic location was sufficient for NFR formation and flanking Htz1 incorporation.
The incorporation of HTZ-1 at sites of transcription initiation suggests that HTZ-1 may be useful for identifying previously unannotated promoters. Our observation of HTZ-1 incorporation upstream of subsequent genes within operons implies the existence of independently regulated internal promoters in at least one-third of all currently annotated operons. Alternatively, operons may be less prevalent than the current genome annotation indicates. Indeed, a recent publication found evidence for functionally distinct internal promoters at 66 out of 238 (27%) downstream operon genes tested [71], a proportion of operons similar to which we found internal HTZ-1 incorporation. Additionally, transcript evidence from a published time-course during early development [31] provides evidence for independent regulation of some internal operon genes (Figure S2).
Immunofluorescence and ChIP-chip experiments reveal a significant under-incorporation of HTZ-1 on the X chromosome relative to the autosomes. We explored three explanations for this under-representation: differential detection of the HTZ-1 protein specifically on the X; a function for HTZ-1 in dosage compensation; or an under-representation of developmentally important genes, which tend to be HTZ-1 targets, on X [47]–[50].
The first possibility is reasonable because mammalian H2A.Z can be ubiquitylated on its C-terminus, and this mark distinguishes H2A.Z incorporated on the heterochromatin and silent X chromosome [19]. The C-terminal residues are conserved in C. elegans HTZ-1, and include the antigen to which the antibody was raised. However, analysis of HTZ-1::YFP localization, in which detection would not be affected by modification state, indicates that under-representation on X is not due to epitope occlusion by ubiquitylation or any other cause (Figure S4).
The second possibility concerns C. elegans dosage compensation, during which the two hermaphrodite X chromosomes undergo chromosome-wide reduction in expression to match the output of the single male X chromosome [54]. While we observed a high degree of overlap between sites of HTZ-1 incorporation and sites of DPY-27 binding, the converse did not hold true. Many sites of DCC binding occur in areas of no HTZ-1 incorporation, including known recruitment sites such as the rex-1 locus. This suggests that Dosage Compensation Complex binding does not require HTZ-1. We interpret the extensive co-localization on X to indicate independent functions of HTZ-1 and the DCC, both of which act upstream of genes active during embryogenesis. This interpretation is further supported by very few instances of overlap between HTZ-1 and DPY-27 at sites away from promoters. For example, only nine HTZ-1 peaks on X overlap with the 219 DPY-27 peaks found in the region downstream of genes. Of the nine overlaps, all may be explained as having promoter function: six occurred in regions near another gene promoter, and the remaining three were bound by RNA Polymerase II despite the absence of a gene annotation. Finally, HTZ-1 is under-incorporated on X both prior to and subsequent to the activation of somatic dosage compensation during embryogenesis.
The simplest explanation for the under-representation of HTZ-1 on X is that during embryogenesis HTZ-1 and the DCC both tend to bind at the transcription initiation sites of active genes important for development (Figure 7K). Genes essential for development are approximately 2-fold under-represented on the X [47],[48],[50], which is consistent with the approximately 2-fold under-incorporation of HTZ-1 on X. Although there are fewer sites of HTZ-1 incorporation on X, the individual sites of HTZ-1 incorporation on X do not differ in any quantifiable way from sites of incorporation on autosomes (Figure S8).
A ChIP-chip or ChIP-seq experiment revealing similar under-representation of HTZ-1 on the male X would provide further evidence against the direct involvement of HTZ-1 in dosage compensation. As stated in the results, HTZ-1 is likely to have an indirect function in dosage compensation and many other developmental processes by virtue of its incorporation at a wide spectrum of developmentally important genes. Nonetheless, our results do not completely rule out a direct positive or negative role for HTZ-1 in dosage compensation complex targeting or function. For example, the HTZ-1 on X could be post-translationally modified differently from HTZ-1 on autosomes [19],[72],[73], thereby conferring a distinct function for HTZ-1 depending on where in the genome it is incorporated.
One challenge in understanding H2A.Z function is integrating very diverse types of data, each of which lends clues to H2A.Z function but has its own limitations. For example, our experiments were performed in an unsynchronized population of embryos composed of multiple cell types. Therefore, the results presented here are a static projection of the dynamic activation and repression events that are occurring at gene promoters. How this might be manifested in our dataset can be illustrated by considering a previous time-course study of HTZ-1 at the myo-2 promoter [23]. HTZ-1 was not incorporated into the myo-2 promoter when it was repressed, but was transiently incorporated at the onset of transcription. HTZ-1 was then lost as myo-2 became highly expressed later in development. In a temporal projection of these results, as occurs in our dataset, it appears as if HTZ-1 was very weakly incorporated (if at all) into the myo-2 promoter (Text S1). Our genome-wide data indicates a preferential incorporation of HTZ-1 at developmentally dynamic promoters, and a loss of HTZ-1 at very highly transcribed genes. We infer that the general conclusions implied by the previous time-course study conducted at the myo-2 locus are now extended to the entire genome by our data [23].
The biophysical properties of nucleosomes containing H2A.Z provide clues about how H2A.Z could facilitate precise, coordinated developmental transcriptional programs. Incorporation of H2A.Z into an otherwise canonical nucleosome appears to have slight stabilizing effects, but incorporation of H2A.Z into nucleosomes containing H3.3, a mark of active transcription, is reported to cause instability [4],[5]. Furthermore, H2A.Z incorporation may alter associations between the histone proteins within the nucleosome, regulating the formation of higher-order chromatin fibers [18],[74]. The H3.3 mediated instability could allow H2A.Z to facilitate or maintain a nucleosome-free region at promoters upon activation, while the effect of H2A.Z on higher-order structures could promote the maintenance of transcription by promoting more precise nucleosome positioning at promoters or by promoting the assembly of a specialized higher-order chromatin state [4],[15],[18],[21],[42],[75]. In this way, HTZ-1 may aid the C. elegans embryonic genome in executing rapid transitions between quiescence and activity as developmental programs are executed.
The RNA polymerase II monoclonal antibody 8WG16 was obtained from Covance. A mouse ascites polyclonal anti-HTZ-1 antibody was made to the HTZ-1 specific C-terminal peptide (N-PGKPGAPGQGPQ-C) by Invitrogen-Zymed. Rabbit DPY-27 polyclonal antibodies used for immunofluorescence were generously provided by Dr. B.J. Meyer (UC Berkeley). Rabbit polyclonal MES-4 antibodies were generously provided by Dr. S. Strome (UC Santa Cruz). AlexaFluor donkey anti-mouse 488 IgG (Invitrogen A-21202), AlexaFluor donkey anti-rabbit 594 IgG (Invitrogen A-21207) were used as secondary antibodies for immunofluorescence. Anti-mouse HRP and ECL Plus (Amersham) were used for western blot visualization.
ChIP-chip analysis was performed in the N2 Bristol strain. htz-1(tm2469) was obtained from Shohei Mitani and balanced over nT1(qIs51) IV,V. KW1665 (htz-1(tm2469) IV/nT1(qIs51) IV,V) is maintained by selecting GFP-positive heterozygotes. All strains were cultured under standard conditions [76] at 20°C using E. coli strain OP50 or HB101 as a food source.
Adult hermaphrodites gravid with embryos were dissected in 1× PBS (137 mM NaCl, 2.7 mM KCl, 8 mM Na2HPO4 and 2 mM KH2PO4) on a slide. Paraformaldehyde was then added to 5%. The slide was incubated at room temperature for 2 minutes with a cover slip in place, and placed on dry ice for approximately 20 min. The cover slip was removed rapidly with a razor, and the slide was then placed into 95% ethanol for 2 minutes, followed by incubation in PBST (1× PBS+0.1% Tween-20) for 30 minutes. Slides were incubated with primary antibody at 1∶500 (HTZ-1) or 1∶100 (DPY-27 and MES-4) dilution overnight and with secondary antibody (1∶500) for approximately 3 hours. 4′,6-diamidino-2-phenylindole (DAPI) was used to stain DNA. Slides were mounted using ProLong® Gold antifade reagent (Invitrogen P36934). Staining was visualized using a Leica DMRXA microscope outfitted with a Cooke Sensicam. Capture and analysis of immunofluorescence images was performed using either Volume Scan (Vaytek) and Image-Pro Plus (Media Cybernetics) or SimplePCI.2 (Hamamatsu Corporation) imaging software. Nomarski DIC microscopy imaging was performed with a Leica DMRXA microscope and SimplePCI.2 software.
dsRNA was generated by in vitro transcription reaction using a Promega RiboMAX Large Scale RNA Production Systems T7 kit (Promega #P1300). Direct injection of concentrated dsRNA into adult gonads was required to obtain significant levels of embryonic lethality and larval arrest, as standard feeding and soaking methods did not result in sufficient depletion of the maternal HTZ-1. 1.2 µg/µL htz-1 dsRNA was injected into young adult eri-1(mg366) or N2 animals. The injected animals were allowed to recover and lay embryos overnight, then transferred to a new plate for collection and phenotypic scoring of affected embryos for 9–12 hrs. Following the 9–12 hour period, the adults and the embryos in utero were dissected and processed for immunofluorescence. The phenotypes of hatched larvae were observed and analyzed by DIC light microscopy 2–3 days after hatching. N2 animals were used for phenotype counts and staining experiments; eri-1(mg366) animals were used for DIC RNAi phenotype experiments.
Embryos were prepared by bleaching from gravid N2 adults grown in S-basal media liquid culture. Live embryos were cross-linked using 2% formaldehyde for 30 minutes at room temperature followed by quenching with 125 mM glycine for 15 minutes. Embryos were then washed twice with M9 Buffer, once by ChIP buffer, and frozen at −80°C. Extracts were prepared by resuspending embryo pellets in 1 volume ChIP Buffer (50 mM HEPES-KOH pH 7.5, 300 mM NaCl, 1 mM EDTA pH 8.0, 1% TritonX-100, 0.1% sodium deoxycholate, 10% glycerol, protease inhibitors (Calbiochem)), followed by dounce homogenization (50×) and sonication (4×, 1 s on, 0.5 s off, at 20% amplitude on ice) using a Branson Digital Sonifier 450. In a volume of 500 µL, 2 mg extract was used for each ChIP. 100 mg (5%) of the extract was set aside as “Input” and 400 µL elution buffer (0.1 M NaHCO3, 1% SDS) was added. Two (anti-RNA Pol II) or six (anti-HTZ-1) µg of antibody was added to each IP sample and incubated overnight at 4°C. Immune complexes were purified with 10 µL protein-A sepharose (Amersham) and washed 5 minutes with 1.5 mL of each of the following solutions: ChIP Buffer, ChIP Buffer with 500 mM NaCl, ChIP Buffer with 1 M NaCl (HTZ-1 IPs only), LiCl solution (10 mM Tris-HCl pH 8.0, 250 mM LiCl, 0.5% NP-40, 0.5% sodium deoxycholate, 1 mM EDTA), and TE (10 mM Tris-HCl pH 8.0, 1 mM EDTA). Samples were treated with 20 µg RNAse for 30 minutes at 37°C. IP samples were eluted twice with 200 µL elution buffer. NaCl was added to 200 mM and crosslinks were reversed by incubation overnight at 65°C. DNA was purified using Zymo DNA purification columns and amplified using LM-PCR [51].
Microarrays used were previously described (GEO GPL4614 and GPL4619; [51]). Four independent HTZ-1 ChIP biological replicates were performed, one of which was a dye-swap (ChIP 4). RNA Polymerase II ChIPs were performed from extracts used for HTZ-1 ChIPs 1 and 2. DPY-27 and “no antibody” datasets were published previously (GEO GSE6739; [51]). HTZ-1, RNA Polymerase II, and no antibody raw intensities were normalized by median centering log2 ratios (IP/input). Normalized log2 ratios from each experiment were converted to standardized z-scores and combined by taking the median of experiments. Raw data for HTZ-1 and RNA Polymerase II can be found at NCBI GEO accession number GSE10201. Peaks were derived using a Perl implementation of ChIPOTle (https://sourceforge.net/projects/chipotle-perl/) [35] using a window size of 500 bp, step size 86 bp, at a Bonferroni corrected p-value of 1×10−9. Any HTZ-1 peaks overlapping a peak found in the mock “No antibody” IP were removed from analysis. Peaks were annotated using Wormbase genome release 120 (Table S3). Maximum probe centers of peaks were annotated either to an intergenic or coding region, exclusively. Annotation distribution statistics were calculated using an unpublished C. elegans implementation of Cis-Element Annotation Software (CEAS) ([77], X. Shirley Liu, unpublished). Genome browser views were generated using the UCSC genome browser (http://genome.ucsc.edu), using the ce2 (March 2004)/ws120 genome build. Analysis for overrepresentation of Gene Ontology terms was done with GOTermfinder [43], accessed November 11, 2007 at http://go.princeton.edu/.
The figure of 23% of C. elegans genes being incorporated with HTZ-1 is based on annotating HTZ-1 peaks to the closest promoters and coding regions of 4650 genes (23.4%). In the case of bidirectional promoters (1800), both genes downstream were counted. Therefore, this number may be an overestimate if HTZ-1 functions only at one of the two genes in these cases.
HTZ-1 or RNA Polymerase II occupancy at each gene promoter was scored by averaging all probes within a 1-kb window centered on translation start sites. Transcript abundance data was obtained from a published study [31] and compared by averaging the last 3 timepoints from this study. To avoid ambiguity from maternally loaded RNAs, genes with high maternal transcript abundance (>100 parts per million) were removed from the clustering analysis. Each time point was divided by the 0 minute time point (the onset of zygotic transcription) and log2 transformed. K-means clustering (k = 6, 1000 iterations, similarity metric = spearman rank correlation) was performed using Cluster 3.0 [78] and visualized using Treeview [79].
|
10.1371/journal.pgen.1003070 | Insertion/Deletion Polymorphisms in the ΔNp63 Promoter Are a Risk Factor for Bladder Exstrophy Epispadias Complex | Bladder exstrophy epispadias complex (BEEC) is a severe congenital anomaly; however, the genetic and molecular mechanisms underlying the formation of BEEC remain unclear. TP63, a member of TP53 tumor suppressor gene family, is expressed in bladder urothelium and skin over the external genitalia during mammalian development. It plays a role in bladder development. We have previously shown that p63−/− mouse embryos developed a bladder exstrophy phenotype identical to human BEEC. We hypothesised that TP63 is involved in human BEEC pathogenesis. RNA was extracted from BEEC foreskin specimens and, as in mice, ΔNp63 was the predominant p63 isoform. ΔNp63 expression in the foreskin and bladder epithelium of BEEC patients was reduced. DNA was sequenced from 163 BEEC patients and 285 ethnicity-matched controls. No exon mutations were detected. Sequencing of the ΔNp63 promoter showed 7 single nucleotide polymorphisms and 4 insertion/deletion (indel) polymorphisms. Indel polymorphisms were associated with an increased risk of BEEC. Significantly the sites of indel polymorphisms differed between Caucasian and non-Caucasian populations. A 12-base-pair deletion was associated with an increased risk with only Caucasian patients (p = 0.0052 Odds Ratio (OR) = 18.33), whereas a 4-base-pair insertion was only associated with non-Caucasian patients (p = 0.0259 OR = 4.583). We found a consistent and statistically significant reduction in transcriptional efficiencies of the promoter sequences containing indel polymorphisms in luciferase assays. These findings suggest that indel polymorphisms of the ΔNp63 promoter lead to a reduction in p63 expression, which could lead to BEEC.
| Bladder exstrophy epispadias complex is a severe congenital abnormality. The affected babies' bladders are born open, leaking urine constantly. Treatment involves multiple major reconstructive surgeries and the need for lifelong care for the complications of the disease. Although a number of studies have suggested a genetic cause of the disease, the genetic and molecular mechanism underlying the formation of BEEC remains unknown. One gene, TP63, plays a crucial role in the early bladder development. Two different genetic promoters of TP63 produce different forms of the protein with opposing properties. We have shown mice lacking p63 displayed a deformity complex identical to human BEEC. There are no genetic mutations in the p63 protein in BEEC, so genetic variants in the promoter could alter protein expression. Our hypothesis was that loss of p63 expression due to sequence polymorphisms in a promoter is a risk factor for BEEC. We found promoter sequence variants that were statistically associated with the disease and the sequence variant location varied between Caucasian and non-Caucasian patients. This is particularly important as Caucasian populations have a higher risk of BEEC. These findings provide an explanation of BECC and a base for further study of TP63 related genes in this disease.
| Bladder exstrophy epispadias complex (BEEC; MIM600057) is a serious congenital anomaly present in 1 in 36,000 live births [1]. BEEC is manifested as a cluster of ventral midline defects including: 1) ventral bladder and abdominal wall defects, 2) epispadias (split external genitalia), 3) separation of the pubic bones and the rectus abdominis muscles, 4) exomphalos, and 5), ventrally displaced anus [2]. Treatment of BEEC requires a series of major reconstructive surgeries with high morbidity rate. Without treatment, the affected babies continuously leak urine through the bladder defects, resulting in skin excoriation, a progressive decline in renal function, a constant stench, and severe psychosocial strain for both patients and parents. Untreated BEEC patients develop chronic bladder mucosal irritation and have a 700-fold increased chance of developing bladder cancer. The genetic and molecular mechanisms underlying the formation of BEEC remain unclear. To date, studies have shown an increased incidence of BEEC in children of older mothers [3], in Caucasian populations [4], and a familial genetic component of pathogenesis [5]. Chromosomal abnormalities have been suggested as possible cause of BEEC and a genome wide linkage study suggested more than one gene was involved [6]. Among siblings and offspring of BEEC patients, the risk of BEEC increase dramatically from 1 in 10,000–50,000 to 1 in 100 and 1 in 70 respectively, representing a 500-fold increase in incidence [7]. The concordance rate among the monozygotic twins is much higher than that of dizygotic twins (62% vs. 11%), representing a 4500-fold increase in incidence compared to that of the general population [8]. Brought together these data clearly suggest a genetic component in BEEC pathogenesis.
A member of the p53 tumor suppressor family, the protein p63, is expressed in all stratified epithelia, including the bladder urothelium and the skin overlying the external genitalia during development [9]. The protein p63 plays a key role in initiating epithelial stratification during development [10]. Expression of TP63 is regulated by two promoters, TAp63 and ΔNp63, located upstream to exon 1 and exon 3 respectively [11]. TAp63 protein isotypes are pro-apoptotic whereas ΔNp63 isotypes are anti-apoptotic and both isotypes may compete for the same set of target genes [11]. Mouse embryos lacking p63 have thin, non-stratified skin, a short tail, truncated limbs, and cleft palate and are perinatally fatal [11], [12]. p63 is required for stratification of epithelia, including the bladder urothelium [10]; furthermore the epithelial-mesenchymal interaction is instrumental in bladder mesenchymal (smooth muscle) development [13]. We have demonstrated that p63−/− mouse embryos exhibit ventral midline defects identical to those of human BEEC, including ventral bladder and abdominal wall defects, separation of external genitalia, separation of pubic bones and rectus abdominis muscles, exomphalos, and ventrally translocated anus [9]. p63 is expressed in the bladder urothelium during development, TAp63 at an earlier stage (E9–11) and ΔNp63 later (E11–19). ΔNp63 is preferentially expressed along the ventral midline in the epithelium overlying the genital tubercle and ventral bladder. Moreover, the apoptotic activity of ventral urothelium of p63−/− bladder is markedly increased whereas cell proliferation is much reduced. We believe that bladder epithelial apoptosis and failure of induction of the adjacent mesenchyme lead to the development of a BEEC-like phenotype in mice [9]. This study investigates whether the loss of p63 expression due to genetic variants in the ΔNp63 is a risk factor for BEEC. In this study we show the expression of the two different TP63 promoters in BEEC patient tissue, and explore sequence of the ΔNp63 promoter in BEEC patients and controls. Our studies suggest that ΔNp63 is the dominant promoter in human tissue and its expression is significantly reduced in BEEC tissue in early bladder formation. We also find many sequence variants within the ΔNp63 promoter and three insertion/deletion polymorphisms were significantly associated with increased risk of BEEC. Furthermore the sites of these polymorphisms varied between Caucasian and non-Caucasian patients.
To establish if TP63 plays a role in human bladder exstrophy, we first confirmed by real time-PCR that normal human foreskin (from circumcision) predominantly expresses ΔNp63 mRNA, whereas the TAp63 isoform was expressed at lower levels (Figure 1A). Compared with normal controls, the ΔNp63 expression in the dorsal foreskin (adjacent to the epispadias) was decreased and also decreased compared to the patient's own ventral foreskin (opposite side) (Figure 1B). Down-regulation of ΔNp63 appears to be mainly in the foreskin mucosa (non-keratinized epithelium) in an 8-year old BEEC patient (Figure 1C). Strikingly, in a tissue sample taken from the ventral bladder of a 2 day old Caucasian BEEC patient, ΔNp63 expression was decreased whereas TAp63 was increased compared with normal controls (Figure 1D and 1E). Tissue taken from the bladder of a<1 year old Caucasian BEEC patient also had deceased ΔNp63 and increased TAp63 compared with normal controls (Figure 1F). Expression of ΔNp63 and TAp63 in older Caucasian BEEC patients showed more varied expression (Figure 1G). The older patients haven undergone Mitroffanof procedures (using appendix as a conduit for bladder catheterization) and bladder augmentation (using small bowel patch to increase bladder volume). The irritation from small bowel mucus secretion and various degree of cystitis (from bacteria introduced by catheters) may affect TP63 expression in the urothelia sampled. The genotypes of patient samples from Figure 1D–1G are shown in Table 1. We found that ΔNp63 expression in some of the BEEC patients' urothelia is reduced. Although the post-natal expression does not necessarily represent that of the developing embryos and the sample size of post-natal bladder urothelia will not be large enough to draw statistically convincing conclusion, our data does demonstrate that TP63 is expressed in neonatal bladder urothelium and neonatal foreskin of normal individuals and BEEC patients. The possible reduced ΔNp63 expression during development could be one of the possible mechanisms of BEEC pathogenesis.
Our data corroborate a recently published study where three out of the five BEEC patients had no ΔNp63 expression detected in their bladders [14]. Our results showing reduced levels of ΔNp63 expression led us to examine if any mutations were present within the coding sequence of the TP63 gene in BEEC patients. We therefore sequenced all 15 exons of TP63 gene [11] in 15 BEEC patients but found no mutations (data not shown). This finding confirms that of a recent study where no exon mutations were found in a study of 22 BEEC patients [14].
To explain reduced ΔNp63 expression in the absence of any exon mutation, the ΔNp63 promoter (2700 nucleotides upstream of exon 3) was sequenced in BEEC patients and normal controls. DNA was extracted from buccal swab samples from 163 BEEC patients and 285 ethnicity-matched controls from India, Bangladesh, China, Australia, Spain, Canada and USA. We found 7 single nucleotide polymorphisms (SNPs; 2 of which were novel ss#541026548 and ss#541027120) and 4 insertion/deletion (indel; 1 novel ss#541028600) polymorphisms in the ΔNp63 promoter in both BEEC and control sequences (Table 2). There were no significant deviations from Hardy-Weinberg equilibrium in the controls' genotype distributions when tested by a goodness of fit χ-square test [15]. In a number of cases we were unable to obtain complete sequences from patient or control samples.
Significantly, three indel polymorphisms (rs6148242, rs5855273, ss#541028600) were associated with increased risk of BEEC (Table 3). The indel polymorphisms showed no linkage disequilibrium (12 base pair (bp) deletion (del) vs. 4 bp insertion (ins), r2 = 0.0754; 1 bp vs 4 bp, r2 = 0.145; 12 bp vs 1 bp, r2 = 0.0273) [16]. We stratified the cohort into Caucasian and non-Caucasian groups based on the increased incidence in Caucasian populations [4]. The non-Caucasian group contained mainly Indian and Chinese patients and there were no significant differences in the three indel polymorphisms between the Indian and Chinese patients. A significant ethnicity bias of ΔNp63 promoter base pair (bp) indel polymorphisms was observed in patients compared to their ethnically matched controls. The 12 bp deletion (rs6148242) was only associated with increased risk of BEEC among Caucasians but not in non-Caucasian patients. Conversely, the novel homozygous 4 bp insertion (ss#541028600) was found to be only associated increased risk of BEEC amongst non-Caucasians. The homozygous 1 bp insertion polymorphism (rs5855273) was significant across both cohorts (Table 4; Table 5). Interestingly, two heterozygous SNPs (ss#541026548, rs1464117) were associated with decreased BEEC risk (p = 0.0043 Odds Ratio (OR) = 3.060 and p = 0.0379 OR = 2.164 respectively) (Table 2; Figure 2).
To assess the effect of different indel polymorphisms on transcriptional efficiency, we performed luciferase assays to test the ΔNp63 promoters. We sub-cloned four variations of indel polymorphisms (Table 6) into the pGL3 luciferase vector and transfected human embryonic kidney (HEK-293) cells. We found a consistent and statistically significant reduction of the transcriptional efficiencies of the promoters containing indel polymorphisms compared with the control sequence lacking indel polymorphisms (Figure 3; Table 7).
A number of studies suggest a genetic component in the etiology of BEEC. Precise regulation of ΔNp63 expression is required for development and differentiation of the ventral bladder urothelium during human development. In addition, p63−/− mice also have phenotypes identical to the associated anomalies of BEEC patients (Table 8). The only genetic model of the BEEC provides valuable insight into the possible mechanism of BEEC in humans. The mouse model showed that the anti-apoptotic isotype of p63, ΔNp63, is predominantly expressed in fetal murine bladder epithelium. Loss of ΔNp63 expression in p63−/− mice increased apoptosis in the ventral bladder epithelium causing reduced mesenchymal induction (indicated by reduced expression of the mesenchymal markers Msx-1 and Fgf8) [9]. The reduction in mesenchymal cell proliferation and failure of smooth muscle formation in turn resulted in ventral midline defects, i.e. BEEC [9]. Brought together this evidence strongly suggests that TP63 is a candidate gene for human BEEC and dysregulation of p63 expression could be a contributing factor in BEEC pathogenesis.
Our murine model shows that the urogenital tubercle, the embryological origin of the foreskin, is one of few anatomical sites where ΔNp63 is expressed [9]. Foreskin in our human study is one of p63 expressing tissues and is logistically more accessible. Our study has shown that ΔNp63 is the dominant isotype in human foreskin, that ΔNp63 expression was reduced in the urothelium of BEEC patients, and expression of the pro-apoptotic TAp63 was increased. No exon mutations have been discovered in TP63 suggesting that dysregulation may be caused by other regions of the gene such as the promoter regions. Sequence variants in promoters have been shown to be risk factors in numerous diseases including pneumoconiosis [17], auto-immune diseases [18], asthma [19], and β-thalassaemia [20]. A six-nucleotide promoter polymorphism has been described as a risk factor in multiple cancers such as lung, esophagus, stomach, colorectum, breast and cervix in Chinese populations [21]. However, studies on breast, prostate, and colorectal cancer in European and USA populations have not shown the same association with cancer risk [22]–[24]. Sequencing of the ΔNp63 promoter region revealed three indel polymorphisms in the 163 BEEC patients associated with a statistically significant increase in BEEC risk. The prevalence and the role of indel polymorphisms differed between Caucasian and non-Caucasian ethnic populations. While further studies are required to explain this, we speculate ethnicity-specific polymorphisms of up-stream or down-stream genes may further modify the final effect of the ΔNp63 promoter polymorphisms. One possible explanation may be that the polymorphisms may interfere with transcription binding sites which may differ in Caucasian and non-Caucasian populations. An example is the deletion of six-nucleotides in the CASP8 promoter destroys the Sp1 transcription factor binding site [21]. We searched transcription factor binding sites in the ΔNp63 promoter using MATCH software. Computational analysis predicted the 12 bp indel might affect binding of Hand1, GATA-1,2,3,6, Gfi1, LEF1, TCF1 and SOX10, whereas the 4 bp indel may affect binding of SREBP, EGR and CBF transcription factors. Further studies would be needed to verify these suggested interactions.
In this study, indel polymorphisms in ΔNp63 promoter are associated with increased risk of BEEC, most likely due to decreased transcriptional efficiency and therefore decreased expression of anti-apoptotic ΔNp63 isoforms during bladder development. The consequences of decreased expression of p63 isotypes could be complex and result in stimulation or negative regulation of a number of genes. One such gene, PERP, is a p63 regulated gene central to epithelial integrity and homeostasis [25]. In skin deficient in PERP, desmosomal deficits are observed in addition to epithelial blistering [25]. Genome-wide expression profiling has revealed a great number of desmosomal-linked genes in addition to PERP, SYNOP2, and the Wnt pathway as potentially contributing to BEEC etiology [26]. P63 mutation has been implicated in human disease such as Ectodactyly, Ectodermal dysplasia and Cleft palate/lip syndrome (EEC) [27]. Maas et al., reported that, out of 14 members of a family with EEC syndrome, 10 suffered from micturition problem [28]. After reviewing 24 previous reports of urogenital anomalies in EEC patients, Maas concluded that structural anomalies of urogenital system may be part of EEC syndrome. The report included a histological figure of “atrophic” urothelium, which showed area of thin urothelium, reminiscent of the non-stratified bladder epithelium of p63−/− BEEC knockout model [28]. The association is further supported by a report from Chuangsuiwanich et al., which showed a case of EEC foetus with markedly hypoplastic bladder, lined at it lower part with thin urothelium [29]. Sub-clinical variations of BEEC may be more prevalent in EEC patients and/or other conditions than previously thought.
We conclude that insertion/deletion polymorphisms of ΔNp63 promoter are associated with increased risk of Bladder Exstrophy Epispadias Complex. We believe our results provide a base for further study of p63 related genes in this debilitating condition
The Research Ethics Committee of Royal Children's Hospital, Melbourne, Australia reviewed and approved the study. Informed consents were obtained from the patients, parents/guardians and the normal controls. Buccal swabs were used to collect DNA extracted with the BuccalAmp DNA Extraction Kit (Epicentre Biotechnologies, Madison, WI, USA). All samples were codified and subject identities kept confidential. Collaborators from overseas centres (Canada, USA, Spain, India, Bangladesh, China, and Malaysia) had the study approved by their institutional ethic committees prior to sample collection.
Exon PCR amplification was performed using published protocols [27] and in-house primers designed using Primer 3 and Premier Netprimer (Premier BioSoft International). Four pairs of primers were designed to complete the PCR and sequencing of the 2785 bp promoter in four parts. The extent of the ΔNp63 promoter by aligning and comparing the 5′ upstream regions of exon 3 of P63 gene sequences from humans, mouse and pig. The most conserved region 2785 bp (−2696 to +89) was selected as the putative region of ΔNp63. Sequencing was carried out in both directions however in a number of cases sequences from patients or controls were unreadable using a number of different combinations of primers. The PCR products were purified and then used as templates for direct DNA sequencing by the automated ABI Prism 3100 Genetic Analyzer and the Big Dye Terminator kit (Applied Biosystems, Mulgrave, VIC, Australia). Sequences were compared with reference sequences of p63 GenBank (Locus: NC_000003) using the BioEdit software (Ibis Biosciences). Data were analyzed with contingency tables, χ-square, odds ratio and 95% confidence intervals (GraphPad Prism, GraphPad Software, Inc.).
As previously described [9], RNA was extracted from foreskin and bladder urothelial tissue samples from BEEC patients. Circumcision foreskin and normal bladder urothelium from cystectomy (cancer) specimens were used as controls. Tissue samples were snap frozen in liquid nitrogen, ground to powder, mixed with 10 µl of β-mercaptoethanol (Sigma-Aldrich Pty. Ltd, Castle Hill, NSW, Australia) in buffer and centrifuged. Samples were purified with QIAshredder (Qiagen Pty. Ltd., Doncaster, VIC, Australia). RNA was then extracted with RNeasy (Qiagen Pty. Ltd., Doncaster, VIC, Australia) or Trizol (Invitrogen, Mulgrave, VIC, Australia). cDNA synthesis was accomplished using Oligo dT (Invitrogen, Mulgrave, VIC, Australia), dNTPs (Invitrogen, Mulgrave, VIC, Australia), Superscript II First-Strand Synthesis Kit (Invitrogen, Mulgrave, VIC, Australia), and RNAse inhibitor (Invitrogen, Mulgrave, VIC, Australia). Real-time qPCR was carried out using SYBR Green (Applied Biosystems, Mulgrave, VIC, Australia) on a MJ Research Bio-Rad Chromo 4 cycler (Gladesville, NSW, Australia), according to published TAp63 and ΔNp63 primer sequences and published PCR conditions [30]. Relative expression will be analyzed by the Pfaffl methodology with β-actin or GAPDH as the endogenous control [31].
ΔNp63 promoter sequences with indel polymorphisms were cloned into pGL3 luciferase reporter vector (KpnI/XhoI sites Promega, Madison, WI, USA). Human embryonic kidney (HEK-293) (Sigma-Aldrich Pty. Ltd, Castle Hill, NSW, Australia) was used for luciferase assays. Cells were cultured with DMEM (Invitrogen, Mulgrave, VIC, Australia), 10% fetal bovine serum at 37°C with 5% CO2. Cells were plated on a 96-well plate (BD Biosciences, North Ryde, NSW, Australia). The following day cells were co-transfected with 200 ng of pGL3-promoter constructs containing ΔNp63 promoters and TK Renilla plasmid DNA with 5 µl Lipofectamine Transfection Reagent (Invitrogen, Mulgrave, VIC, Australia). Cells were harvested 48 hours later and cell lysates assayed for luciferase activity with the Dual Luciferase Reporter Assay system (Promega, Madison, WI, USA). Luminescent signals from the firefly and Renilla luciferase reactions were measured in a FLUOstar Optima luminometer (BMG Labtech Pty. Ltd., Mornington, VIC, Australia). Luciferase assay reagent was added, firefly luciferase luminescence measured, followed by Stop & Glo Reagent (Promega, Madison, WI, USA). The signal intensity from Renilla luciferase was used to normalize the signal from the firefly luciferase. One-way ANOVA with post-hoc Bonferroni's multiple comparisons test was applied to compare the relative expressions (GraphPad Prism, GraphPad Software, Inc.).
The following submitter National Center for Biotechnology Information (NCBI) SNP (ss) accession numbers were assigned to the SNPs and indel observed in this study: promoter position −2657, ss#541026548; −2651, ss#541027004; −2431, ss#541027120; −2293-2282, ss#5411027737; −2260, ss#541027867; −1944, ss#541027977; −1287, ss#541028071; −1209, ss#541028198; −1059, ss#541028317; −71, ss#541028452 (2 bp); and −71, ss#5410228600 (4 bp).
|
10.1371/journal.pgen.0030177 | Genome-Wide Screen for Modifiers of Ataxin-3 Neurodegeneration in Drosophila | Spinocerebellar ataxia type-3 (SCA3) is among the most common dominantly inherited ataxias, and is one of nine devastating human neurodegenerative diseases caused by the expansion of a CAG repeat encoding glutamine within the gene. The polyglutamine domain confers toxicity on the protein Ataxin-3 leading to neuronal dysfunction and loss. Although modifiers of polyglutamine toxicity have been identified, little is known concerning how the modifiers function mechanistically to affect toxicity. To reveal insight into spinocerebellar ataxia type-3, we performed a genetic screen in Drosophila with pathogenic Ataxin-3-induced neurodegeneration and identified 25 modifiers defining 18 genes. Despite a variety of predicted molecular activities, biological analysis indicated that the modifiers affected protein misfolding. Detailed mechanistic studies revealed that some modifiers affected protein accumulation in a manner dependent on the proteasome, whereas others affected autophagy. Select modifiers of Ataxin-3 also affected tau, revealing common pathways between degeneration due to distinct human neurotoxic proteins. These findings provide new insight into molecular pathways of polyQ toxicity, defining novel targets for promoting neuronal survival in human neurodegenerative disease.
| Spinocerebellar ataxia type-3 is the most common dominantly inherited movement disorder and is caused by a CAG repeat expansion within the gene ATXN3, encoding the Ataxin-3 protein. This leads to a protein with an expanded polyglutamine domain, which confers a dominant toxicity on the protein, leading to late onset, progressive neural degeneration in the brain. Although some modifiers of Ataxin-3 toxicity have been defined, little was known about their molecular mechanisms of action. The fruit fly Drosophila recapitulates fundamental aspects of the human disease. Here, we performed a genome-wide screen for new modifiers of Ataxin-3 toxicity using the fly and defined 25 modifiers in 18 genes. The majority of the genes belong to chaperone and ubiquitin proteasome pathways, which modulate protein folding and degradation, but the remaining modifiers have a broad range of predicted molecular functions. Assays in vivo revealed that the biological activity of all modifiers converge on aiding in situations of protein misfolding, despite distinct predicted molecular functions. Select modifiers of Ataxin-3 toxicity also modulated tau toxicity associated with Alzheimer disease. These findings underscore the importance of protein homeostasis pathways to disease and provide the foundation for new therapeutic insight.
| Spinocerebellar ataxia type 3 (SCA3) is the most common dominantly inherited ataxia worldwide and is caused by a CAG repeat expansion encoding glutamine within the ATXN3 gene [1,2]. The expanded polyglutamine (polyQ) is thought to confer toxicity on the protein Ataxin-3, leading to neural dysfunction and loss [3]. At least nine human diseases, including additional spinocerebellar ataxias and Huntington's disease, share this mechanism. Studies on such pathogenic proteins reveal that the long polyQ domain alters protein conformation, causing an enriched beta sheet structure [4]. Mutant polyQ protein also dynamically associates with chaperones and colocalizes with proteasome subunits, indicating that the protein is misfolded or abnormally folded [5,6]. Such accumulation of misfolded protein is a common pathology of human degenerative disorders, including Alzheimer, Parkinson, and prion diseases [7–9], indicating that these diseases may share molecular and cellular mechanisms.
Models for human neurodegenerative diseases in simple systems have provided valuable tools to dissect molecular mechanisms of disease pathology [10–13]. Directed expression of pathogenic human Ataxin-3 in Drosophila recapitulates key features of disease, with late-onset neuronal dysfunction and degeneration accompanied by ubiquitinated inclusions [14,15]. Neurotoxicity is more severe with increasing length of the polyQ repeat, similar to the human disease where longer repeats are associated with more severe and earlier onset disease [16,17]. In the fly, increased levels of expression of the disease protein leads to more severe degeneration and earlier onset protein accumulation, suggesting that abnormal accumulation of the mutant protein is central to disease and degeneration.
A number of modifiers of select polyQ disease proteins have been identified using animal models, including chaperones, transcriptional coregulators, and microRNAs [18–22]. Although these approaches have revealed genes that modulate polyQ toxicity, little is known regarding how the modifiers act biologically to modulate polyQ degeneration. Among polyQ proteins, Ataxin-3 is unique because it has been implicated in ubiquitin pathways, and its normal activity may impinge on protein degradation pathways [23–26]. Truncation of Ataxin-3 to remove the ubiquitin protease domain, or mutation of the ubiquitin protease activity, dramatically enhances toxicity [15], indicating that the normal activity of Ataxin-3 may be critical in Ataxin-3-induced degeneration. To reveal insight into pathways that modulate Ataxin-3 neurodegeneration, we performed a genetic modifier screen in Drosophila. These studies revealed a range of modifiers that, despite some broadly diverse predicted molecular functions, converge on protein misfolding with a subset mitigating toxicity through proteasome and/or autophagy pathways. These findings underscore the critical role of protein quality control in SCA3 pathogenesis and provide potential new targets toward therapeutics.
To define modifiers that may reveal new insight into mechanisms of human SCA3 disease, we performed an overexpression screen for modifiers of Ataxin-3-induced neurodegeneration in Drosophila. SCA3trQ78 causes late onset progressive degeneration characterized by loss of pigmentation and collapse of the eye (Figure 1A and 1B) [14]. We initially screened a subset of 2,300 available EP-element insertion lines, each carrying a transposon engineered to direct expression of the downstream gene in the presence of the yeast GAL4 protein [27]. Because reproducibility was variable with this collection, we then performed a screen de novo, selecting for new EP-insertions that modified SCA3trQ78 toxicity. This approach identified 17 suppressors and one enhancer (Figures 1C–1J and S1), which affected both external and internal retinal degeneration.
Plasmid rescue was performed to identify the genes affected. BLAST searches with genomic sequence from the integration sites revealed that the lines bore insertions in the 5′ regulatory region of select genes, and all were in an orientation to direct GAL4-dependent gene expression (Text S1; Table S1). Northern and reverse transcription PCR analysis confirmed upregulation of the targeted genes; using a variety of tests we confirmed that the modifiers did not appear to affect transcription of the Ataxin-3 transgene or general GAL4-UAS transgene expression (Figure S2). Both molecular and genetic analyses confirmed that the insertions were single insertions (see Materials and Methods). Reversion analysis proved that the EP-elements were causal in modification. Taken together, these data indicated that the EP modifiers resulted in increased expression of the targeted downstream genes, which modulated SCA3trQ78 toxicity and neurodegeneration.
Analysis of the targeted genes revealed that the majority fell into two major classes of chaperones and ubiquitin-pathway components (Table 1; Figures S1 and S3). The remaining modifiers were placed in a third category of miscellaneous functions. Class 1 (Figure 1C–1E) included an Hsp70 family member, Hsp68E407; two Hsp40 genes, DnaJ-1B345.2 and mrjE1050; a small heat shock protein αB crystalline CG14207EP1348; and the cochaperone Tpr2EB7-1A. The class 2 ubiquitin-pathway components included: polyubiquitin CR11700EP1384; a ubiquitin-specific protease Ubp64EE213-1A; two ubiquitin ligases, CG8209B3-Sa and Faf E659; and an F-box protein CG11033EP3093, the only enhancer of the group. Class 3 (Figure 1F and 1G) included genes with a variety of predicted molecular functions: the nuclear export protein Embargoed, embE2-1A and embE128-1A; three transcriptional regulators Sin3AB9-E, NFAT (NFATEP1335 and NFATEP1508), and debra (dbrEP456 and dbrEP9); three translational regulators, including four alleles of the lin-41 homologue dappled (dpldJM120, dpldJM265, dpldEP546, and dpldEP2594), a polyA binding protein orb2B8-S, and insulin growth factor II mRNA binding protein ImpEP1433; and finally a fatty acid oxidation enzyme palmitoyl co-A oxidase CG5009B227.2. This indicated that, although chaperone and ubiquitin-pathway components are major modifier categories, a variety of functional pathways are implicated in Ataxin-3 pathogenesis.
The screen selected for modifiers that, upon upregulation, affected toxicity. To determine whether the activity of these genes may normally play a role in SCA3 toxicity, we examined whether reduction in the level of the modifier genes had an effect. To do this, we reduced gene expression by 50% using loss-of-function alleles where available or deficiency lines. Among these, reduction with a deficiency of DnaJ-1 and Ubp64E (within the same deficiency; reduction of DnaJ-1 has previously been shown to enhance with a dominant-negative construct [16]), Trp2, emb (with an allele), and dbr dramatically enhanced degeneration (Figure 1K–1O; Table S2). Although the deficiencies reduce the level of a number of genes, these data suggest that the endogenous activity of these genes may normally help to protect against degeneration.
To address whether morphological rescue correlated with functional rescue, we determined the ability of select suppressors to restore function in a phototaxis assay. When flies bearing the SCA3trQ78 protein were given a choice between a light and dark chamber, they distributed randomly, indicating that they are functionally blind (Figure 1P). However, when mrjE1050, which dramatically rescued degeneration, was co-expressed, normal vision was restored. A milder suppressor that anatomically restored less retinal tissue, CG14207EP1348, restored vision partially. Thus, anatomical rescue correlated with functional rescue. These and other studies indicated that the modifiers not only modulated toxicity of the external eye, but also that of the neuronal cells. To confirm this in another situation, we tested select modifiers for the ability to mitigate polyQ toxicity when directed exclusively to the nervous system with elav-GAL4 (Figure S4). These studies confirmed that the modifiers mitigated neuronal toxicity of the Ataxin-3 protein.
We performed select additional experiments with dpld, for which we obtained many independent EP overexpression alleles. These detailed studies confirmed activity of the EP alleles of dpld with independent UAS-dpld transgenic lines. Further, suppression by dpld was not limited to development but also extended to the adult timeframe (Figures S2, S5, and S6).
Because the genes were isolated as modifiers of a truncated Ataxin-3 protein, we tested whether they could mitigate toxicity of full-length Ataxin-3, which is largely a neuronal toxicity [15]. Modifiers that phenotypically strongly suppressed toxicity of truncated Ataxin-3 also strongly suppressed full-length Ataxin-3 toxicity; however, a number of modifiers that were weak or moderate suppressors of the truncated protein were, in contrast, strong suppressors of full-length Ataxin-3: CG14207EP1348 (αB crystalline), CR11700EP1384 (polyubiquitin), CG8209B3-Sa (putative ubiquitin ligase), and Sin3AB9-E (Figure S7 and unpublished data). Consistent with strong anatomical rescue, CG14207EP1348 also robustly suppressed functional vision defects of full-length Ataxin-3 (Figure 1Q). The enhancer, however, CG11033EP3093 had a minimal effect on toxicity of full-length Ataxin-3 (unpublished data). This indicated that the modifiers varied in strength and selectivity depending upon whether the Ataxin-3 protein was intact or truncated. As truncation may be a feature of SCA3 disease [28–30], the effectiveness of modifiers against different forms of Ataxin-3 has implications for disease pathogenesis.
Previous studies have shown that the molecular chaperone Hsp70 is a potent suppressor of SCA3 toxicity [31]. Therefore, we considered that the class 1 modifiers may function similar to Hsp70 to help cells handle the misfolded disease protein, whereas those of class 2 likely have a role in ubiquitin-dependent pathways that process misfolded proteins. However, class 3 presented a range of potential activities. To address how the modifiers were functioning biologically, we tested whether the modifiers could affect a general protein misfolding phenotype: compromised chaperone activity with a dominant-negative form of Hsp70 (Hsp70.K71E). This situation results in an eye phenotype that resembles severe polyQ degeneration (Figure 2A and [32]). Thus, modifiers that affected both SCA3 and Hsp70.K71E toxicity would likely include those whose mode of action was to modulate protein misfolding.
Strikingly, we found that most of the suppressors of polyQ toxicity also mitigated the Hsp70.K71E phenotype, as well as or better than directed expression of Hsp70 itself (Figure 2; Table 1). Interestingly, DnaJ-1B345.2 and Tpr2EB7-1A enhanced this phenotype; we interpreted this to indicate that these genes, which encode proteins thought to act as cochaperones of Hsp70, may compromise residual Hsp70 in the dominant-negative situation. The chaperone mrjE1050, although an Hsp40 class chaperone, acted in a manner distinct from DnaJ-1B345.2, as it suppressed rather than enhanced the Hsp70.K17E phenotype. Moreover, the enhancer of SCA3trQ78 toxicity, CG11033EP3093, suppressed the misfolding phenotype. Only one modifier did not affect this phenotype, the ubiquitin protease Ubp64EE213-1A; interestingly, normal Ataxin-3 also has ubiquitin protease activity that mitigates its own pathogenicity, and similarly, has no effect on Hsp70.K71E [15]. We considered that one mechanism by which the modifiers may mitigate the Hsp70.K71E phenotype is by upregulating Hsp70/Hsc70 chaperones; however, none of the modifiers appeared to act in this way (Figure 2K and unpublished data). These results indicated that the majority of modifiers of SCA3trQ78 toxicity appeared to function biologically by aiding in situations of compromised chaperone activity and/or protein misfolding.
The degree of neurodegeneration induced by pathogenic polyQ protein is typically correlated with the level of accumulation of the protein in animals in vivo. We reasoned, therefore, that the modifiers may affect protein levels. We therefore examined protein accumulation by immunohistochemistry and western analysis. Although nuclear inclusions may not be causal in disease [33], later onset, smaller nuclear inclusions are typically reflective of reduced pathogenicity of the protein in vivo. Hsp70 and Hsp40 have also been shown to increase the solubility of pathogenic polyQ protein by western blots, concomitant with reducing toxicity [16]. We therefore examined protein accumulation using rh1-GAL4 or the full-length Ataxin-3 protein—both situations that allow sensitive analysis of protein accumulation [15,22]. In these studies, we limited analysis to the strong and moderate modifiers.
Immunohistochemical analysis revealed that select modifiers had striking effects to reduce NIs. These included the class I chaperones Hsp68E407, DnaJ-1B345.2, mrjE1050, but did not include αB-crystalline CG14207EP1348 or Tpr2EB7-1A (Figure 3; Table 1). Of the class 2 ubiquitin-pathway components, polyubiquitin CR11700EP1384 and the ubiquitin protease Ubp64EE213-1A reduced NIs, but the other strong modifiers of this class did not. Of class 3 modifiers tested, all reduced NI except ImpEP1433 (Figure 3B and 3F). We then analyzed solubility of the pathogenic protein by immunoblot. This approach revealed that, although the modifiers had no effect on protein levels at early time points prior to inclusion formation, all suppressors increased the level of monomeric protein over time, thus all increased the solubility properties of the pathogenic protein (Figure 3E). The effect was specific, as co-expression of a control protein (green fluorescent protein [GFP]) had no effect (unpublished data). Similarly, the enhancer reduced monomer levels (unpublished data). These findings indicate that the modifiers either affected pathogenic protein accumulation or altered the solubility properties of the toxic protein, concomitant with altering protein pathogenicity.
Given that many modifiers mitigated protein accumulation, we asked whether there were interactions with genes of protein degradation pathways. A key pathway thought to modulate the pathogenic polyQ toxicity is the ubiquitin-proteasomal system (UPS) [34]. We therefore tested whether suppression by modifiers that lowered protein levels was dependent upon a fully functional proteasome. Proteasome activity can be reduced by a dominant temperature-sensitive mutation in a proteasome protein subunit (DTS5) [35]. In a situation where limiting proteasome activity using this mutation had no effect on SCA3 toxicity on its own, we found that a striking number of modifiers still suppressed polyQ toxicity, thus indicating that they were not sensitive to inhibition of the proteasome by this assay; these included dpld alleles (Figure 4; Table 1). In contrast, a striking exception was the class 2 ubiquitin-pathway suppressors: all of these modifiers lost the ability to suppress upon proteasome inhibition with the DTS mutation.
Autophagy, or lysosome-mediated protein degradation, has also been implicated in polyQ toxicity and cell survival in situations of stress [36,37]. We therefore asked whether the modifier genes had an effect on this process. First, we determined whether normal or pathogenic Ataxin-3 itself induced lysosomal accumulation reflective of autophagy, by examining the fat body tissue from larvae, a standard assay for autophagy [38]. Normally, well-fed animals show minimal lysosomal induction, whereas starved animals show a dramatic increase, reflected by the uptake of dye (Figure 4F and 4G). Whereas expression of normal Ataxin-3 (SCA3Q27) had minimal effect, expression of pathogenic Ataxin-3-induced autophagy (Figure 4J). To further address the role of autophagy, we determined whether limiting the activity of autophagy genes affected SCA3 toxicity. Key genes to which RNA interference transgenic lines are available include Atg5. Whereas reduction of Atg5 activity on its own had little effect, reducing Atg5 in the presence of pathogenic SCA3 protein appeared to enhance toxicity, with increased loss of retinal integrity (Figure 4L–4N). This suggests that, normally, autophagy may mitigate toxicity of the pathogenic protein. Reduction of autophagy also enhanced aggregation of the protein by western immunoblot and enhanced cytoplasmic protein accumulation along photoreceptor axons (Figure 4O–4Q).
We then determined whether strong modifier genes also modulated autophagy. We examined two situations: (1) to determine whether strong modifier genes could induce autophagy in well-fed animals when autophagy is normally minimal; (2) to determine whether they could block autophagy under starvation, when autophagy is stimulated as a protective mechanism. We tested these as we considered that a modifier may affect Ataxin-3 pathogenicity either by inducing autophagy-mediated lysosomal degradation of the pathogenic protein, or alternatively, by blocking autophagy if autophagy contributes to loss of the cells in response to mutant polyQ protein. These studies revealed that select modifiers induced, whereas others mitigated, autophagy (Figure 4H, 4I, and 4K). Of the class 1 chaperone modifiers, the two Hsp40 genes (DnaJ-1B345.2 and mrj E1050) increased autophagy, whereas Hsp70 and the class 3 modifier ImpEP1433 reduced autophagy. Although Dpld showed no effect in these assays, limiting autophagy by reduction of Atg5, Atg7, or Atg12 mitigated Dpld suppression, suggesting that its activity was dependent on autophagy (Figure 4R, 4S, and unpublished data). These and other data (Figures S8 and S9) suggested that Dpld may act upstream of autophagy genes to activate autophagy in select situations. We also tested available GFP protein trap lines to examine localization of modifier proteins. Although none of these lines showed GFP immunostaining, one line with a protein trap insertion in Hsc70Cb enhanced SCA3 neuronal toxicity and increased protein accumulation in the neurophil similar to autophagy genes (Figure S8G–S8K). Taken together, these studies indicated that the modifiers, despite broad molecular nature, mitigated situations of protein misfolding; in some cases their activity appeared dependent on the proteasome, whereas others may involve autophagy-based protein clearance or autophagy-related cell loss.
The modifiers were selected based on ability to modulate SCA3 degeneration; however, our studies suggested that the modifiers may have broader functions in protein misfolding. We therefore determined whether they could modulate toxicity of tau. Abnormal accumulation of tau in neurofibrillary tangles or mutations in tau are associated with Alzheimer disease and frontotemporal dementia [39]. Tau-induced degeneration is mitigated by the caspase inhibitor P35 and DIAPs, implicating programmed cell death pathways in tau toxicity [40]. Expression of normal (tau.wt) or mutant (tau.R406) tau causes toxicity reflected in a reduced and degenerate eye [41]. Co-expression of the class 1 chaperone Tpr2EB7-1A and the Hsc70Cb line class 2 modifier polyubiquitin CR11700EP1384, and class 3 modifiers dpld JM265, ImpEP1433, and CG5009B227.2 strikingly suppressed toxicity of tau (Figure 5; Figure S8). The class 3 modifier embE2-1A enhanced tau (Figure 5), whereas NFAT EP1335 enhanced tau.wt, but had no effect on mutant tau.R406W (unpublished data). We then examined the ability of the modifiers to affect programmed cell death. Several class 3 modifiers had an effect on hid-induced eye loss: co-expression of embE2-1A and NFATEP1508 enhanced hid, whereas ImpEP1433 and dpldJM265 alleles suppressed hid-induced cell death (Figure 5 and unpublished data). Alleles of those genes that modulated tau and programmed cell death similarly (emb, dpld, NFAT, and Imp) may modulate tau toxicity by altering cell death. In contrast, the others (Tpr2, polyubiquitin CR11700, and Hsc70Cb) likely modulate tau toxicity through other means. Taken together, these data indicate that select modifiers that influence cell survival and protein misfolding may be common to both SCA3 and tau-induced degeneration.
We present a detailed functional analysis of genetic modifiers of pathogenic Ataxin-3 induced neuronal toxicity. The majority of modifiers fell into chaperone and ubiquitin pathways, consistent with the proposed function of Ataxin-3 in the ubiquitin-proteasome pathway, as well as pathways implicated in polyQ toxicity. Strikingly, however, biological assays for the activity of modifiers, some of which are predicted to function in diverse biological processes, indicated that their activities also converge on protein misfolding. Several protein quality control pathways appear involved, including the UPS and autophagy. These findings underscore the significance of protein quality control pathways to SCA3 and potentially other age-dependent protein conformation diseases.
Our screen was designed to identify upregulation modifiers, targeting genes whose increased activity could modulate neuronal toxicity and degeneration of a pathogenic Ataxin-3 protein. This screen identified activities that may become compromised or normally insufficient in the disease protein situation. A number of modifiers appeared dosage sensitive, in that reduction of the genetic region encoding the gene enhanced degeneration. This suggests that, at least for select modifiers, their normal activity is also critical for maintenance of neurons in disease.
Because of the severity of the degenerate eye, we may have selected for particularly strong modifiers. Modifiers were also initially selected for the ability to mitigate an external eye degeneration, rather than direct neuronal toxicity. Despite this, all of the modifiers mitigated internal neural degeneration. This indicates that the degree of external eye degeneration is a reasonable predictor of internal neuronal integrity. A number of modifiers were identified multiple times (dpld, NFAT, emb, and dbr), although most were found only once, indicating that the screen is not saturated. Additional screens using different mutagens or different types of screening approaches may reveal additional and different classes of modifiers. Although the modifiers were selected for mitigation of toxicity of a truncated Ataxin-3 protein, nearly all effectively mitigated toxicity of the full-length pathogenic protein. Our previous and other studies suggest that truncation may normally occur in disease to remove the N-terminal ubiquitin protease domain and would dramatically enhance degeneration [15,29,30]. Therefore, these are efficacious modifiers that mitigate toxicity of various forms of the pathogenic protein.
The majority of modifiers were either chaperones or, by sequence and functional tests, modulated ubiquitin-mediated protein quality control pathways. Whereas those that have sequence implications in this pathway might be anticipated, a surprise was that many genes with widely divergent molecular activities were functionally implicated in this process. These data suggest that many genes can modulate situations of protein misfolding, and moreover, suggest that protein misfolding is central to polyQ pathogenesis. It is also possible that, given the normal function of Ataxin-3 in ubiquitin-modulated pathways, the screen selected for modifiers that impinge on normal activities of Ataxin-3. Among the modifiers, some decreased protein accumulation, whereas others suppressed with no apparent change in Ataxin-3 inclusions. Of those that decreased Ataxin-3 accumulation, some appeared dependent on functional proteasome activity, whereas others stimulated autophagy, or both, implicating various mechanisms that attenuate Ataxin-3 degeneration. Among the modifiers was polyubiquitin, arguing that the level of ubiquitin itself may normally be insufficient to handle the misfolded protein over prolonged periods. This is consistent with the lack of a robust stress response to pathogenic polyQ [42]. In a Caenorhabditis elegans of model of polyQ, Gidalevitz et al. [43] reveal that polyQ expansions can cause temperature-sensitive alleles of various unrelated genes (paramyosin, dynamin, and ras) to display the mutant effect at what would normally be permissive conditions. This finding suggests that long polyQ runs can cause a general disruption of the cellular protein-folding environment to affect the temperature-sensitive mutant proteins. Our work used a pathogenic human disease protein Ataxin-3 and revealed that many modifiers that mitigate Ataxin-3 toxicity also modulate general protein misfolding. Taken together, these results suggest that, in the absence of upregulation of the various components needed, the cell may become overwhelmed, triggering deleterious effects [32]. Our detailed analysis of modifiers of Ataxin-3 have revealed a variety of biological processes that can help manage pathogenic polyQ protein, as well as specific genes of interest (Figure 6A).
The precise molecular pathways by which those modifiers that are not clearly within protein folding pathways can modulate misfolding to affect neurodegeneration requires further study. We envisage, however, that they function by their predicted molecular mechanisms, but on targets that impinge on cellular protein homeostasis. For example, chromatin modifiers may tweak the expression of a variety of genes in such processes, whereas translational regulators may translationally modify such genes. The finding that a variety of cellular functions are involved would be consistent with RNA interference screens in C. elegans that identified a variety of genes that impinge on cellular protein homeostasis as enhancers of the aggregation of polyQ protein [20]. These findings, along with other studies including those we present here, underscore the critical importance of proper protein homeostasis to most—if not all—cellular functions, such that a variety of genes can influence this process.
Among our modifiers, some that might have been expected to act in a similar manner, appeared to have distinct biological effects. The class 2 modifiers showed complete dependence on proper proteasome activity, while other modifiers, including alleles of DnaJ-1 and Imp, appeared to be insensitive to limiting proteasome activity, but rather affected autophagy in stress conditions. Hsp70 suppresses multiple stress conditions including protein misfolding, starvation-induced autophagy, and paraquat-induced oxidative stress [11], suggesting it may both facilitate UPS activity, but also block autophagy-dependent degeneration. In vivo, these pathways interplay to maintain proper neuronal function in the face of disease, thus further study of the modifiers may allow greater molecular identification of the individual pathways, as well as their integration.
Although only select modifiers affected protein accumulation as assayed by immunohistochemistry, all affected the solubility of the disease protein by biochemical analysis. The relationship between protein inclusions and proposed toxic oligomers is still under investigation, but it seems reasonable to suggest that the change in solubility may reflect activity of the modifiers to buffer or alter toxic conformations of the protein. Our efforts to detect toxic oligomers of Ataxin-3 using available antibodies [33,44,45] are still in progress. We note that, despite gross similarity, SCA3 degeneration is not identical to general misfolding: some suppressors of Ataxin-3 toxicity strikingly enhanced dominant-negative Hsp70 (upregulation of DnaJ-1 and Tpr2), whereas the enhancer CG11033E3093 suppressed it.
The UPS and autophagy pathways function in either degeneration or polyQ protein pathogenicity [46]. Our findings indicate that these pathways at least in part function in the removal or decrease of the toxic protein, thereby reducing degeneration. It is also possible that activities of these pathways may be inhibited during pathogenesis; such inhibition of normal activity would contribute to disease pathology [47–49]. The normal function of Ataxin-3 is in ubiquitin-modulated pathways [23–26]; our data suggest the possibility that Ataxin-3 may also modulate autophagy. Normal physiological levels of autophagy are critical to integrity of neurons, as loss of autophagy causes degeneration associated with ubiquitinated inclusions [50,51]. Our studies also reveal that compromise of autophagy pathways strikingly increased cytoplasmic Ataxin-3 accumulations in the neurophil without an obvious change in NIs. This may indicate that autophagy normally modulates cytoplasmic accumulation of the disease protein. Previous studies show that cytoplasmic polyQ protein is very toxic and blocks axonal transport [52]. These findings suggest that perturbations in autophagy may enhance cytoplasmic toxicity further.
In addition to regulation of the disease protein level and misfolding, results with select modifiers (Hsp70 and Imp) suggest that autophagy may also regulate loss of the cells. Given that our previous findings failed to reveal a clear role of caspase-dependent cell death in SCA3 [22], autophagy may be a mechanism of cell loss in this situation. Identification of other components of UPS and autophagy pathway will give further insights into how the disease protein is degraded and mechanisms of neuronal loss.
Genetic screens have been performed for modifiers of Ataxin-1 [19] and of pure polyQ domains [18], revealing similar components of protein folding and degradation: DnaJ-1 [18,19] and Tpr2 [18], but also RNA binding proteins and transcription factors. Although components of ubiquitin pathways that modulate Ataxin-1 appear distinct from Ataxin-3, this may reflect different regulators required for modifying different proteins or lack of saturation of the screens. Similarly several RNA binding proteins identified as modifiers of Ataxin-1 were suggested to reflect Ataxin-1 function as a putative RNA binding protein; our data suggest that it is also possible that these modifiers work more fundamentally to modulate protein solubility or levels. Our other work, however, has revealed a role for microRNAs in modulating neuronal survival in neurodegenerative situations [22,53], which has recently been extended to vertebrate neuronal integrity as well [54]. A C. elegans screen revealed a large number of modifiers of a pure polyQ aggregation phenotype [20]. Several modifiers identified in that screen affect RNA synthesis, processing, and protein synthesis as modifiers of misfolding. It will be important to test these modifiers against various specific disease proteins, as the action of a pure polyQ repeat may be distinct from a pathogenic repeat within a host protein.
One reason for global commonality among modifiers of polyQ disease proteins may be that the proteins themselves fall within a common interacting protein network. Recent studies using the yeast two-hybrid approach and proteomic databases reveal a protein interaction network of the ataxia proteins [55]. A surprise was that many different ataxia proteins, including Ataxin-3, fall within a few interaction steps from one another, suggesting that their common phenotypes may be a reflection of common interactions, which, when perturbed, contribute to disease. In that study [55], seven proteins were identified as direct interaction partners of Ataxin-3; however none of these appeared in our screen. Another direct binding protein–VCP [56]—was also not identified. However, the human orthologues of the majority of the genetic modifiers we defined fit into the protein interaction network at various points; some are direct interactors of other ataxia disease proteins, and others are one or more steps removed (Table S3).
Alzheimer disease and polyQ diseases are two unrelated human neurodegenerative diseases that cause neuronal degeneration in distinct brain regions [1,2,57]. Therefore, no overlap was expected between modifiers of these disease proteins in flies; indeed previous studies suggest minimal overlapping genes [58]. Surprisingly, we found that select modifiers of Ataxin-3 suppressed tau-degeneration, including the cochaperone Tpr2 and polyubiquitin (Figure 6B). This finding is consistent with cell culture studies and C. elegans RNA interference screens that implicate chaperones as modifiers of tau degeneration [59,60]. Further study of these modifiers, especially those that may be in common among different disease proteins, should provide the foundation for new therapeutic insight.
Fly lines were grown in standard cornmeal molasses agar with dry yeast at 25 °C. Transgenic lines UAS-SCA3trQ78, UAS-SCAtrQ61, UAS-SCA3Q78, and UAS-SCA3Q84 are described [14–16]. General stock lines, deficiency lines, and specific alleles of the EP modifiers were obtained from the Bloomington Drosophila stock center. The emb mutant lines were from C. Samakovlis (Wenner-Gren Institute Stockholm University, Stockholm, Sweden) [61]. rh1-GAL4, gmr-hid, UAS-DTS5, autophagy inverted repeat lines (UAS-Atg5_IR and UAS-Atg7_IR), tau transgenic lines (UAS-htau and UAS-htau.R406W), UAS-brat, and Hsc70C protein trap line were kindly provided by C. Desplan (New York University, New York, New York, United States), H. Steller (Rockefeller University, New York, New York, US), J.M. Belote (Syracuse University, Syracuse, New York, US), T. Neufeld (University of Minnesota, Minneapolis, Minnesota, US), M. Feany (Harvard Medical School, Boston, Massachusetts, US), R. Wharton (Duke University Medical Center, Durham, North Carolina, US), M. Buszczakl (Johns Hopkins University, Baltimore, Maryland, US), respectively. The dominant-negative Hsp70 line is described [32]. UAS-dpld FLAG, untagged, and gfp fusion lines were generated from cDNA clone LD02463 by subcloning into the pUAST vector [62]. Excision lines for reversion were made by crossing the EP insertions to lines bearing transposase, then screening for loss of the EP element.
For the genetic screen, virgin females of the starter line EP55, bearing an EP insertion on the X chromosome [27], was crossed to males with transposase. Males were selected and crossed to virgin females that expressed the disease gene (w/w; gmr-GAL4 UAS-SCA3trQ78 Tft/CyO). F1 progeny were screened for males with either suppressed or enhanced eye phenotypes as compared to controls; any modifiers were then outcrossed and balanced. EP insertions were confirmed to be single insertions through outcrossing, as well as plasmid rescue.
Epon sections, paraffin sections, and cryosections of adult heads were performed as described [14,16]. Primary antibodies for immunostaining were anti-HA primary antibody (Y-11, 1:50, Santa Cruz Biotechnology; and 12CA5, 1:100, Roche), anti-Myc (9E10, 1:100, Santa Cruz Biotechnology), anti-Hsp70 (7FB, 1:1,000,) [63], and anti-Gfp (A6455, 1:50, Molecular Probes). Secondary antibodies included anti-mouse or anti-rabbit conjugated to Alexa Fluor 594 or 488 (1:200 or 1:100, Molecular Probes). Western immunoblots were performed as described [16]. Primary antibodies used were HA-HRP (3F10, 1:500, Roche), rat anti-Hsp70 (7FB, 1:2,000), and mouse anti-tubulin (E7, 1:2,000, Developmental Studies Hybridoma Bank), with the secondary goat anti-rat IgG (1:2,000, Roche) and goat anti-mouse IgG (1:2,000, Chemicon International).
To define the sites of EP insertion, plasmid rescue was performed [27]. To confirm single hits in the genome, multiples of clones from each EP line were sequenced with an EP 3′ P-end specific primer (5′-CAA TCA TAT CGC TGT CTC ACT CA-3′). The flanking DNA sequence was used to query flybase BLAST to define the nearby gene and exact site of insertion. Upregulation of the genes by the EP element was confirmed by northern analysis, comparing the EP line alone, to wild-type fly controls with the EP line in the presence of a GAL4 driver, using gene-specific probes generated by reverse-transcription PCR using an EP-specific primer and gene-specific primers to the target genes. That the EP insertions were single insertions was independently verified genetically by molecular and phenotypic analysis of lines.
The effect of the modifiers on transgene expression levels was examined by real-time PCR (see Text S1 and Figure S2) and by reverse-transcription PCR. For the latter, total RNA was extracted with the RNAeasy kit (74104, QIAGEN) following manufacturer's instructions. cDNA synthesis was done using SuperScript First Strand Synthesis for reverse-transcription PCR (12371–019, Invitrogen). PCR was performed using primers: for the SCA3 transcript, 5′-CTATCAGGACAGAGTTCACAT-3′ (forward) and 5′-CAGATAAAGTGTGAAGGTAGC-3′ (reverse); for the GAL4 transcript, 5′-GTCTTCTATCGAACAAGCATGCGA-3′ (forward) and 5′-TGACCTTTGTTACTACTCTCTTCC-3′ (reverse) and for rp49 control, 5′-CCAGTCGGATCGATATGCTAA-3′ (forward) and 5′-ACCGTTGGGGTTGGTGAG-3′ (reverse).
Phototaxis was performed as described [15]. The percentage of flies in the light and dark chambers represent an average of three independent groups of flies. At least 100 flies were tested for each genotype, 20 flies were used in each experiment.
Third instar larval fat body tissues were stained with Lyso Tracker Red DND-99 (L-7528, Molecular Probes) and Hoechst (H3570, Molecular Probes) as described [38]. For each genotype, well-fed and starved larvae were used as negative and positive controls for the assay conditions. A score of 4 was given to control larvae grown under starvation condition; autophagy for other genotypes was scored relative to this. Controls (driver line alone and/or modifier alone, in starvation and well-fed conditions) were performed for each modifier genes in parallel to the experimental situation. The final autophagy score represents the average of 20 larvae from each genotype.
The accession numbers of the proteins used in the analysis in Figure S5 from the National Center for Biotechnology Information (NCBI) (http://www.ncbi.nlm.nih.gov) are Dm Dappled (NM_165533), DmCG15105 (NM_137546), Dm Brat (NM_057597), Dm Mei-P26 (NM_143765), Ce Lin-41 (NM_060086), Hs Lin-41 (XM_067369), Hs TRIM2 (NM_015271), Hs TRIM3 (NM_006458), and Hs TRIM32 (NM_012210). |
10.1371/journal.pntd.0000711 | Clinical Outcomes of Thirteen Patients with Acute Chagas Disease Acquired through Oral Transmission from Two Urban Outbreaks in Northeastern Brazil | Outbreaks of orally transmitted Trypanosoma cruzi continue to be reported in Brazil and are associated with a high mortality rate, mainly due to myocarditis.
This study is a detailed report on the disease progression of acute Chagas disease in 13 patients who were infected during two micro-outbreaks in two northeastern Brazilian towns. Clinical outcomes as well as EKG and ECHO results are described, both before and after benznidazole treatment.
Fever and dyspnea were the most frequent symptoms observed. Other clinical findings included myalgia, periorbital edema, headache and systolic murmur. Two patients died of cardiac failure before receiving benznidazole treatment. EKG and ECHO findings frequently showed a disturbance in ventricular repolarization and pericardial effusion. Ventricular dysfunction (ejection fraction <55%) was present in 27.3% of patients. After treatment, EKG readings normalized in 91.7% of patients. Ventricular repolarization abnormalities persisted in 50% of the patients, while sinus bradycardia was observed in 18%. The systolic ejection fraction normalized in two out of three patients with initially depressed ventricular function, while pericardial effusion disappeared.
Myocarditis is frequently found and potentially severe in patients with acute Chagas disease. Benznidazole treatment may improve clinical symptoms, as well as EKG and ECHO findings.
| Chagas disease is caused by a parasitic protozoan transmitted to humans by the contaminated feces of blood-feeding assassin bugs from the Triatominae subfamily. It may also be transmitted from mother to baby during pregnancy, by breastfeeding, blood transfusion or organ transplant. In rare cases, the disease can also be caused by accidental ingestion of contaminated food (sugar cane or açaí juice, drinking water, etc.). Acute Chagas disease often presents itself as a mononucleosis-like syndrome, with symptoms including fever, lymph node enlargement and muscle pain. The mortality rate of acute Chagas disease is high, mainly due to heart failure as a consequence of cardiac fiber lesions. There are few studies describing clinical outcomes and the disease progression of patients who receive therapeutic treatment, especially with regard to cardiac exam findings. In this report, the authors describe clinical findings from two micro-outbreaks occurring in impoverished towns in northeastern Brazil. Prior to receiving treatment, patient mortality rate was 28.6% in one of the outbreaks, and one pregnant woman experienced a spontaneous abortion due to the disease in the other outbreak. Most patients complained of fever, dyspnea, myalgia and periorbital edema. After receiving a two-month course of treatment, clinical symptoms improved and the number of abnormalities in cardiac exams decreased.
| American trypanosomiasis, or Chagas disease, is a zoonotic protozoan disease caused by the haemoflagellate Trypanosoma cruzi. The disease is endemic throughout Central and South America where about 17 million people are estimated to be infected and 100 million are at risk of infection [1]. In Brazil, the overall prevalence of Chagas disease is 4.2% and in the northeast region the infection rate can reach more than 5.0% [1]–[3]. In endemic areas, the primary infection usually occurs in children aged 15 years and under. More than 99% of acute Chagas disease cases are asymptomatic or appear as a nonspecific febrile disease. However, in untreated patients with severe symptoms of acute Chagas disease, the mortality rate rises to about 5–10% [1]. Moreover, 30% of infected individuals can develop chronic symptoms of Chagas disease over a lifetime [4].
Vectorial transmission of Chagas disease has decreased over the last decade in Brazil. Other forms of transmission, such as blood transfusion, congenital, organ transplants, and laboratory accidents are reported sporadically [1], [4]. The oral-accidental transmission of Trypanosoma cruzi is becoming increasingly common.
Since 1965, several outbreaks, possibly caused by oral-accidental routes, mainly due to ingestion of food, fresh water, “açaí” (Euterpe oleracea) or sugar cane juice, have occurred in many Brazilian states, including Rio Grande do Sul, Amazonas, Amapá, Santa Catarina, and Bahia [5]–[14]. In other Latin American countries, oral transmission has also been reported [15]–[16]. Oral infection with T. cruzi is associated with a high mortality rate, usually in the first two weeks after infection [8]–[11]. Mortality is mainly due to acute congestive heart failure, myocarditis and meningoencephalitis [4], [11]. Hemorrhagic manifestations and severe gastritis have also been reported [6].
Myocarditis is present in 80% of patients presenting severe symptoms of acute Chagas disease [4], [16]–[17]. Electrocardiography (EKG) and echocardiography (ECHO) show alterations such as atrial fibrillation and pericardial effusion, which are associated with a poor prognosis [11], [17]–[19]. In this study, we describe the clinical outcomes of 13 patients, before and after benznidazole treatment, during two micro outbreaks of acute Chagas disease in two towns located in the State of Bahia in northeastern Brazil.
The patients involved in this study came from two neighboring towns: Macaúbas (46,554 inhabitants) and Ibipitanga (13,109 inhabitants), both located in the south central region of the state of Bahia (approximately 700 km from the capital) in the Brazilian Northeast. In both towns, the annual per capita income is less than U$1,000 and the UN human development index is 0.62521. In May 2006, there was an outbreak of acute Chagas disease involving seven individuals from Macaúbas (cases 1–7) [8]. All individuals were members of the same family. Acute Chagas disease was suspected by a local physician and diagnosis was laboratory-confirmed in five cases (cases 1–5). Two patients (cases 6, 7) died as a consequence of heart failure before Chagas disease was confirmed. Oral contamination with T. cruzi probably occurred via ingestion of improperly stored water, possibly contaminated by feces of infested T. sordida [8].
The Ibipitanga outbreak occurred a few months after that of Macaúbas and involved six cases occurring among a family of 11. On August 9, 2006, the six cases: father (case 9), three sons (cases 10, 12, 13), one daughter (case 11) and his daughter-in-law (case 8) were working on a sugarcane plantation and drank a freshly-made sugarcane juice between 8:00 and 9:00 am, which they prepared in an abandoned sugarcane mill located next to the plantation. The cases developed symptoms between 11 and 21 days (between August 20 and 30, 2006) after the day they drank the sugar cane juice. A diagnosis of acute Chagas disease was suspected by a local physician 49 days after the ingestion of sugar cane juice (September 27, 2006). On October 8, 2006, the Epidemiological Surveillance Department of the Bahia Health Secretariat investigated the outbreak. Twelve specimens of Triatoma sordida were captured at the sugarcane mill, one of which was infested with T. cruzi. The diagnosis of Chagas disease was laboratory-confirmed for all six cases based on positive serological test results from samples collected on October 8, 2006, almost 60 days after exposure (Table 1).
The five family members who did not develop clinical symptoms had repeated negative serological test results for Chagas disease. All five reported to have drunk the same sugar cane juice prepared on August 9, 2006 which was also drunk by the six confirmed cases. However, the five uninfected family members drank the juice more than four hours after it was prepared and two of them had boiled the juice prior to ingestion.
The diagnosis for acute Chagas disease was confirmed using results from positive T cruzi parasitological tests: thick smear or blood culture; or a positive serologic test for IgM anti-T-cruzi antibodies: conventional enzyme-linked immunosorbent assay (ELISA), ELISA with recombinant antigens [20], or an indirect immunofluorescence antibody test (IFAT) (Table 1).
The study was approved by the institutional review board of CPqGM-FIOCRUZ, Bahia, Brazil. All patients and/or parents signed a letter of informed consent prior to examination.
All patients were treated after the Chagas diagnosis, which occurred between seven to 14 days and 27 to 37 days after the onset of symptoms in the Macaúbas and Ibipitanga patient groups, respectively. Oral benznidazole 300mg/day for 60 days was prescribed according to the Brazilian Consensus of Chagas Disease [9].
The EKG and ECHO were carried out before or shortly after beginning treatment, and 180 days after the end of specific treatment for Chagas disease. The criteria to define EKG alterations were based on the AHA/ACCF/HRS Recommendations for the Standardization and Interpretation of the Electrocardiogram and on the Guidelines of the Brazilian Society of Cardiology on Analysis and Report Issuance Electrocardiographic [21], [22]. The classification of the severity of the valve disease in adults was based on the American College of Cardiology/American Heart Association Practice Guidelines [23] and the quantification of cardiac chamber size and ventricular mass followed the criteria from American Society of Echocardiography's Guidelines and the European Association of Echocardiography [24].
The most frequent symptoms in the acute phase were fever (92%) and dyspnea (92%), myalgia (69.2%), periorbital edema (53.9%), headache, systolic murmur (46.2%), nausea, cough, abdominal pain, hepatomegaly (38.5%). Thoracic pain and vomiting were observed in four patients (30.8%), while palpitations were present in three patients (23.1%) (Table 2). Two patients (6, 7) from Macaúbas had heart enlargement, gallop rhythm (3rd sound), tachycardia, hypotension, and cardiac failure, resulting in death. Chest X-rays showed pleural effusion and cardiac enlargement (Figure 1A and B) in both patients. These patients were brothers and were the first in their family to develop the symptoms of this disease. In these cases, the diagnosis of Chagas disease was based on epidemiological findings alone. The mortality rate of Chagas disease, before benznidazole treatment, was 28.6% in Macaúbas and one pregnant woman from Ibipitanga experienced a spontaneous abortion prior to receiving treatment.
Table 3 displays the most frequent EKG and ECHO findings. EKG data were available for 12 out of 13 patients. At initial presentation, all 12 patients had a disturbance of ventricular repolarization. Right bundle branch block was observed in three out of 12 patients (25%) (cases 2, 3, 6) and sinus bradycardia was observed in cases 2, 12 and 13. Atrial fibrillation was present only in case 4 (8.3%).
ECHO results were available for 11 out of 13 patients. Seven out of 11 patients had major alterations: a mild degree of mitral regurgitation was observed in six out of 11 patients (54.6%) (cases 2, 3, 4, 10, 12, 13). Pericardial effusion was observed in cases 2, 3, 4, and 9. Tricuspide regurgitation was found in cases 4, 10 and 13 (27.3%), and dyskinetic septum was observed in cases 3 and 4 (18.2%). Ventricular dysfunction with low ejection fraction <55% was present in cases 9, 10 and 12 (27.3%). ECHO findings were normal for cases 1, 5, 8, and 11.
No adverse events during treatment with benznidazole were observed. As shown in Table 3, EKG results normalized in five out of 11 patients (91.7%), 180 days after treatment ended. Ventricular repolarization abnormalities persisted in six out of 11 patients (50%) while sinus bradycardia was observed in two patients (16.7%). The atrial fibrillation that was present in case 4 ceased after treatment. Regarding ECHO findings, mitral regurgitation persisted only in cases 3, 4 but disappeared in cases 2, 10, 12, and 13. After treatment, mitral regurgitation was present in case 9. Ventricular function normalized in cases 9, 10 and 12, and pericardial effusion was not present.
Acute Chagas disease caused by oral transmission has been increasingly reported in Brazil and other Latin American countries [5]–[16], [25]. However, few studies describe clinical outcomes after treatment with benznidazole [26]. In this report, the post-treatment clinical evolution of acute Chagas disease in patients from two impoverished rural towns in northeastern Brazil was observed. Oral transmission was determined to be the cause of both micro-outbreaks of acute Chagas disease. In the Macaúbas outbreak, patients were purportedly infected by the ingestion of stored water contaminated by the feces of infested triatomines [8]; while in Ibipitanga, oral transmission was due to the ingestion of sugarcane juice prepared in an abandoned sugarcane mill, where specimens of T. sordida contaminated with T. cruzi were captured.
Fever and dyspnea were experienced by nearly all patients. Other symptoms and findings indicating myocardial involvement, such as periorbital edema, chest pain and pericardial effusion, were observed in more than one-third of patients in both outbreaks. Hematological and digestive tract symptoms, including gastrointestinal bleeding and gastritis, were not found in our series, but were observed in patients from the Santa Catarina outbreak [6].
Morbidity and mortality rates of severe symptomatic acute Chagas disease are notably higher in children who contract the disease [11]–[12], [27]. Moreover, orally transmitted Chagas disease causes more severe symptoms in acute phases [8]–[11]. Prior to receiving treatment, two children from Macaúbas died as a consequence of heart failure (28.6% mortality rate), while in Ibipitanga, one pregnant woman experienced a spontaneous abortion. In both outbreaks, almost all exposed individuals developed severe manifestations of acute Chagas disease. In the Macaúbas outbreak the attack rate was 100% [8]. In the Ibipitanga outbreak, the six cases had drunk freshly-made sugarcane juice, while the five uninfected family members drank the juice more than four hours after it was prepared and two of them boiled the juice prior to ingestion. These five individuals remained asymptomatic and had repeated negative serological test results for Chagas disease. Insect-derived metacyclic trypomastigotes have specialized mechanisms that allow mucosal invasion. In orally-infected mice, trypomastigotes are able to invade and replicate in the gastric mucosa, causing a systemic infection [28]–[29]. In addition, metacyclic trypomastigotes taken from a patient who was orally infected with acute Chagas disease caused high parasitemia and a high mortality rate in orally-infected mice [30]. The development of myocarditis in acute Trypanosoma sp. infection is associated with intense edema of the cardiac fibers and the presence of inflammatory infiltrate containing amastigote forms of T. cruzi [4]. High levels of inflammatory cytokines are associated with myocardial damage. A recent study found higher levels of interferon-gamma, tumor necrosis factor-alpha, interleukin-10 and CCL3 in the myocardium of hamsters exhibiting acute symptoms of Chagas disease, when compared to asymptomatic animals [31].
EKG and ECHO abnormalities are frequently observed as a consequence of myocarditis. In the acute phase of Chagas disease, EKG findings may present alterations including low QRS voltage, prolonged PR and/or QT intervals, as well as T-wave changes [4], [17]. Ventricular extra systoles, sinus tachycardia, atrial fibrillation and advanced grade right bundle branch block are all associated with a poor prognosis [27]. In this study, during the acute phase of Chagas disease, ECHO exams were normal in only one-third of patients. Pericardial effusion was observed in 36% of patients. These findings were similar to those observed by Pinto et al during the acute phase of Chagas disease [26].
Six-months after the end of treatment with benznidazole, every patient showed an improvement in clinical symptoms, as well as a decrease in the number of EKG and ECHO abnormalities. Clinical signs of cardiac dysfunction, such as ejection fraction by ECHO, showed improvement in patients with more severe cardiac manifestations (cases 9, 10 and 12). In addition, 45% of patients (5 out of 11) with acute Chagas disease had a normal EKG at the end of treatment. We cannot conclude that improvements in EKG and ECHO findings were directly attributable to therapy, as opposed to the natural course of the disease. Although there is ample information on the clinical evolution of chronic Chagas disease, there are few studies that have evaluated the effect of benznidazole treatment on EKG and ECHO exams in patients with acute Chagas disease who have been followed from initial presentation to convalescence [11], [26]–[27]. In children in the early chronic phase of Chagas disease, after three to four years of follow-up, no conclusive evidence was obtained to indicate that benznidazole treatment, when compared with a placebo, could revert EKG abnormalities [32]–[33]. However, adults with indeterminate and chronic Chagas disease who were treated with benznidazole developed fewer electrocardiographic abnormalities when compared with untreated patients [34]. The present study exclusively involved patients in the acute phase of Chagas disease, for whom treatment is mandatory [8]. As such, it would be unethical to deprive these patients of treatment, making a clinical trial with a placebo control impossible. The efficacy of benznidazole in acute Chagas disease is demonstrated by a reduction in parasite load [35]. The cure rate for parasitological acute Chagas disease ranges from 60 to 80%, depending on patient age, dosage and whether treatment was initiated at the beginning of the infection [36]–[37].
We can conclude that cardiac alterations occur frequently, and are potentially severe, in the acute phase of orally transmitted Chagas disease. Furthermore, EKG and ECHO findings may have an impact on the clinical management of the disease, enabling monitoring of disease progression both during and after benznidazole treatment. Further studies are necessary to evaluate the persistence of EKG and ECHO abnormalities over the long term, as well as morbidity and mortality.
|
10.1371/journal.pntd.0003534 | Repurposing Auranofin as a Lead Candidate for Treatment of Lymphatic Filariasis and Onchocerciasis | Two major human diseases caused by filariid nematodes are onchocerciasis, or river blindness, and lymphatic filariasis, which can lead to elephantiasis. The drugs ivermectin, diethylcarbamazine (DEC), and albendazole are used in control programs for these diseases, but are mainly effective against the microfilarial stage and have minimal or no effect on adult worms. Adult Onchocerca volvulus and Brugia malayi worms (macrofilariae) can live for up to 15 years, reproducing and allowing the infection to persist in a population. Therefore, to support control or elimination of these two diseases, effective macrofilaricidal drugs are necessary, in addition to current drugs. In an effort to identify macrofilaricidal drugs, we screened an FDA-approved library with adult worms of Brugia spp. and Onchocerca ochengi, third-stage larvae (L3s) of Onchocerca volvulus, and the microfilariae of both O. ochengi and Loa loa. We found that auranofin, a gold-containing drug used for rheumatoid arthritis, was effective in vitro in killing both Brugia spp. and O. ochengi adult worms and in inhibiting the molting of L3s of O. volvulus with IC50 values in the low micromolar to nanomolar range. Auranofin had an approximately 43-fold higher IC50 against the microfilariae of L. loa compared with the IC50 for adult female O. ochengi, which may be beneficial if used in areas where Onchocerca and Brugia are co-endemic with L. loa, to prevent severe adverse reactions to the drug-induced death of L. loa microfilariae. Further testing indicated that auranofin is also effective in reducing Brugia adult worm burden in infected gerbils and that auranofin may be targeting the thioredoxin reductase in this nematode.
| Onchocerciasis or river blindness, and lymphatic filariasis, which can lead to disfiguring elephantiasis, are two neglected tropical diseases that affect millions of people, primarily in developing countries. Both diseases are caused by filariid nematodes; onchocerciasis is caused by Onchocerca volvulus and lymphatic filariasis is caused by Brugia malayi, B. timori, and Wuchereria bancrofti. Currently, there are no drugs available that are highly efficacious against adult worms; existing drugs mainly kill the first-stage larvae (microfilariae). While these drugs can reduce the transmission of infections in a population, the adult filariids (macrofilariae) can continue to produce microfilariae and perpetuate the cycle of infection. Finding a drug that could kill the adult worms would be an important tool in eliminating onchocerciasis and lymphatic filariasis. To identify potential macrofilaricidal drugs, we developed a high throughput screening method to test FDA-approved drugs on adult Brugia spp., which serves as a model for O. volvulus. Using this screening method, we identified a drug called auranofin that kills adult Onchocerca and adult Brugia spp. in vitro, inhibits the molting of O. volvulus L3s, and reduces the worm burden in an in vivo gerbil-B. pahangi model system. Auranofin is known to inhibit a critical enzyme called thioredoxin reductase in some parasite species, and subsequent testing of the effects of auranofin on the thioredoxin reductase of Brugia indicates that this may be auranofin’s mode of action in this nematode as well.
| River blindness and lymphatic filariasis (LF) are two major neglected diseases caused by filariid nematodes that, together, affect an estimated 145 million people worldwide in mostly poor, developing countries [1,2]. River blindness, caused by the filariid nematode Onchocerca volvulus, is a chronic, debilitating disease and a major cause of infectious blindness. The adult worms, or macrofilariae, reside in subcutaneous tissues where females release the early larval stage, microfilariae, into the skin. Adult worms can reproduce for up to 10–14 years, releasing millions of microfilariae over an infected individual’s lifetime [3]. Microfilariae migrate throughout the tissues and those that accumulate in the eyes induce an inflammatory response that eventually leads to blindness [4]. LF is caused by several species of filariid nematodes: Wuchereria bancrofti, Brugia malayi and B. timori. The adult worms reside in the lymphatic tissues where females release microfilariae into the circulation. The microfilariae are then ingested by mosquitoes and develop into the infectious larval stage. With LF, the chronic condition is characterized by pain and severe lymphedema often involving the arms, legs, breasts and genitalia, as well as elephantiasis, all of which may lead to social stigma and economic loss to those afflicted [4,5].
Currently, global health programs that aim to eliminate these diseases distribute ivermectin, diethylcarbamazine (DEC), and albendazole through mass drug administration (MDA) to reduce transmission and ideally break the cycle of infection [6]. However, these drugs mainly target the microfilarial stage of the parasite, leaving the adult worms to continue to reproduce. DEC can cause adverse effects in patients infected with O. volvulus, so it can only be used to treat LF in areas where onchocerciasis is not endemic [4,6]. There is also an increased risk of serious adverse events, including encephalopathy and death, in those individuals who are treated with ivermectin or DEC and are co-infected with Loa loa with high microfilaraemia (greater than 30,000 microfilariae per mL) [7–10]. Recently, the veterinary drug, moxidectin has been investigated as a potential new therapeutic for filarial infection. Awadzi et al (2014) found that moxidectin was an effective microfilaricidal drug in a small-scale study, but it could not be concluded that moxidectin was macrofilaricidal or caused sterility in adult worms [11]. The antibiotic, doxycycline, has been shown to be safe and efficacious in treating both lymphatic filariasis and onchocerciasis, and can sterilize and eventually kill adult worms. However, doxycycline requires long treatment periods of upwards of 4–6 weeks, which is unlikely to be feasible for MDA [4]. These factors, in addition to the difficulty of attaining sufficient coverage through MDA, make discovering effective macrofilaricidal treatments to cure infections a high priority in stopping the transmission of filariasis. An ideal drug candidate is one that has high specificity for Onchocerca and Wuchereria/Brugia macrofilariae, but has little to no effect on the microfilariae of L. loa.
The overall goal of our program is to identify lead candidates for the treatment of river blindness and LF. Previously, we developed an in vitro worm assay [12] using Brugia pahangi and B. malayi as a primary screen to identify compounds that inhibit worm motility. The WormAssay apparatus and computer software (Worminator) enables us to screen compounds against adult Brugia in 24-well plates in less than one minute and assess worm killing in an objective manner. Compounds that strongly inhibited adult worm motility in a 3-day assay were then tested against molting O. volvulus third-stage larvae (L3) and adult O. ochengi. Adult O. ochengi, which naturally infect cows and develop in subcutaneous nodules, serve as a model organism for O. volvulus, which only infects humans and non-human primates [13–15].
In this study, we screened a library of over 2,000 FDA-approved compounds and found that auranofin was highly effective in inhibiting adult Brugia motility. Auranofin is an FDA-approved, gold-containing compound (2,3,4,6-tetra-O-acetyl-1-thio-beta-D-glucopyranosato-S (triethylphosphine) gold) that has been used to treat rheumatoid arthritis for over 25 years [16,17]. Orally dosed auranofin is rapidly metabolized in vivo but its active metabolite is not known. It has been suggested that triethylphosphine gold or deacetylated auranofin could be the biologically active metabolites and that some form of the gold from auranofin circulates bound to plasma protein [18–20]. Since gold is known to be necessary for auranofin’s drug activity, studies of its pharmacokinetics employ elemental analysis for gold [19,21–24]. Previous studies have shown that the likely target of auranofin is thioredoxin reductase (TrxR) [25,26], which is a key enzyme involved in reducing oxidative damage in cells. We also found that auranofin is effective in killing adult Brugia in an in vivo gerbil model and that TrxR is most likely the target of auranofin in Brugia.
Adult female and male Brugia (B. malayi and B. pahangi) were shipped from TRS Labs Inc., Athens, GA and assayed using methods described by Marcellino et al. (2012) [12]. Individual females were placed in each well of a 24-well plate with media (RPMI-1640 with 25 mM HEPES, 2.0 g/L NaHCO3, 5% heat inactivated FBS, and 1X Antibiotic/Antimycotic solution). Excess media was removed from plates using a Biomek FxP, leaving 500 μL of media per well. Initial screening of a library of FDA-approved compounds, compiled by the Small Molecule Discovery Center at the University of California San Francisco, was conducted at 10 μM per compound, and 1% DMSO was used as a negative control. All drugs including auranofin (Enzo Life Sciences, Farmingdale, NY) were dissolved in DMSO (Sigma-Aldrich, St. Louis, MO) and 10 mM stock solutions were stored at -20°C. Four worms were used as replicates for each concentration and worm plates were kept in a 37°C, 5% CO2 incubator for four days. Auranofin was also tested against male Brugia worms under the same conditions after initial screening against female Brugia revealed its high level of inhibitory activity.
To determine the effect of a compound on worm motility, individual worm movements were counted as the number of pixels displaced per second by each worm in each well using the Worminator. Each plate of worms was video recorded for approximately 60 seconds, and mean movement units (MMUs) were determined for individual worms. Percent inhibition of motility was calculated by dividing the MMUs of the treated worms by the control average MMUs, subtracting the value from 1.0, flooring the values to zero and multiplying by 100%. Videos were recorded for 4 days, including the first day of the assay (Day 0). IC50 determinations were conducted at 10 μM, 3 μM, 1 μM, 0.3 μM, 0.1 μM and 0.03 μM, with 1% DMSO used as a control. IC50 assays were repeated at the same concentrations and at six point, three-fold dilutions from 1 μM to 0.003 μM or 3 μM to 0.001 μM to ensure that activity was consistent between assays. Prism 4.0 was used to calculate IC50 values using a non-linear regression curve fit. The means of all IC50s with R2 values greater than or equal to 0.7 are reported.
Cows that had grazed in northern Cameroon where O. ochengi is highly endemic were brought to abattoirs located in Douala, Cameroon. Subcutaneous nodules containing adult O. ochengi worms were identified on the umbilical skin of infected cows. Adult worm masses containing one viable, adult female and zero to several adult males were carefully recovered by dissection of the nodule with a sterile razor blade. The masses were then incubated in 4 mL of complete culture medium (CCM), which was comprised of RPMI-1640 (Sigma-Aldrich), 5% newborn calf serum, 200 units/mL penicillin, 200 μg/mL streptomycin and 2.5 μg/mL amphotericin B (Sigma-Aldrich), in standard 12-well culture plates. Masses were maintained in the medium in a 37°C, 5% CO2 incubator overnight during which period most of the smaller and more agile adult males migrated out of the masses while the females remained in the nodules. Worm viability was checked microscopically by observing the movement of adult male worms or emergence of viable microfilariae from the nodular masses. The next day, 2 mL of the CCM was removed and replaced with 60 μM auranofin in 2 mL CCM in each well to generate a final drug concentration of 30 μM. The compound and controls were tested in quadruplicate at each concentration and the experiments were repeated twice on different days. The negative control wells received only 1% DMSO. Cultures were terminated on day 7 post addition of drug. Adult male worm viability was visually scored on day 5 as percent reduction of motility ranging from 100% (complete inhibition of motility), 90% (only head or tail of worm moving or vibrating), 75% (worm very sluggish), 50% (worm sluggish), 25% (little change in motility), to 0% (no observable reduction in motility). Adult female worm viability was assessed on day 7 by the standard MTT/formazan assay in which each nodular mass was placed in a well of a 48-well microtiter plate containing 500 μL of 0.5 mg/mL MTT (Sigma-Aldrich) in incomplete culture medium, and then incubated in the dark at 37°C for 30 minutes. Adult female worm viability was evaluated visually by the extent to which the female worm mass was stained with MTT. Mean percent inhibition of formazan formation was calculated relative to the negative control worm masses after 7 days in culture. Adult worm death positively correlated with inhibition of formazan formation.
To calculate the IC50 of auranofin, quadruplicate worm masses were incubated with final concentrations of 30 μM, 10 μM, 3 μM, 1 μM, 0.3 μM, 0.1 μM and 0.03 μM and assays were conducted as described above. Prism 4.0 for Windows was used to calculate IC50s.
O. ochengi microfilariae were obtained from the umbilical skin of infected cattle and cultured on confluent monkey kidney epithelial cells for drug testing as previously described [27].
Loa loa microfilariae were purified from the blood of a heavily infected subject (having approximately 10,000 microfilariae/mL of blood) using Percoll (GE Healthcare, Piscataway, NJ) gradient centrifugation. Venous blood (10 mL) was collected from consenting, infected individuals in an EDTA tube. The blood was layered on a step-wise Percoll gradient (46% and 43% Percoll prepared in CCM) followed by centrifugation at 400 rcf for 20 minutes. The L. loa microfilariae were recovered in the 43% layer, washed 3 times in CCM and counted.
Microfilariae (10–15 per well) were cultured in 96-well culture plates in duplicate under the same conditions and drug concentrations as were used for the adult O. ochengi, except that 10 μg/mL ivermectin was used as a positive control. Microfilariae viability was visually scored based on motility reduction using the same scale described above for adult male O. ochengi. Scores were recorded every 24 hours after the addition of drugs for 5 days using an inverted microscope.
L3 stage larvae previously collected and cryopreserved in Cameroon were rapidly thawed in a 37°C water bath and washed in incomplete media comprised of a 1:1 ratio of Medium NCTC-109 and IMDM + GlutaMax-I containing 1X glutamine, penicillin, and streptomycin (all from Gibco by Life Technologies, Grand Island, NY). The number of worms was adjusted to about 10 worms per 50 μL in complete medium containing 20% heat inactivated FCS. Worms were distributed into the wells of a 96-well plate containing 50 μL of 1.5 × 105 normal human PBMCs. 100 μL of 2X auranofin (final concentrations of 30 μM, 10 μM, 3 μM, 1 μM and 0.3 μM) were added to each well for a final volume of 200 μL. Each concentration was tested in triplicate. Controls included 0.05% DMSO in complete medium and complete medium only with neither DMSO nor compound added. The 96-well plates were then incubated at 37°C in a 5% CO2 incubator for 6 days, then molting was assessed using an inverted microscope. Molting was determined in each well by counting the presence of fourth-stage larvae (L4) and empty casts of the L3. The percent inhibition of molting was calculated based on the number of treated larvae that were able to molt in comparison to the number of control larvae that had successfully molted. Prism 4.0 for Windows was used to calculate IC50s.
Adult female B. pahangi worms were incubated with either 1 μM, 0.3 μM, or 0.1 μM auranofin, 10 μM flubendazole (as a positive control [28]), or 1% DMSO overnight, then cut into 3 segments separating the anterior, middle and posterior sections. The middle sections were further cut into 1 mm sized pieces in fixative (2.5% glutaraldehyde and 2% paraformaldehyde in 0.1 M sodium cacodylate buffer, pH 7.3–7.4) and stored at 4°C. Middle sections were subsequently treated with 1% tannic acid for 1 hour, followed by three buffer washes before post fixation staining with 2% osmium tetroxide for 1 hour. The samples were washed three times in buffer before dehydration in an ethanol series. Worm sections were then infiltrated with propylene oxide, embedded in epon 812 resin and polymerized in a vacuum oven at 60°C overnight. Ultrathin sections were cut using an RMC MTX ultramicrotome with a Diatome diamond knife followed by post staining of the grids with saturated ethanolic uranyl acetate and Reynolds lead citrate. Samples were imaged on a FEI Tecnai 12 spirit TEM operated at 80 kV. A similar procedure was performed on adult female O. ochengi worm masses that were cultured for 7 days with 10 μM auranofin before fixation of cut pieces of the adult female mass. Untreated adult female masses cultured for 7 days and fixed by the same procedure served as the control.
Animal studies were performed under IACUC approval #AN085723–02 to test the efficacy of auranofin in vivo. Male Mongolian gerbils (Meriones unguiculatus, Charles River Laboratories International, Inc., Wilmington, MA) were injected intraperitoneally (IP) with 300 B. pahangi L3 (Filariatech, Inc., Athens, GA) and treated 3 months post-infection. Auranofin was dissolved in 100% ethanol at 4 mg/mL and mixed 1:1 with PBS. Vehicle doses consisted of the same mixture of ethanol and PBS but without auranofin. Doses (up to 200 μl) were given to gerbils orally at 5 mg/kg body weight, BID weekdays and SID weekends for a total of 48 doses over 4 weeks.
Two studies (Study 1 and Study 2) were conducted using the same protocols and the same dosing schedule except that in Study 1, two gerbils from the auranofin treatment group and two gerbils from the vehicle group were treated for 14 days and were necropsied 2 hours after their last dose (interim necropsy) to determine plasma gold levels (from auranofin). The remaining gerbils in Study 1 were treated for 28 days and were necropsied 11, 14, or 16 days after the end of dosing. In Study 2, all gerbils were treated for 28 days and were necropsied 16 days after the end of dosing. For both of these in vivo studies, worms were collected from the gerbil’s peritoneal cavity, counted, sexed and examined under a dissecting microscope. For each study a two-tailed Student’s T-test assuming equal variance was conducted using Microsoft Excel to determine the statistical significance of the difference in mean worm retrieval between the auranofin treated and vehicle treated groups. Gerbil blood was collected by cardiac puncture and plasma was sent to NMS Labs, Willow Grove, PA to determine plasma gold levels (elemental gold analysis) by graphite furnace atomic absorption spectroscopy.
Thioredoxin reductase activity of worm lysates was assayed using female B. malayi treated in vitro with either 0.3 μM, 0.1 μM, or 0.03 μM auranofin or 1% DMSO. After 5 hours of treatment, worm motility was measured using the Worminator, and then worms (24 in each group) were pooled, washed three times in PBS, and lysed by douncing in a glass homogenizer in assay buffer (Abcam Thioredoxin Reductase Assay kit, ab83463) with 1 mM PMSF. The crude lysates were centrifuged at 10,000 rcf for 15 minutes at 4°C to pellet insoluble material. The total protein concentrations of soluble lysates were measured using the Bradford assay. The soluble lysates were incubated for 20 minutes in assay buffer or assay buffer with a proprietary thioredoxin reductase specific inhibitor before adding a specific substrate, DTNB (5, 5′-dithiobis (2-nitrobenzoic) acid), and measuring activity at 20 second intervals for 40 minutes using the SpectraMax Plus Microplate Reader (Molecular Devices, Sunnyvale, CA) at λ = 412 nm. Lysates were tested in duplicate. TrxR activity was calculated based on the linear amount of TNB produced per minute per mg of total protein and adjusted for background activity from enzymes other than TrxR in the lysates.
Thioredoxin reductase activity was also analyzed in worms that were treated with auranofin or vehicle in vivo. Adult male and female worms were transplanted intraperitoneally, and gerbils were treated with auranofin or vehicle for 28 days as was done in the previous in vivo studies. Gerbils were necropsied 16 days after the final dose, and lysates were prepared from recovered worms and assayed as above.
The open reading frame for B. malayi TrxR (XM_001898694.1) was synthesized (GenScript) with codons optimized for expression in Escherichia coli. The two C-terminal amino acids (selenocysteine (Sec)-Gly), missing in XM_001898694.1, were added along with a bacterial SECIS (selenocysteine insertion sequence) to allow expression of the Sec protein in E. coli in pET100 (Invitrogen by Life Technologies) [29]. For PCR, the reverse primer was 5’-GGCCGCATAGGTTAACGATTGGTGCAGACCTGCAACCGATTATTAACCTCAGCATCCCGTTGCTTTC-3’ and forward primer was 5’-CACCATGCTGCTGCGTTCCAATGC-3’.
To determine if the B. malayi TrxR is a selenoprotein, as are some known thioredoxin reductases, a bioinformatics search was conducted to find a SECIS in the B. malayi genome near the thioredoxin reductase gene. For a detailed description, please see supplementary information (S1 Text).
Recombinant 6-His-tagged B. malayi TrxR (rBmTrxR) was produced in E. coli BL21(DE3) in the presence of pSUABC [29] in LB medium supplemented with 20 μg/mL riboflavin under conditions for optimal selenoprotein expression [30]. An overnight starter culture in LB with 50 μg/mL ampicillin and 34 μg/mL chloramphenicol was diluted (1:100) into in LB medium with the same antibiotics. When the culture reached an OD600 = 0.8, the medium was supplemented with 5 μM sodium selenite and 100 μg/mL L-cysteine. When the culture OD600 = 2, riboflavin (20 μg/mL) was added and protein expression was induced by the addition of isopropyl β-D-1-thiogalactopyranoside (50 μM). At this point the cultures were shifted to 24°C and incubated for 24 hr. Cells were collected by centrifugation, lysed by alternative freeze-thaw cycles, and resuspended in lysis buffer (50 mM potassium phosphate, pH 7.8, 500 mM NaCl, 30 mM imidazole, 1 mg/mL lysozyme, 1 mM phenylmethanesulfonylfluoride) supplemented with 100 μM flavin adenine dinucleotide. The sample was sonicated and cellular debris pelleted at 25,000 x g at 4°C for 25 min. The supernatant was collected and filtered through a 0.45 μm filter before purification by immobilized metal ion affinity chromatography using a His-Trap FF column (GE Healthcare). The column was washed with 10 column volumes binding buffer (50 mM potassium phosphate, pH 7.8, 500 mM NaCl, 30 mM imidazole) and then with 5 column volumes of buffer A (binding buffer with 100 mM imidazole). TrxR protein was eluted in 3 × 1 mL buffer B (binding buffer with 500 mM imidazole).Protein was concentrated (Amicon Ultra-4 10K) and purity was verified by SDS-PAGE and quantified by absorbance at 280 nm (ε = 69.76 mM-1 cm-1).
rBmTrxR activity was assayed in 0.1 M potassium phosphate (pH 7.2) with 10 mM EDTA and 25 nM rBmTrxR. rBmTxrR was pre-incubated for 20 min with NADPH (100 μM) and auranofin (ICN Pharmaceuticals, now Valeant Pharmaceuticals, Bridgewater, NJ) or aurothioglucose (USP Reference Standards, Rockville, MD) in DMSO in 100 μL, followed by addition of an equal volume of buffer with NADPH (200 μM) and DTNB (6 mM) with reaction progress monitored at λ = 412 nm for TNB production. The concentration of DMSO in all reactions was 3.5%.
The Loa loa microfilariae donors were all adult male and female patients, aged 21 or older, residing in the Edea Health District of the Littoral Region of Cameroon. Ethical and administrative clearances were obtained from the Cameroon National Ethics Committee (N°2013/11/371/L/CNERSH/SP) and the Cameroon Ministry of Health, respectively. Written and signed informed consent was obtained from each participating patient, and all of them had 2000 L. loa microfilariae per mL of blood or greater. The patients were employed in the study as microfilariae donors only.
Animal studies were performed under the University of California San Francisco Institutional Animal Care and Use Committee (IACUC) approval #AN085723–02 and adhere to guidelines set forth in the NIH Guide for the Care and Use of Laboratory Animals and the USDA Animal Care Policies. Animals were euthanized by carbon dioxide inhalation followed by bilateral thoracotomy.
Results of the adult worm assay showed that the motility of female B. pahangi and B. malayi was inhibited by 97% within 18 hours of incubation with 3 μM of auranofin. Following our prescribed screening funnel, after this primary screen, auranofin was then assayed with adult female and male O. ochengi, O. volvulus L3, and O. ochengi and L. loa microfilariae.
Auranofin was highly effective in killing both male and female adult Brugia and Onchocerca worms and inhibiting molting of O. volvulus third-stage larvae to the fourth stage with IC50 values less than or equal to 1.1 μM (Table 1). Auranofin, however, was not very effective in killing O. ochengi and L. loa microfilariae. Auranofin’s IC50 value for adult female O. ochengi was 10 times lower than its IC50 value for O. ochengi microfilariae and 42.7 times lower than its IC50 value for L. loa microfilariae. This is an important consideration when treating individuals with auranofin in L. loa endemic areas.
Adult female B. pahangi incubated with 1 μM, 0.3 μM, or 0.1 μM auranofin overnight and adult female O. ochengi worms encapsulated in nodules incubated with 10 μM auranofin for 7 days were subjected to transmission electron microscopy to compare the internal morphology with their respective control female worms. Auranofin-treated B. pahangi worms showed considerable damage in the hypodermal region compared to control worms (Figs. 1a-1d). The hypodermal area of treated worms was highly vacuolated with remnants of swollen mitochondria containing dark bodies as well as shrunken Wolbachia containing dark condensed material. The hypodermal chord region of B. pahangi female worms treated with 10 μM of flubendazole contained normal Wolbachia with very few mitochondria containing dark bodies (Fig. 1e). In contrast, the hypodermal chord region in control worms (Fig. 1f) contained numerous Wolbachia without the condensed material observed in auranofin treated worms.
Similar morphology was also observed in the O. ochengi auranofin treated worms (Fig. 2). Numerous vacuoles with inclusion bodies were observed in the muscle tissue below the hypodermal chord. Numerous vacuoles and a complete lack of mitochondria were also observed in the hypodermal chord region directly below the cuticle.
Two in vivo studies were performed using the same dosing regimen of 5 mg/kg BID weekdays and SID weekends for 28 days (for a total of 48 doses). Study 1 and Study 2 are replicate studies, except that in Study 1 an interim necropsy was conducted to determine the plasma levels and level of infection 14 days after the first dose. The number of worms collected from these vehicle treated gerbils was 43 (13 male worms and 30 female worms, a ratio of approximately 1:2) and the total number of worms from the auranofin treated gerbils was 11 (4 males and 7 females, a ratio of approximately 1:2).
In Study 1, the average number of worms from all vehicle treated animals (n = 7) was 9.4 worms and the average number of worms from all treated animals (n = 9) was 4.0 worms (Fig. 3A). There was a 58% overall reduction in worm burden in the auranofin treated group in comparison with the vehicle treated group but difference between the two groups was not statistically significant (p > 0.05). In the control group the ratio of male to female worms at terminal necropsy was 1:2, similar to the ratios found in the control group and treated group at the interim necropsy. In the treated group however, the ratio of male to female worms was 12:1 at terminal necropsy. This sex ratio bias was also observed in the auranofin treated group in Study 2 (Fig. 3B).
In Study 2, there was a 91% reduction in worm burden in the auranofin treated group compared to the control group, which was statistically significant (p = 0.01) in a Student’s T-Test. There were 161 total worms recovered from the vehicle group (mean = 32 worms per gerbil), of which 55 were males and 106 were females (ratio of 1:2). In the auranofin treated group, there were a total of 12 worms recovered (mean = 3 worms per gerbil): 11 were males and only 1 was a female worm (ratio of 11:1) (Fig. 3D). This remaining female was encapsulated with host tissue.
Plasma collected from the necropsies from Study 1 and Study 2 was submitted for elemental gold analysis (Table 2). Gold was not detected in the vehicle group. Plasma taken 2 hours after gerbils were given an auranofin dose (but had been treated for 14 days) had gold levels of 5.08 μM and 8.63 μM. In Study 1 and Study 2, the mean plasma gold levels 16 days after the last dose were 701 nM and 609 nM, respectively. There were 2 animals in each of the treatment groups that did not have detectable levels of gold in their plasma but this may be due to the limit of detection in the assay, where any value less than 100 μg/L (508 nM) gold is given as zero.
Thioredoxin reductase activity in Brugia females cultured for 5 hours with 0.3 μM, 0.1 μM or 0.03 μM of auranofin in vitro was significantly reduced (p < 0.05) to 15%, 33% and 69% of endogenous activity, respectively, compared to the activity in DMSO-treated worms (Fig. 4A).
When Brugia worms were removed 16 days after the last dose from gerbils treated with auranofin in vivo, endogenous enzyme activity was reduced significantly (p < 0.05) by 49% compared to worms collected from vehicle treated gerbils (Fig. 4B). These data further suggest that endogenous Brugia TrxR is specifically inhibited by auranofin.
Recombinant B. malayi TrxR (rBmTrxR) was overexpressed in E. coli at approximately 10 mg of protein per liter of culture following His-Trap affinity chromatography. Two organic gold compounds, auranofin and aurothioglucose, were assayed with rBmTrxR and both were found to be effective inhibitors suggesting that gold is the active component of auranofin as expected from previous studies with TrxR and thioredoxin glutathione reductase [24,25]. Both compounds had inhibitory activity in the low nanomolar range with auranofin IC50 = 3 nM and aurothioglucose IC50 = 9 nM. Production of eukaryotic Sec-proteins in bacteria is not 100% efficient. The misreading of the Sec codon (UGA) results in premature termination of the peptide resulting in an enzymatically inactive product [29]. Since the active and inactive proteins both bind metal affinity resins and differ in size by only two amino acids, recombinant protein is a mixture of both active and inactive enzyme forms. Based on previous studies [29–31] between 10% and 20% of the protein is active, with the remainder inactive. The inhibitory activity of both compounds indicates that they irreversibly inhibit rBmTrxR at a one-to-one molar ratio, with potencies similar to those found for other TrxR and thioredoxin glutathione reductase enzymes [32,33].
The main goal of our study was to identify macrofilaricidal drugs for the treatment of onchocerciasis and LF. Two major challenges in developing new drugs for these neglected diseases are finding suitable animal models for preclinical studies and limiting the costs of drug development and production. To date, the only animals in which O. volvulus can develop to patency are chimpanzees and mangabey monkeys [34–36]. O. ochengi, which infects cows, is thought to be closely related to O. volvulus [37], and previous studies have used O. ochengi as a model for O. volvulus infection [13–15]. Brugia malayi and B. pahangi, as members of the Filariidae family, are also closely related to O. volvulus [38]. Because of the large number of compounds required to identify preclinical candidates and with the accessibility of large numbers of adult worms that can be collected from gerbils, we selected adult Brugia for our primary screens. Following our funneling scheme, we first identify compounds screened with adult female Brugia in the Worminator assays. Compounds that inhibit motility by 75% compared with control worms are then screened against O. volvulus molting larvae and O. ochengi adult worms in an MTT assay and motility assay.
In an effort to identify candidate drugs that could be more rapidly moved into clinical trials, we screened an FDA-approved library of compounds and found that auranofin was effective in killing adult Brugia and O. ochengi worms and in inhibiting larval O. volvulus from molting from L3s to L4s in vitro. Microfilariae of O. ochengi and L. loa were used in a counter screen to determine the effects of auranofin on the microfilarial stage. We found that the IC50s for O. ochengi and L. loa microfilariae were approximately 10 and 42.7 times higher, respectively, compared with the IC50s of adult female O. ochengi. These results may have important implications, should auranofin be used for treatment in areas endemic for both onchocerciasis and loaiasis to avoid severe adverse events.
Auranofin was then tested for its efficacy in secondary screens with infected gerbils. Results of the in vivo studies showed that dosing animals for 28 days at 5 mg/kg was effective in reducing worm burden by 58% and 91% in the two studies. Gold plasma levels in gerbils obtained at 2 hours post-dose after 2 weeks of treatment indicated that the plasma gold levels were in the micromolar range (5.08 μM and 8.63 μM), approximately 5 to 10-fold higher than the IC50s from the in vitro worm assays. These gerbils continued to maintain gold levels in their blood approximately 2 weeks after the last dose (0.66 μM) which may suggest that a sustained level of gold is necessary for worm killing.
Transmission electron micrographs of adult Brugia incubated overnight with 1 μM auranofin showed that there was extensive damage to Wolbachia in the hypodermal area, in contrast to worms treated with 10 μM flubendazole. Flubendazole at this concentration did not cause vacuolization but only minor changes to the mitochondria, which appeared to contain black bodies. Loss of integrity in muscle tissue and the hypodermal chord were also observed when O. ochengi adults were incubated with auranofin at 10 μM for 7 days. Thus, the structural damage caused by auranofin is similar in both species, except that presumably due to the large size of O. ochengi, auranofin takes a much longer time and higher concentrations of drug to have an effect.
Auranofin is an FDA-approved drug that was originally developed to treat rheumatoid arthritis. There is strong evidence in several species of parasites that thioredoxin reductase and a similar enzyme, thioredoxin glutathione reductase (TGR), are targeted by auranofin [26,33,39–41]. Previous studies have shown that this drug is an effective antiparasitic agent against a number of organisms, including Schistosoma mansoni and S. japonicum [33,42], Echinococcus granulosus [43], Taenia crassiceps [44], Plasmodium falciparum [45], Leishmania spp. [46], Trypanosoma brucei [47], Giardia lamblia [39,48] and Toxoplasma gondii [49]. In animal studies, auranofin was highly efficacious in treating amoebic colitis in mice and amoebic liver abscesses in hamsters [26]. Auranofin treatment also significantly decreased worm burdens in mice infected with S. mansoni [33] and suppressed footpad lesion formation and reduced existing lesions in a mouse model of cutaneous leishmaniasis [46].
The thioredoxin system is integral to maintaining a reduced state and managing oxidative stress within the cell, which makes this system critical for organism survival [50]. Thioredoxin, which is reduced by thioredoxin reductase, is a substrate for redox enzymes including peroxidases in filarial worms [18,51]. Inhibition of TrxR by auranofin alters the redox state of the cell leading to an increased production of hydrogen peroxide and oxidation of the components of the thioredoxin system thereby enhancing apoptosis [52]. Sayed et al (2006) found that silencing peroxiredoxins, downstream redox partners of TrxR, in schistosomes led to detectable protein and lipid oxidation [53]. Inhibition of Brugia TrxR by auranofin may disrupt this process in filarial worms, which can then lead to worm death. Interestingly, there were significantly fewer female worms than male worms from gerbils treated with auranofin. The preferential killing of female worms may be due to the host’s immune response against females when they release microfilariae [54,55]. It is also possible that as female worms develop and molt from the larval stage to the adult stage, they elicit an immune response that, together with auranofin, preferentially kills female worms over males.
The mode of action of auranofin is thought to be a specific inhibition of the selenoenzymes thioredoxin reductase (TrxR) and thioredoxin glutathione reductase (TGR). No TGRs from Brugia have been identified thus far. Kuntz et al (2007) showed that auranofin inhibited TGR in adult schistosomes in vitro but had no effect on the activities of another selenoenzyme, glutathione peroxidase, or the abundant enzyme lactate dehydrogenase [33]. Loss of TGR activity preceded parasite death, indicating that specific inhibition of TGR by auranofin was responsible for parasite death in schistosomes.
Auranofin inhibition has also been shown to be less specific to glutathione peroxidase and glutathione reductase, which have about 1000-fold higher IC50s compared to TrxR isolated from human placenta [32]. Other thioenzymes, such as the cysteine protease cathepsin B, also had significantly higher IC50s when tested with auranofin (approximately 250 μM) [56] compared with the IC50 of auranofin with rBmTrxR.
Thioredoxin reductase enzyme activity of B. malayi adult worms treated with auranofin was significantly lower compared to with vehicle-treated worms in the in vitro assays. TrxR activity was also decreased by 49% in worms removed from gerbils 16 days after treatment with auranofin, supporting the hypothesis that auranofin specifically targets TrxR in these worms.
Targeting the thioredoxin system by inhibiting thioredoxin reductase may be a promising strategy for treating filarial infections, since the enzyme appears to be necessary for worm survival. It is possible that auranofin treatment increases the susceptibility of the parasite to oxidative damage, which in turn allows the host’s immune system to eliminate the parasite.
Since auranofin is already an FDA-approved drug, the path to clinical trials is streamlined. Patients with rheumatoid arthritis who were treated with auranofin for an average of 6 months had few side effects, with the most common side effect being diarrhea [20]. In the present study auranofin was shown to be efficacious in the Brugia/gerbil model when given for 28 days. Additional studies will be conducted to determine efficacy with shorter treatment regimens and to obtain pharmacokinetic data. Auranofin will also be evaluated for any synergistic effects with other drugs such as doxycycline and for its use as part of a macrofilaricidal cocktail.
|
10.1371/journal.pntd.0000265 | Acute Schistosoma mansoni Infection Increases Susceptibility to Systemic SHIV Clade C Infection in Rhesus Macaques after Mucosal Virus Exposure | Individuals living in sub-Saharan Africa represent 10% of the world's population but almost 2/3 of all HIV-1/AIDS cases. The disproportionate HIV-1 infection rates in this region may be linked to helminthic parasite infections that affect many individuals in the developing world. However, the hypothesis that parasite infection increases an individual's susceptibility to HIV-1 has never been prospectively tested in a relevant in vivo model.
We measured whether pre-existing infection of rhesus monkeys with a parasitic worm would facilitate systemic infection after mucosal AIDS virus exposure. Two groups of animals, one consisting of normal monkeys and the other harboring Schistosoma mansoni, were challenged intrarectally with decreasing doses of R5-tropic clade C simian-human immunodeficiency virus (SHIV-C). Systemic infection occurred in parasitized monkeys at viral doses that remained sub-infectious in normal hosts. In fact, the 50% animal infectious (AID50) SHIV-C dose was 17-fold lower in parasitized animals compared to controls (P<0.001). Coinfected animals also had significantly higher peak viral RNA loads than controls (P<0.001), as well as increased viral replication in CD4+ central memory cells (P = 0.03).
Our data provide the first direct evidence that acute schistosomiasis significantly increases the risk of de novo AIDS virus acquisition, and the magnitude of the effect suggests that control of helminth infections may be a useful public health intervention to help decrease the spread of HIV-1.
| To test the hypothesis that infection with helmiths may increase host susceptibility to infection with HIV-1, we quantified the amount of a clade C simian-human immunodeficiency virus needed to infect rhesus macaques that had acute Schistosoma mansoni infections. Compared to control animals exposed to virus alone, monkeys with schistosomiasis required exposure to 17-fold lower levels of virus to become infected. The schistosome-infected monkeys also had significantly higher levels of initial virus replication and loss of a certain subset of memory T cells, both predictors of a more rapid progression to immune dysfunction. These results suggest that worm infections may increase the risk of becoming infected with HIV-1 among individuals with viral exposures. Furthermore, they support the idea that control programs for schistosomiasis and perhaps other parasitic worm infections may also be useful in helping to reduce the spread of HIV/AIDS in developing countries where helminths are endemic.
| Sub-Saharan Africa represents only 10% of the world's population but more than 62% of the world's HIV/AIDS cases [1]. While it remains controversial whether HIV transmission and/or disease progression in sub-Saharan Africa differ from what is observed in industrialized countries [2], one factor that may contribute to any exacerbation of HIV/AIDS is the high prevalence of parasitic worm infections, such as schistosomiasis [3],[4]. However, because it is not possible to directly test this hypothesis in humans, studies to date have evaluated in vitro exposure of cells to virus [5],[6], effects of schistosome infection on established viral infection [7],[8], or epidemiologic evaluations of the effect of praziquantel treatment on viral load [9]–[12]. None of these approaches address the issue of whether a helminth infection increases the susceptibility of the host to acquire de novo infection with an immunodeficiency virus after mucosal exposure, the predominant route of HIV transmission in humans.
In order to directly investigate this essential question, we tested the effect of Schistosoma mansoni infection on host susceptibility to immunodeficiency virus infection using schistosome-infected and control macaques exposed intrarectally to successively lower doses of the recently described R5-tropic SHIV-C, SHIV-1157ipd3N4 [13]. This strain is highly relevant for our study because an estimated 90% of all new HIV-1 infections occur by mucosal transmission, which almost exclusively involves R5 strains [14], and clade C strains cause >50% of all HIV-1 infections worldwide [1]. We also assessed the impact of schistosome infection on viral loads and immune cell profiles in monkeys with SHIV infection. These studies represent the first prospective evaluation of the impact of helminth infection on immunodeficiency virus transmission in a relevant model.
Chinese-origin adult female rhesus macaques were housed at the animal facility of the Centers for Disease Control and Prevention (CDC) in Atlanta, GA. Protocols were approved and animals were maintained in accordance with the guidelines of the Institution Animal Care and Use Committees for both CDC and the Dana-Farber Cancer Institute (DFCI). All procedures employed were consistent with the Guide for Care and Use of Laboratory Animals. Animals were free of helminth infection prior to our study.
SHIV-1157ipd3N4 [13], an R5-tropic SHIV-C infectious molecular clone, is a monkey-adapted, late form of SHIV-1157i, which encodes most of the env sequences of a primary HIV-1 clade C strain isolated from a recently infected Zambian infant. SHIV-1157ipd3N4 was engineered with an additional NF-κB site per long terminal repeat (LTR) in order to enhance viral replication. The methodology used to construct this virus as well as its parental biologic strains are described by Song et al. [13]. Both the early and the late forms of our SHIV-C were pathogenic in rhesus monkeys, although disease progression was somewhat slow, with AIDS developing approximately 2.5–5.5 years post-inoculation.
Animals were anesthetized with ketamine and placed in a prone position. SHIV-1157ipd3N4 dilutions were prepared in RPMI in a total volume of 1 ml. The inoculum was loaded into a gastric feeding tube and inserted 5 cm into the rectum with the aid of lubricant. Animals were infused with 1 ml of the virus dilution followed by 3 ml of plain RPMI and 2 ml of air. All animals were exposed to virus dilution within 1 hr after the vial of stock virus was thawed.
Animals were anesthetized with ketamine and percutaneously exposed to 500 cercariae of a Puerto Rican strain of S. mansoni. An area on the abdomen was shaved, and cercariae were placed on the skin within a metal ring for 30 min to allow penetration. To monitor infection, fresh stool was obtained and processed by formalin-ethyl acetate sedimentation and concentration. Schistosome eggs were counted by microscopic examination. White blood cell counts (WBC) and percent eosinophils were calculated using standard methods. Hematology results, as well as CD4 and CD8 T-cell counts and ratios were determined by the Pathology Laboratory at Yerkes National Primate Research Center (Atlanta, GA). There was no evidence of fever, diarrhea, weight loss, or dysentery in these animals.
Peripheral blood samples were obtained by venipuncture and collected into Vacutainer cell preparation tubes containing sodium citrate (Becton Dickinson, Rutherford, N.J.). Immediately after collection, plasma and PBMC were separated and quick frozen. Plasma samples from infected monkeys were stored at −80°C until vRNA loads were assessed by real-time RT-PCR [15]. Cells were washed once in RPMI 1640 (Gibco, Grand Island, NY), lysed with RLT lysis buffer (Qiagen, Valencia, CA) containing 1% β-mercaptoethanol (Sigma Chemical Company, St. Louis, MO), and stored at −80°C until they were processed to measure cytokine mRNA levels. A quantitative real-time RT-PCR assay based on TaqMan chemistry was utilized to measure mRNA levels for the cytokines interleukin IL-2, IL-4, IL-6, IL-10, IFN-γ, and TNF-α and the chemokine RANTES using previously described primers and protocols [16]. mRNA expression was normalized using primers unique for the housekeeping gene, phosphate dehydrogenase (PDH).
The following antibodies were used for both phenotyping and sorting of peripheral blood mononuclear cells (PBMC): CD3-Alexa700, CD4-PERCP-Cy5.5, CD8-APC-Cy7, CCR7-FITC, CD45RA-ECD, CD28-PE, CD95-APC; for cell sorting, we stained PBMC for CD4+ T-cell subsets only, using CD3, CD4, CD28 and CD95 mAbs. All reagents, except for CCR7 (R&D Systems, Minneapolis, MN), were obtained from BD Biosciences (San Jose, CA). For phenotyping and/or cell sorting, PBMC were isolated from blood collected and separated in Na citrate CPT tubes (Becton Dickinson). Cells were washed, counted, and resuspended in phosphate-buffered saline (PBS) containing 2% FCS with the appropriate mix of antibodies. For phenotyping, 2×106 cells were stained, for cell sorting, a minimum of 10×106 cells were evaluated. Following incubation with antibodies for 15 min under reduced light, cells were washed and resuspended in PBS containing 1% paraformaldehyde prior to analysis. Data were collected on a Becton Dickinson cell sorter FACSVantage with DiVa option (BD Biosciences) and analyzed with FlowJo analysis software (Tree Star, Inc., Ashland, OR). In both cases, we used an electronic gate on the forward scatter (FSC-H/FSC-A) to avoid doublets.
Calculation of AID50 values and statistical comparison of infectious doses were performed using the method of Spouge [17]. Peak vRNA loads and single time point T-cell subsets were compared using the Wilcoxon rank-sum test. Statistical comparisons of vRNA loads and T-cell subsets over time between schistosome-infected and parasite-free groups were performed using repeated measures analysis. Modeling was performed with generalized estimating equations [18].
After preparation and storage of multiple vials of a large SHIV-C (strain SHIV-1157ipd3N4) stock, the minimal and AID50 necessary to establish systemic viral infection after mucosal challenge were determined in a group of 9 parasite-free animals (Table 1). Plasma vRNA loads were monitored prospectively by real-time RT-PCR (15). The minimal infectious dose for the parasite-free controls was 1 ml of virus stock diluted 1∶50; only 2 of the 3 animals exposed to this dilution became systemically infected. Based on the statistical method of Spouge [17], we determined that the AID50 of SHIV-C was 0.025 ml (95% confidence interval (CI), 7×10−3 to 8.7×10−2) for these control animals.
To evaluate the impact of helminth infection on the susceptibility to SHIV-C, a second group of 8 macaques was infected with schistosomes by percutaneous exposure to 500 cercariae. Parasite infection was confirmed by measuring signs consistent with acute schistosomiasis: egg excretion in feces, eosinophilia (Figure 1A), and increased PBMC IL-4 mRNA levels (Figure 1B). In contrast, mRNA levels for IL-2, IL-6, IFN-γ, TNF-α, and RANTES did not differ between baseline and week 7 after infection (data not shown). Levels of IL-10 mRNA were increased as a result of schistosome infection at week 7 but were not statistically different. However, by 14 weeks of schistosome infection (7 weeks after exposure to SHIV) the increase in IL-10 mRNA was statistically significant (p = 0.047) compared to baseline. Monkeys were exposed to SHIV-C between 7 and 9 weeks after schistosome infection. The minimal infectious dose in animals with schistosomiasis was 1 ml of SHIV-C diluted 1∶300; all 3 animals exposed to this dose became systemically infected (Table 1). The AID50 for animals with schistosomiasis (0.0014 ml, 95% CI, 4×10−4 to 5×10−3) was 17-fold lower than that for parasite-free animals (P<0.001). The minimal infectious virus dose differed by a factor of 6 between the two groups of monkeys.
The mean peak vRNA load was >1 log higher in animals with acute schistosomiasis than in controls (P<0.001) even though the mean viral inoculum that led to systemic infection in animals with schistosomiasis (1∶185) was lower than that in parasite-free controls (1∶23) (Figure 2A). This is consistent with our previous findings that the magnitude of the peak vRNA load in infected animals depends on host factors and not on the viral concentration of the inoculum [19]. Elevated vRNA loads in parasite-infected animals were maintained through 10 weeks after virus exposure (P<0.001) (Figure 2B), similar to the elevated viral replication we observed in schistosome-infected animals that had been exposed intravenously to equal, high doses of a related SHIV [7].
Although vRNA loads were higher in animals with schistosomiasis, levels of both CD4+ (Figure 2C) and CD8+ (Figure 2D) T cells were also elevated in these animals compared to virus-only controls, consistent with the generalized immune activation caused by schistosomiasis. CD4/CD8 ratios were similar between the two groups and remained steady over time (data not shown).
PBMC from coinfected monkeys and from animals infected with SHIV-C alone were analyzed by multi-color flow cytometry after a month of viral inoculation (Figure 3A). The levels of CD4+ central memory (CM) T cells were significantly higher in coinfected animals, while parasite-free animals had more naïve CD4+ T cells (Figure 3B). We observed a similar pattern in CD8+ naïve and memory T cells but these differences were not statistically significant (data not shown).
Next, we sought to determine the prevalence of vRNA in different CD4+ T-cell subsets. We sorted CD4+ naïve and memory T-cell subsets for each group of animals (Figure 3A) and determined vRNA levels by RT-PCR (Figure 3C). Both naïve and memory CD4+ T cells from coinfected animals contained more vRNA per 106 cells than the corresponding cells of animals infected with SHIV-C alone; the difference was statistically significant only in CD4+ CM subset (CD28+ CD95+) of T cells (P = 0.03) (Figure 3C). Consistent with the higher vRNA levels in CD4+ CM T cells of coinfected animals, 4 out of the 6 of these monkeys also had losses to abnormal levels (<10%) of the CD4+CD29+ subset of memory T-cells in peripheral blood, which is an early sign of immune dysfunction in lentivirus-infected-macaques [13],[20],[21]. In contrast, none of the 5 animals infected with virus alone fell below this threshold (data not shown). Together, these results suggest that CD4+ T cells from coinfected animals are more permissive to viral infection and replication, leading to more rapid destruction of memory T cells.
This study is the first direct evidence showing that helminth infection significantly increases host susceptibility to mucosal AIDS virus transmission in primates. Systemic infection of S. mansoni-infected rhesus macaques was established by low doses of virus that remained sub-infectious in parasite-free hosts, and the difference in viral dose needed to successfully establish systemic infection was surprisingly large. Furthermore, peak vRNA levels were also significantly elevated in schistosome-infected monkeys, consistent with previous in vitro, epidemiologic and primate model studies that suggest increased viral replication in schistosome-infected hosts or cells from persons with helminth infections [5]–[9]. However, none of the earlier studies was able to address whether schistosome infection increased the likelihood of de novo immunodeficiency virus infection in hosts harboring helminths. The increased susceptibility to mucosal immunodeficiency virus transmission we observed in schistosome-infected macaques, elevated viral replication in coinfected hosts, and accelerated loss of memory T cells all have profound public health implications for areas of the world where both parasitic worms and HIV-1 are endemic. Our data support the suggestion that better control of helminth infections may favorably impact efforts to reduce the spread of HIV/AIDS [3],[4].
Our observations are consistent with the hypothesis that helminth infections increase a hosts' susceptibility to infection with immunodeficiency viruses and is associated with a Th2-type immunologic phenotype and increased viral replication within these cells [22]. However, it is possible that any infection resulting in systemic immune activation or increased peripheral blood CD4+ T cell numbers (Figure 2C) would have yielded similar results [23]. Other data to support the hypothesis that Th2-type responses increase susceptibility to virus infection include the observation that PBMC from Kenyan schistosomiasis patients with HIV-1 coinfection produce decreased levels of IL-4 and IL-10 compared to patients with schistosomiasis alone. The magnitude of this effect correlates with the decrease in CD4+ T cells [24], suggesting that it is the CD4+ Th2 cells that are preferentially infected and killed by the virus. Alternatively, the shift to a predominant Th2-type response in schistosome-infected hosts may result in down-regulation of virus-controlling cytotoxic T lymphocytes [25]. We will evaluate these possibilities more directly in future studies.
We also considered whether the biology of parasite egg excretion, combined with the rectal route of viral exposure, may have increased host susceptibility to virus. Fecal egg excretion in hosts infected with S. mansoni or S. japonicum increases the number of activated T cells associated with egg granulomas that are in close juxtaposition to the intestinal lining [26]. Furthermore, the passage of eggs from mucosal tissue into the gut lumen may compromise the integrity of the epithelial lining, which in turn could lead to microbial translocation, release of immune-activating bacterial products into the bloodstream, and facilitate intrarectal SHIV-C transmission.
A limitation of our primate system is that we are only able to model the effect of the acute phase of schistosomiasis on susceptibility to viral transmission. This is because rhesus monkeys, while able to develop a fully patent infection, self cure their schistosomiasis at 20 to 25 weeks after exposure to cercariae [7],[8],[27]. Thus, we were not able to assess whether animals exposed to immunodeficiency virus during the chronic, more immunologically regulated phase of schistosome infection similarly display increased susceptibility to viral transmission. If the increased susceptibility to viral infection is related to the shift towards a Th2-type immune response, the increase in susceptibility to virus may be more modest during chronic infections when the phenotypic shift is less dramatic. Nevertheless, in human hosts with chronic schistosomiasis, both the immunologic stimulation and Th2-type responses (e.g., eosinophilia) persist. In addition, surface levels of the chemokine coreceptors CCR5 and CXCR4 are elevated on CD4+ T cells and monocytes of persons with chronic schistosomiasis and decrease following praziquantel treatment [28].
Our data strengthen the hypothesis that helminth infection may be a risk factor for increased susceptibility to de novo HIV-1 infection and support control of schistosomiasis and perhaps other helminths in persons living in areas endemic for these parasites. In the absence of an effective AIDS vaccine, there are several other strategies that may help decrease the risk of HIV-1 transmission, including control of sexually transmitted diseases [29] and male circumcision [30],[31]. Treatment of helminth infections is inexpensive, safe, and easily administered to large populations. In addition to the benefit of reducing host morbidity caused by the parasites, our data support control of helminths as a public health intervention for individuals at risk for acquiring HIV-1.
|
10.1371/journal.pcbi.1003411 | Intrinsic Noise Induces Critical Behavior in Leaky Markovian Networks Leading to Avalanching | The role intrinsic statistical fluctuations play in creating avalanches – patterns of complex bursting activity with scale-free properties – is examined in leaky Markovian networks. Using this broad class of models, we develop a probabilistic approach that employs a potential energy landscape perspective coupled with a macroscopic description based on statistical thermodynamics. We identify six important thermodynamic quantities essential for characterizing system behavior as a function of network size: the internal potential energy, entropy, free potential energy, internal pressure, pressure, and bulk modulus. In agreement with classical phase transitions, these quantities evolve smoothly as a function of the network size until a critical value is reached. At that value, a discontinuity in pressure is observed that leads to a spike in the bulk modulus demarcating loss of thermodynamic robustness. We attribute this novel result to a reallocation of the ground states (global minima) of the system's stationary potential energy landscape caused by a noise-induced deformation of its topographic surface. Further analysis demonstrates that appreciable levels of intrinsic noise can cause avalanching, a complex mode of operation that dominates system dynamics at near-critical or subcritical network sizes. Illustrative examples are provided using an epidemiological model of bacterial infection, where avalanching has not been characterized before, and a previously studied model of computational neuroscience, where avalanching was erroneously attributed to specific neural architectures. The general methods developed here can be used to study the emergence of avalanching (and other complex phenomena) in many biological, physical and man-made interaction networks.
| Networks of noisy interacting components arise in diverse scientific disciplines. Here, we develop a mathematical framework to study the underlying causes of a bursting phenomenon in network activity known as avalanching. As prototypical examples, we study a model of disease spreading in a population of individuals and a model of brain activity in a neural network. Although avalanching is well-documented in neural networks, thought to be crucial for learning, information processing, and memory, it has not been studied before in disease spreading. We employ tools originally used to analyze thermodynamic systems to argue that randomness in the actions of individual network components plays a fundamental role in avalanche formation. We show that avalanching is a spontaneous behavior, brought about by a phenomenon reminiscent to a phase transition in statistical mechanics, caused by increasing randomness as the network size decreases. Our work demonstrates that a previously suggested balanced feed-forward network structure is not necessary for neuronal avalanching. Instead, we attribute avalanching to a reallocation of the global minima of the network's stationary potential energy landscape, caused by a noise-induced deformation of its topographic surface.
| An important problem in many scientific disciplines is understanding how extrinsic and intrinsic factors enable a complex physical system to exhibit a bursting behavior that leads to avalanching [1], [2]. Avalanching is a form of spontaneous behavior characterized by irregular and isolated bursts of activity that follow a scale-free distribution typical to systems near criticality. In the brain, this mode of operation is thought to play a crucial role in information processing, memory, and learning [2]–[6].
Although avalanche dynamics have been extensively studied in vitro [7] and in vivo [8], [9] for cortical neural networks, it is not clear what causes avalanching. A recent in silico attempt to address this issue [10] was based on approximating the dynamics of a Markovian model of nonlinear interactions between noisy excitatory and inhibitory neurons by Gaussian fluctuations around the macroscopic system behavior using the linear noise approximation (LNA) method of van Kampen [11]. This led to the conclusion that the cause of neural avalanches is a balanced feed-forward (BFF) network structure. We argue here that the Gaussian approximation used to arrive at this conclusion is not appropriate for studying avalanching, thus leading to deficient results. As a consequence, understanding the underlying causes of avalanching in silico is still an open problem.
To address this challenge, we introduce a theoretical framework that allows us to examine the role of intrinsic noise in inducing critical behavior that leads to avalanching. Although the idea that noise may induce avalanching has been proposed more that a decade ago [12], our framework leads to a novel understanding of the underlying causes of avalanching in a particular class of complex networks. We focus on a general Markovian network model, which we term leaky Markovian network (LMN), with binary-valued state dynamics. These dynamics are described by a time-dependent probability distribution that evolves according to a well-defined master equation [13] (see Methods for details). It turns out that a LMN is a continuous-time stochastic Boolean network model with a state-dependent asynchronous node updating scheme (we provide details in Text S1). LMNs can model a number of natural and man-made systems of interacting species, such as genetic, neural, epidemiological, and social networks.
Recent work has clearly demonstrated the importance of stochastically modeling physical systems using Markovian networks. The main reason is that intrinsic noise produced by these networks may induce behavior not accounted for by deterministic models [14]–[17]. Examples of such behavior include the emergence of noise-induced modes, stochastic transitions between different operational states, and “stabilization” of existing modes.
In this paper, we study the effect of intrinsic noise on avalanching by using a LMN model. We do so by employing the notion of potential energy landscape [13], [18], [19] and by establishing a connection between statistical thermodynamics and the kinetics of bursting. We quantify the landscape by calculating logarithms of the ratios between the stationary probabilities of individual states and the stationary probability of the most probable state. To reduce computational complexity, we follow a coarse graining approach that transforms the original LMN model into another (non-binary) LMN model with appreciably smaller state-space. To accomplish this task, we partition the nodes of the LMN into homogeneous subpopulations and characterize system behavior by using the dynamic evolution of the fractional activity process, which quantifies the fraction of active nodes (nodes with value 1) in each subpopulation. Moreover, we parameterize the LMN in terms of the network size , where is the net number of nodes in the network and is a normalizing constant such that can be approximately considered to be continuous-valued. We refer the reader to the Methods section for details.
The behavior of the fractional activity process is fundamentally affected by . In general, the strength of stochastic fluctuations (intrinsic noise) in the activity process may be thought of as the probability of moving uphill on a fixed potential energy surface, which decays exponentially with increasing . At sufficiently large network sizes , the LMN operates around a ground state of the potential surface located at a fixed point predicted by the macroscopic equations associated with the LNA method, which we assume to be unique, nonzero, and stable (see Methods and Text S1 for details). In this case, a new mode of operation is introduced in the system, as the network size decreases, in the form of a potential well in the topographic surface of the energy landscape, located at the inactive state . This is a “noise-induced” mode, since it appears at small network sizes at which the fractional activity process is subject to appreciable intrinsic fluctuations.
We show in this paper that noise-induced deformation of the stationary potential energy landscape is the underlying cause of avalanching in LMNs. For sufficiently large network sizes, the potential energy landscape can be approximated by a quadratic surface centered at . In this case, the LMN operates within the potential well associated with this mode, except for rare and brief random excursions away from that mode. As a consequence, the fractional activity process will fluctuate in a Gaussian-like manner around the macroscopic mode. At smaller network sizes, the fractional activity process is characterized by a bistable behavior between the macroscopic and noise-induced modes, spending most time within the potential well associated with the macroscopic mode, at which the potential energy surface attains its global minimum, while occasionally jumping inside the potential well associated with the noise-induced mode at . As a consequence, the fractional activity dynamics take on a bursting behavior characterized by long periods of appreciable activity followed by short periods of minimal (almost zero) activity. When the network size decreases further, the noise-induced mode becomes the main stable operating point (i.e., the point at which the potential energy surface attains its global minimum), whereas the macroscopic mode becomes shallower and eventually disappears. In this case, the system is trapped within the potential well associated with the noise-induced mode, except for random and brief excursions away from that mode. As a consequence, the fractional activity process will still exhibit bursting, but now characterized by long periods of minimal (almost zero) activity followed by short bursts of appreciable activity.
Thermodynamic analysis reveals critical behavior in LMNs (we provide details in the Methods section and Text S1). By employing a number of statistical thermodynamic quantities, such as internal and free potential energies, entropy, internal pressure, pressure and bulk modulus (inverse compressibility), we effectively summarize the stochastic behavior of a LMN as its size decreases to zero. We also use these summaries to quantify network robustness and the stability of a given state. In agreement with the classical theory of phase transitions, the previous thermodynamic quantities evolve smoothly as a function of until a critical network size is reached. At this size, a discontinuity is observed in the system pressure, which produces a spike in the bulk modulus demarcating loss of thermodynamic robustness. Critical behavior is caused by reallocation of the ground states (global minima) of the potential energy landscape due to noise-induced deformation of its topographic surface. In particular, observed critical behavior produces two distinct phases: one in which the fixed point predicted by the macroscopic equations associated with the LNA method constitutes the ground state of the potential energy landscape and one in which the ground state is reallocated to the noise-induced mode at . We conclude that avalanching is a complex mode of operation that dominates system dynamics at near-critical and subcritical network sizes due to deformations of the potential energy landscape as the network size decreases to zero, caused by appreciable levels of intrinsic noise.
It is important to mention here that our work provides a novel stochastic perspective to the well-known phenomenon of self-organized criticality (SOC) [20]; i.e., the spontaneous emergence of critical behavior without tuning system parameters whose values are influenced by external factors. We approach SOC from the perspective of nonlinear Markovian dynamics and directly associate self-organization properties of a complex network with the existence of a unique probability distribution at steady-state, which leads to a unique stationary potential energy landscape. Our work demonstrates that SOC may emerge as a consequence of two interweaved adaptive processes that may take place on separate timescales [21], [22]: a short timescale convergence to dynamic equilibrium (stationarity), during which the topological structure of a neural network subsystem is kept relatively fixed (is quasistatic), and longer timescale alterations in topological properties that lead to changes in the size of the network. For example, the neural activities modeled by our LMNs occur on a short timescale compared to the timescale of neural development, where it is well known that programmed cell death plays a major role [23]. This large reduction in the number of neurons could presumably serve to bring neural subsystems within proximity of their critical sizes in accordance with their underlying connectivity structures. On the other hand, an adaptive interplay between the short timescale neural dynamics and the quasistatic topological network structure may lead to a longer timescale topological self-organization that enlarges or contracts the neural subsystem with an objective to keep the system robustly close to criticality [22]. We show that such long timescale alterations can result in a spontaneous reallocation of the ground states in the stationary potential energy landscape due to a noise-induced deformation of its topographic surface. Our approach associates SOC with observable stochastic multistability, which is directly related to the phenomenon of phase transition, thus bridging the gap between “self-organized” and “classical” criticality (see also [24]).
Finally, we would like to point out that, in the neural network literature, it is commonly said that avalanching occurs in the supercritical regime near the critical point. On the other hand, we show in this paper that avalanching occurs in a subcritical regime near the critical point. To avoid confusion, the reader must keep in mind that these statements do not contradict each other. Correctly using the terms “subcritical” and “supercritical” depends on the parameter employed to take a system from one regime to the other. Contrary to existing works that study criticality in terms of functional parameters (e.g., in terms of the firing rate of all neurons in a neural network), we study in this paper criticality in terms of a structural parameter, the system size, which is inversely related to the strength of intrinsic noise. In this case, avalanching occurs at network sizes smaller but near a critical value, which forces us to use the aforementioned terminology.
We consider a directed weighted network with nodes from a set , characterized by an adjacency matrix . The element of this matrix assigns a value to the edge leaving the -th node and entering the -th node whose importance will become clear shortly. Each node represents a species (e.g., an individual or neuron) which, in some well-defined sense, can be active or inactive at time with some probability. We use to denote the state of the -th node of the network at time , taking value 1 if the node is active and 0 if the node is inactive. Then, we represent the state dynamics of the network by an -dimensional random process whose -th element takes binary 0–1 values. We refer to as the activity process.
We assume that, within an infinitesimally small time interval , the state of the -th node is influenced by the net input to the node, where is the state of the network at time and is a real-valued scalar function. In particular, we assume that the probability of the -th node to transition from the inactive to the active state within is proportional to , given by , where is known as the propensity function and is a term that goes to zero faster than . We set(1)for some nonnegative parameter and a nonnegative function . The term ensures that transition to the active state is possible only when the -th node is inactive (i.e., when ), whereas the function describes how the net input affects the probability of transition. On the other hand, when , the parameter forces the node to be “leaky,” in the sense that it has a fixed propensity to transition from the inactive to the active state, even when the net input is zero. “Leakiness” is a property observed in many applications, including the ones discussed in this paper.
We also assume that the probability of the -th node to transition from the active to the inactive state within is given by , where the propensity function is given by(2)for some nonnegative parameter and a nonnegative function . The term ensures that transition to the inactive state is possible only when the -th node is active (i.e., when ), whereas the function describes how the net input affects the probability of transition. On the other hand, when , the parameter forces the node to be “leaky,” in the sense that it has a fixed propensity to transition from the active to the inactive state even when the net input is zero.
In general, the weights are used to determine the net input to node . As a matter of fact, must not depend on when . In particular, we set , for every , which implies that the nodes are not self-regulating. In some applications (such as the ones considered in this paper), we can set(3)where is the -th row of the adjacency matrix and is a constant. In this case, may represent the influence of external sources on the node (which we assume for simplicity to be fixed and known), whereas represents the influence of all active nodes in the network on the state of the -th node.
is a Markov process. By assuming that all nodes in the network are initially inactive at time , we can show that the probability distribution satisfies the master equation(4)for , initialized with the Kronecker delta function [i.e., ], where is the -th column of the identity matrix. The model described by this equation is a continuous Boolean network model with state-dependent asynchronous node updating (more details can be found in Text S1). It can also be viewed as a special case of an interacting particle system (IPS) model [25] and it thus follows IPS-like dynamics. Unfortunately, solving this equation is a notoriously difficult task, especially when the number of nodes in the network is large. This is due to the fact that we need to calculate the probabilities , for , at every point in the state space , whose cardinality grows exponentially as a function of , since .
We address the previous problem by employing a “coarse graining” procedure, similar to that suggested in [26] (see also [27]), which allows us to appreciably reduce the size of the state space while retaining key properties of the system under consideration. We assume that we can partition the population of all species in the network into homogenous sub-populations , , where . Due to the homogeneity of each sub-population, it may not be of particular interest to track the states of individual species in a given sub-population . Instead, it may be sufficient to track the fraction of active species in , defined by(5)where . In this case, we may replace the original network with a smaller directed weighted network comprised of nodes from the set that represent the homogeneous sub-populations. We assume that, for every , there exists a function such that , for all , where is a vector whose -th element is given by . In this case, the stochastic process is also Markovian, governed by the following master equation (details in Text S1)(6)initialized with the Kronecker delta function [i.e., ], where and is the -th column of the identity matrix multiplied by . The new propensity functions are given by(7)(8)where , , , and are such that, for every , and , for . We refer to as the fractional activity process.
When the input to a node of the network is given by , there is indeed a function so that , for every . This function is given by , where and are such that , for every , and , for every , (details in Text S1). Note also that takes values in . Therefore, the fractional activity process takes values in . As a result, the state-space will be appreciably smaller than , since , and solving the master equation of the fractional activity process will be easier than solving the master equation of the activity process.
We define the fractional “size” of the -th sub-population as . The thermodynamic limit is obtained by taking , for every , such that all 's remain fixed. In this case, becomes a continuous random variable in the -dimensional closed unit hypercube . Furthermore, since , for every , one might expect that the intrinsic noise at each node of the coarse network will be averaged out due to coarse graining. As a matter of fact, it can be shown that converges in distribution to the deterministic solution of the macroscopic differential equations(9)for , initialized by . For simplicity of notation, we denote the thermodynamic limit by .
If the macroscopic equations have a unique and stable fixed point in the interior of the unit hypercube , then for large enough but finite , the linear noise approximation (LNA) method of van Kampen [11] allows us to approximate the fractional activity process by adding correlated Gaussian noise to the macroscopic solution . In this case,(10)for , where, for each , are zero-mean correlated Gaussian random variables with correlations that satisfy a system of Lyapunov equations (details in Text S1). As a consequence, is approximated by a multivariate Gaussian random vector with mean and covariance matrix , where is a diagonal matrix with elements , and is the correlation matrix of random vector .
We consider the probability distribution of the fractional activity process at time , where we explicitly denote the dependence of this distribution on the network size . Let be a state in at which attains its (global) maximum value at time and define the function(11)Note that with equality if and only if is a state at which attains its (global) maximum value, known as a ground state. Moreover,(12)for , , where . In this case, is a Boltzmann-Gibbs distribution with potential energy function and partition function . The (local or global) minima of are associated with “potential wells” (basins of attraction) in the energy surface, which correspond to peaks in the probability distribution . We may therefore view the fractional activity dynamics as fluctuations on a time-evolving potential energy landscape in the multidimensional state-space , where downhill motions (towards the bottom of a potential well) are preferred with high probability, but random uphill motions can also occur with increasing probability as the population size decreases.
We are interested in the stationary potential energy , since its landscape remains fixed once the stochastic dynamics reach this point. At steady-state, the fractional activity dynamics simply perform a random walk on . To compute and , we solve the master equation of the fractional activity process numerically.
We characterize the stability, robustness, and critical properties of a LMN by studying its behavior when nodes are removed from the network. To do so, we introduce a number of quantities from classical thermodynamics which are used to describe the behavior of a physical system as its volume contracts or expands (see Text S1 for details). We focus our interest here at steady-state (irreducibility properties of the LMN model are discussed in Text S1). We define the internal potential energy by , where denotes expectation with respect to the stationary probability distribution . Moreover, we define the free potential energy by , where is the entropy of the network, given by . In thermodynamic terms, the free potential energy measures the portion of the energy, not accounted for by the energy of the most likely state, available in the LMN to do work at fixed size.
We also define the internal pressure , pressure , and bulk modulus . The internal pressure quantifies the rate of change in internal potential energy with respect to a change in the number of nodes, whereas the pressure quantifies the rate of change in free potential energy. Finally, the bulk modulus measures the network's resistance to changing pressure.
It turns out that , where is the self-information of state that quantifies the amount of information associated with the occurrence of state at steady-state. Therefore, and from an information-theoretic perspective, the internal potential energy of a LMN measures how far the self-information of the most likely state at steady-state is from the expected self-information of all network states (which is the entropy). Note that zero internal potential energy implies zero self-information for the most likely state. In this case, the LMN will be at the most likely state with probability one. As a consequence, we may consider the internal potential energy as a thermodynamic measure of the “stability” of a particular ground state of the potential energy landscape with smaller values indicating increasing stability of that state.
We can use the pressure as a measure of (thermodynamic) robustness of a LMN with respect to the network size . We say that a LMN is robust against variations in network size if there is no appreciable change in pressure when adding or removing nodes. Therefore, a LMN is robust if the derivative of the pressure or the bulk modulus is small (especially at small network sizes). This implies that a robust network must significantly resist changes in pressure.
Finally, we can use the bulk modulus to detect network sizes at which a LMN exhibits critical behavior. As a matter of fact, it is well-known that an intensive thermodynamic quantity, such as the pressure, may experience a sharp discontinuity when another thermodynamic variable, such as the network size, varies past a critical value. If the pressure of a LMN experiences such a discontinuity as the size varies past a critical value , then will effectively capture this discontinuity by a pulse at , thus indicating that the network experiences phase transition at .
We explored our methods by considering two examples: a stochastic version of a one-dimensional SISa model of Methicillin resistant staphylococcus aureus (MRSA) infection [28], [29] with one homogenous population of individuals, and a two-dimensional stochastic neural network (NN) model with two homogeneous populations of an equal number of excitatory and inhibitory neurons [10] (details in Text S1). With the first example, we were able to demonstrate for the first time that avalanching can also occur in epidemiology, even when simple models are used. In the second example, we show that the LNA method is not an appropriate tool for explaining the emergence of bursting and avalanching in neural network models. Despite a difference in dimensionality and their functional form, the two examples produce surprisingly similar results. The code, written in , used to produce the results can be freely downloaded from www.cis.jhu.edu/~goutsias/CSS%20lab/software.html.
For each model, we computed the probability distribution and the potential energy landscape , defined by Eq. (11), parameterized by the network size , where is a state at which attains its (global) maximum. Figure 1 depicts four computed thermodynamic quantities as a function of , whereas, the supporting Figures S1, S2, S3, S4 depict movies of the dynamic evolutions of the stationary potential energy landscapes and probability distributions with respect to decreasing . The results for the two models are qualitatively identical, despite the fact that the dimensionality of their state spaces are different. Note that the internal and free potential energy plots exhibit a deflection point at network size (population size ), for the SISa model, and at (), for the NN model, revealing critical behavior. This is also evident from the pressure, which experiences a discontinuity at and produces a spike in the bulk modulus. On the other hand, the values of the bulk modulus are very close to zero at all other network sizes. For this reason, we can conclude that both models are robust with respect to network size (and hence to variations in the strength of intrinsic noise) away from the critical value .
What is the underlying cause of this critical behavior? The previous results suggest that the slope of the self-information support curve will experience a discontinuity at the critical size and a large curvature at that size. This is a consequence of the fact that equals the pressure (see Text S1). Hence, loss of network robustness near indicates that there is a change in the ground state (global minimum) of the stationary potential energy landscape at . The movies depicted in Figures S1 & S3 corroborate the validity of this point. In particular, Figure 2 confirms that critical behavior in the SISa model is caused by the ground state of the potential energy landscape changing from the fixed point 0.4719 of the macroscopic equation [Eq. (S83) in Text S1] to the origin 0 of the state-space as the network size decreases past the critical value . Likewise, critical behavior in the MM model is caused by the ground state changing from to at .
Figure S2 demonstrates that, for large , the LNA method provides a reasonable approximation to the stationary probability distribution of the SISa model. This observation however becomes questionable upon closer examination of the potential energy landscape dynamics depicted in Figure S1. Although the LNA potential energy landscape approximates well the true energy landscape over an appreciable region around the macroscopic ground state , which accounts for about 99.8% of probability mass, there are substantial differences at the left and right tails of the landscape. These tails characterize rare large deviations from the macroscopic ground state and do not conform to the parabolic shape predicted by LNA. As a matter of fact, state values smaller than reside over a lower and flatter landscape than the one predicted by LNA, whereas state values larger than reside over a higher and steeper landscape. As a consequence, if the fractional activity process moves to a state at the left end tail of the potential energy landscape (i.e., close to the inactive state ), it may stay there for an appreciable amount of time before returning back to the macroscopic ground state. On the other hand, if the fractional activity process moves to a state at the right end tail (i.e., close to the active state ), it may quickly return back to the macroscopic ground state. This behavior, which is not well-predicted by LNA, is corroborated by Figure 3, which shows the actual stationary potential energy landscape as a function of when (), the potential energy landscape predicted by the LNA method, and the inverse mean escape time from a state [computed from Eqs. (S73), (S81) & (S82) in Text S1]. Similar remarks hold for the NN model.
As the network size decreases towards the critical value , the approximation produced by the LNA method begins to break down, due to the emergence of a second well in the potential energy landscape located at the inactive state (see the movies depicted in Figures S1 & S3). This potential well becomes increasingly dominant, as compared to the well located at . Since the ground state of the potential energy landscape transitions from to at the critical size and remains at for all , we expect the inactive state to be the most stable state at subcritical network sizes.
The internal potential energy remains fixed at supercritical network sizes; see Figure 1. This is predicted by Eq. (S61) in Text S1 and the fact that the LNA method provides a good approximation to the solution of the master equation at supercritical sizes (note that for the SISa model and for the NN model). At subcritical sizes, the internal potential energy monotonically increases initially to a maximum value at some network size ( for the SISa model and for the NN model) and subsequently monotonically decreases to zero. As a consequence, the internal pressure (which is the derivative of the internal potential energy with respect to ) is negative for and positive for . Positive internal pressure (decreasing internal potential energy) signifies the fact that removing nodes (individuals or neurons) from the network results in decreasing the distance between the self-information of the most likely state (i.e., the amount of information associated with the occurrence of the inactive state) from the average self-information of all states and thus increasing the stability of this state (details can be found in Text S1). As a consequence, and for network sizes below , increasing levels of intrinsic noise result in increasing the stability of the inactive state
To investigate the emergence of bursting in the SISa model, we depict in the first column of Figure 4 realizations of the fractional activity process (red lines) and the macroscopic dynamics (blue lines), superimposed over the potential energy landscape, for three network sizes, namely () in A, () in C, and () in E. Moreover, we depict in the second column of Figure 4 the corresponding stationary potential energy landscapes.
The stationary energy landscape depicted in Figure 4B exhibits two potential wells. A shallow and narrow well located at and a relatively deep and wide well located at the stable fixed point of the macroscopic equation. Transitions from into are dubious, since such transitions require appreciable stochastic deviations, which are not likely to take place. On the other hand, transitions from to are easier, requiring smaller stochastic fluctuations (mean escape time from 0 is 200 days). In this case, the fraction of infected individuals will fluctuate in a Gaussian-like manner around , although it may sometimes become zero for a relatively short period of time; see Figure 4A.
Figures 4D & 4F indicate that, as the network size decreases, the first potential well becomes deeper and wider, whereas the second well becomes shallower and eventually disappears; see also the movie depicted in Figure S1. When , the two potential wells achieve the same depth. In this case, the fractional activity processes may remain inside longer than before, since transitions from into become more difficult (mean escape time from 0 is now 286 days). As a consequence, the fraction of infected individuals will fluctuate in a Gaussian-like manner around as before, although it may now become zero for a longer period of time; see Figure 4C.
On the other hand, Figure 4F indicates that, when , the potential well becomes extremely shallow. In this case, the fractional activity process will spend most time within with infrequent and very short excursions outside this well (mean escape time from 0 is 500 days). As a consequence, the fraction of infected individuals will mostly be zero with occasional and brief switching to nonzero values. This bursting behavior is clear from figure 5E and is expected in the SISa model since, in a hospital setting or in a swine herd, one often speaks of unpredictable “outbreaks” of an infection, such as MRSA. The deterministic SISa model is fundamentally incapable of predicting such complex behavior. Similar remarks apply for the NN model; see the supporting Figure S5.
Because bursting occurs primarily at steady-state (see Figure 4), we computed avalanche statistics from a single trajectory of the fractional activity process obtained from a long sample of this process. This helped us reduce the computational effort required when calculating avalanche statistics from multiple runs. We simulated the SISa model using the Gillespie algorithm for a period of 300,000 years and used an avalanching threshold to compute the presence of an avalanche (details in Text S1). This allowed us to characterize the SISa model as being active if at least 1 out of 100 individuals was infected. In Figure 5A, we depict log-log plots of the estimated probability distributions of the fractional avalanche size, for sizes between 0.01 and 10 and for various choices of . We also depict the rate of avalanche formation for each case, calculated as the number of avalanches that occurred per day. For the three subcritical network sizes below 0.175, the distributions exhibit scale-free behavior (i.e., the log-log plots are linear) for fractional avalanche sizes below 1 (i.e., when the number of infections that occur during an avalanche is at most ). This observation is clearly corroborated by the results depicted in Figure 5B. On the other hand, for the three supercritical network sizes above 0.175, we observe increasingly lower rates of avalanching, indicating that avalanche formation becomes eventually a rare event as increases. Moreover, this is accompanied with a loss of the scale-free behavior of the size distribution, as it is clearly indicated by Figure 5B. We obtained similar results for the NN model; see Figures 6A & 6B.
It has been argued in the literature that the empirical observation of a power law in a statistical distribution is not sufficient to establish criticality [30], [31]. This has been particularly scrutinized in the case of neural networks and specifically in studying bursting [32]–[34]. Although our analysis clearly shows that a LMN model will experience critical behavior as its size decreases from the supercritical to the subcritical regime, quantified by an observed abrupt behavior in the value of the bulk modulus, an important question that arises at this point is whether this prediction concurs with real experimental data. By employing a method known as avalanche shape collapse [7], [34], we can demonstrate that our thermodynamic predictions lead to near-critical avalanching behavior that can be confirmed by real experimental data.
Avalanche shape collapse is used to demonstrate whether the power laws observed in neural bursting is the result of a mechanism operating near criticality. The basic idea behind this method is that the distributions of some variables of interest in a neural system near criticality are characterized by power laws. Given avalanche data, we can organize avalanches into groups of the same duration . We can then draw a log-log plot of the average size of observed avalanches of duration as a function of and expect that a linear trend will emerge with slope [i.e., we expect that the distribution of will be proportional to ]. Note that, in order to compare our results with the ones obtained in [7], we employ here the definitions for avalanche duration and size obtained by partitioning time into bins of equal duration ; see Text S1. Within each avalanche group of size , we can compute the avalanche shape by plotting the average number of neurons firing within the time interval as a function of , for . Note that right before and right after an avalanche there are no neurons firing. Therefore, the avalanche shape describes the average pattern of non-zero neuron firings between these two lulls in activity.
If observed avalanche cascades are truly critical, then we expect that the computed avalanche shapes will be self-similar, in which case [7](13)where is a scaling function that does not depend on . In this case,(14)By testing whether available experimental data confirm these conditions, we can increase our confidence that the data come from a neural network which operates near criticality. We can accomplish this by shape collapse: we multiply each avalanche shape with and rescale time to be between 0 and 1 by dividing with .
Avalanche shape collapse has been recently applied on real experimental data and has confirmed Eq. (14) – see Figure 3 in [7]. To investigate whether our LMN neural network model concurs with these results, we have simulated our LMN neural network model with neurons () for a period of 80,000,000 ms ( hours). We then implemented the shape collapse method by partitioning time into 20,000,000 bins of ms each. The results depicted in Figure 7 agree with the corresponding results depicted in Figures 2 & 3 of [7]. Our model captures well the inverted parabolic nature of avalanche shapes observed in real experimental data and our simulation results confirm that the dynamics of our model undergo a shape collapse similar to that observed in [7]. As a consequence, our in silico experiment provides a conclusive argument that our model operates near criticality, confirming the validity of our thermodynamic analysis and further supporting the applicability of the proposed modeling and analysis approach to avalanching.
The previously discussed results for the SISa model are based on setting . This parameter quantifies the influence of extrinsic factors (other than direct transmission from other infected individuals) on the rate of infection. We therefore investigated the effect of on bursting.
The mean escape time from the inactive state depends inversely proportional on the network size and parameter . This implies that, for a fixed value of , the stability of the inactive state increases for decreasing , in agreement with our previous discussion.
For fixed , , as , and moving away from the inactive state becomes increasingly difficult. When , the SISa model reduces to the standard SIS model of epidemiology, which enjoys far simpler dynamics: infections will always die out and never appear again, since . As a consequence, the stationary probability distribution of the SIS model assigns all probability mass to the inactive state. On the other hand, , as , which implies that, for sufficiently large , the SISa model will be moving away from the inactive state almost instantaneously.
Figure 8A depicts a plot of the critical network size as a function of , for . Clearly, decreasing increases the value of . In particular, , as . As a consequence, and for sufficiently small values of , , which implies that no avalanching is expected since the network will be operating in the subcritical regime far away from the critical point. This can also be explained by considering the fact that, for very small values of , infection from sources other than infected individuals will be so small that the state of zero infective individuals will have such high probability that excursions from the origin (and thus avalanches) will be rare events.
Similarly, increasing decreases . In particular, , as . This implies that, for sufficiently large values of , , which implies that no avalanching is expected since the network will be operating in the supercritical regime far away from the critical point. This can also be explained by considering the fact that, for very large values of , infection from sources other than infected individuals will be so prevalent that the state of zero infective individuals will have such low probability that excursions from the origin (and thus avalanches) will be rare events.
For the values of encountered in practice, we may have that . This implies that avalanching will be prevalent since the network will be operating in the subcritical regime close to the critical point. Similar results have been obtained for the NN model (data not shown).
In a previous work [10], analysis of the NN model using the LNA method led to the conclusion that for a NN to exhibit bursting it is required that , where , are two appropriately defined parameters (details about these parameters can be found in Text S1). When , the neural network is balanced, in the sense that excitation is very close to inhibition. Moreover, it has been shown that, when the LNA method is valid, fluctuations in the average difference of the fractional activity processes of the excitatory and inhibitory neurons feed-forward into the evolution of the average sum . It was then argued that a balanced feed-forward (BFF) structure is necessary for avalanching in relatively large NNs and that this is achieved through amplification of low levels of intrinsic noise.
Our thermodynamic analysis demonstrates that bursting could be a noise-induced phenomenon that cannot be characterized by the LNA method. This is due to the fact that, at supercritical network sizes, the LNA method may not sufficiently approximate the potential energy landscape in a neighborhood of the inactive state, whereas the method breaks down completely at subcritical network sizes. As a matter of fact, the supporting Figure S6 shows that LNA produces a poor approximation to the potential energy landscape close to for the model considered in [10]. This is not surprising, since the LNA method always results in a negligible probability for the activity process to reach the inactive state [35], which is a predicament that fundamentally contradicts the very basic and experimentally observed nature of avalanching.
It turns out that the BFF condition is not necessary for bursting in NNs. Instead, we have argued that bursting can be attributed to the gradual formation of the noise-induced mode at with decreasing network size.
To further confirm this point, note that BFF behavior is controlled by when is held fixed [10]. With , the analysis in [10] implies that the NN model will exhibit bursting only when . However, our results show that this is also true when ; see Figure 6. In Figure 8B, we depict the computed critical system size for a fixed value as a function of . This result demonstrates that, increasing the value of increases the critical network size. Therefore, for a NN with large to exhibit bursting it is required that the value of be sufficiently larger than the value of . This implies that the NN must be balanced. Although the feed-forward condition is not necessary for bursting, it ensures that, in large NNs, the noise induced mode at remains stable.
In this paper, energy landscape theory, combined with thermodynamic analysis, has led to a powerful methodology for the analysis of Markovian networks. By introducing leaky Markovian networks, we developed an in silico approach for understanding the origins of bursting. We have quantified topographic deformations of the energy landscape as a function of network size and showed that bursting is a complex behavior caused by the emergence of noise-induced modes and reallocation of ground states. This led to a novel view of avalanching as a complex behavior that dominates system dynamics at near-critical or subcritical network sizes caused by appreciable levels of intrinsic noise. Future improvements in computer hardware and software will allow our methods to be used in more complicated problems than the ones considered here in an effort to theoretically understand and experimentally evaluate bursting as well as other complex phenomena.
The main objective of our work is to demonstrate that intrinsic noise in complex networks of interacting elements can induce critical behavior leading to avalanching. The strength of intrinsic noise in a given network is inversely proportional to its size, quantified by the number of nodes in the network. As the network size decreases, a critical behavior in thermodynamic behavior is observed that leads to avalanching. This is contrary to what has been done so far in the literature, which is mainly focused on how extrinsic influences to a network lead to criticality and avalanching [5]. The methodology proposed in this paper provides a link between statistical and dynamic criticality, two well-known notions of critical behavior [5]. Our results clearly demonstrate that principles fundamental to statistical mechanics can be effectively used to study criticality in phenomena that are dynamic by nature, such as avalanching.
We should also note that living biological systems, such as neuronal networks, must necessarily exchange matter and energy with their surroundings (i.e., they are open thermodynamic systems) and, as such, they should operate away from thermodynamic equilibrium (see Text S1 for details). As a consequence, it is common to refer to a homeostatic state in biological systems as non-equilibrium steady state (NESS) to make explicit the fact that living biological systems in dynamic equilibrium are not in thermodynamic equilibrium.
In this paper, we utilize familiar tools (e.g., the Boltzmann-Gibbs distribution) originally introduced in statistical mechanics to study gasses at thermodynamic equilibrium. As a consequence, the reader may think that our analysis is limited to complex networks at equilibrium. Note however that our methods are based on a Boltzmann-Gibbs distribution defined in terms of the solution of the master equation, according to Eqs. (11) & (12), which are dynamic in nature. As a consequence, our analysis is not limited to equilibrium dynamics, because the master equation is capable of describing systems far from dynamic and thermodynamic equilibrium (for more details on these two types of equilibrium, the reader is referred to Text S1). As a matter of fact, in Figure 4, we depict results away from dynamic equilibrium. Moreover, in Figure 9, we depict the maximum absolute affinity in the SISa and NN models, defined by Eq. (S72) in Text S1, for various system sizes . Figure 9A indicates that the SISa model is near thermodynamic equilibrium at network sizes below 0.3. However, this model moves away from thermodynamic equilibrium as increases beyond this value. On the other hand, Figure 9B shows that the neural network model is away from thermodynamic equilibrium for all network sizes considered, as expected. Clearly, the tools suggested in this paper are fully capable of analyzing non-equilibrium systems and NESSs, such as those that arise in biology.
To conclude, we finally point out that, in some regions of the brain, the ratio of the excitatory to inhibitory neurons is closer to 4∶1 rather than the ratio of 1∶1 used in this paper. The main reason for our choice is to compare our results to the ones in [10], which are based on the 1∶1 ratio. Qualitatively, we do not expect that changing the ratio will alter our conclusions, and this was indeed confirmed by analyzing the leaky Markovian neural network model using a 4∶1 ratio (data not shown).
|
10.1371/journal.ppat.1001295 | DNA Damage and Reactive Nitrogen Species are Barriers to Vibrio cholerae Colonization of the Infant Mouse Intestine | Ingested Vibrio cholerae pass through the stomach and colonize the small intestines of its host. Here, we show that V. cholerae requires at least two types of DNA repair systems to efficiently compete for colonization of the infant mouse intestine. These results show that V. cholerae experiences increased DNA damage in the murine gastrointestinal tract. Agreeing with this, we show that passage through the murine gut increases the mutation frequency of V. cholerae compared to liquid culture passage. Our genetic analysis identifies known and novel defense enzymes required for detoxifying reactive nitrogen species (but not reactive oxygen species) that are also required for V. cholerae to efficiently colonize the infant mouse intestine, pointing to reactive nitrogen species as the potential cause of DNA damage. We demonstrate that potential reactive nitrogen species deleterious for V. cholerae are not generated by host inducible nitric oxide synthase (iNOS) activity and instead may be derived from acidified nitrite in the stomach. Agreeing with this hypothesis, we show that strains deficient in DNA repair or reactive nitrogen species defense that are defective in intestinal colonization have decreased growth or increased mutation frequency in acidified nitrite containing media. Moreover, we demonstrate that neutralizing stomach acid rescues the colonization defect of the DNA repair and reactive nitrogen species defense defective mutants suggesting a common defense pathway for these mutants.
| Studies on intracellular bacterial pathogens have shown the need for maintaining genomic fidelity to promote colonization. Loss of DNA repair functions often leads to attenuation and rapid clearing of the invading pathogen. However, for some pathogens, an increased mutation rate has been shown to be beneficial for promoting host colonization, presumably by allowing the pathogen to rapidly adapt to adverse host conditions. We asked if the non-invasive pathogen V. cholerae experienced increased DNA damage during infection and if so, how the increased damage influenced host colonization and from where the source of the damage was derived. Our results demonstrate that V. cholerae experiences increased DNA damage during infection in the infant mouse model and that loss of ability to repair this damage results in attenuation of virulence. We specifically show that V. cholerae requires both base excision repair and mismatch repair for efficient intestinal colonization. Furthermore, we present evidence that the source of the DNA damage is derived from reactive nitrogen species (RNS) formed by acidified nitrite in the mouse gut and in doing so we identify a new RNS defense protein found in V. cholerae and several other pathogenic bacteria.
| Maintaining genomic integrity during infection is important for several bacterial pathogens to colonize their hosts. DNA repair defects in Listeria monocytogenes, Salmonella typhimurium, Helicobacter pylori and others leads to decreased or even a complete attenuation of virulence [1]–[6]. While there are several types of DNA repair in bacteria [7], many of the studies showing a requirement for DNA repair in pathogenesis focus on three pathways: the SOS response, base excision repair, and mismatch repair [1]–[10].
The SOS response is a well studied and conserved stress response in bacteria that is elicited following DNA damage and replication fork arrest [for review [11]]. The SOS response is controlled by positive and negative regulators. RecA positively regulates the SOS response by binding to single-stranded DNA fragments generated by attempted replication past DNA lesions. The RecA/ssDNA nucleoprotein filament induces the auto-cleavage of the negative regulator LexA, a transcriptional repressor. Cleavage of LexA allows for expression of 57 genes in the E. coli SOS regulon including translesion DNA polymerases that are able to replicate DNA past noncoding base lesion, and proteins involved in the inhibition of cell division [for review [11]].
Base excision repair (BER) is the most common form of repair for single base damage [for review [7]]. In BER, a DNA N-glycosylase first excises the damaged base from the deoxyribose moiety in the DNA strand creating an abasic site. Class II apurinic/apyrimidinic (AP) endonuclease then hydrolyzes the phosphodiester bond immediately 5′ to the abasic site [for review (Kornberg and Baker 1992, Friedberg 2005)]. Subsequent actions process this site to prime and repair the abasic site ultimately by DNA synthesis with DNA polymerase I and ligation by DNA ligase.
Normal replication can also introduce errors in the form of mismatched DNA base-pairs. These mismatches can lead to permanent mutations after a subsequent round of DNA replication. Mismatch repair (MMR) specifically identifies and corrects these base pairing errors increasing the fidelity of the replication pathway nearly ∼1000-fold [for review [7]].
While extensive work has shown the benefit of maintaining genomic integrity for an invading bacterium, there appear to be instances where lapses in genomic fidelity are beneficial for a pathogenic bacterium [12]–[14]. In order to colonize and thrive in a mammalian host, a bacterium must be able to adapt and respond to the conditions and stresses associated its new environment. Genomic mutations support this by allowing current gene products to gain or alter their functions. The utility of mutation(s) and a pathogen's ability to grow in the human environment has been a source of discussion for several years [13], [15]. Giraud et al. showed that a high mutation rate was initially beneficial for Escherichia coli to colonize the mouse gut, but this benefit became a liability once adaptation had been reached [12]. Oliver et al. demonstrated that Pseudomonas aeruginosa from chronically infected individuals often has an increased mutation frequency, suggesting an increased mutation rate can be beneficial to P. aeruginosa to allow rapid adaptation to the hostile host environment [14]. Thus depending on the pathogen, the mode and duration of the infection, defects in DNA repair may be detrimental or beneficial to the infecting bacterium.
Several studies have indicated that host produced reactive oxygen species (ROS) and reactive nitrogen species (RNS) cause DNA damage to the invading bacterium [4], [16], [17]. Not surprisingly bacteria have several defense mechanisms to detoxify ROS and RNS. Each enzyme detoxifies a specific type of ROS or RNS. For example catalases/peroxidases decompose H2O2, superoxide dismutases dismutate superoxide and ferrisiderophore reductase removes nitric oxide [18], [19]. As with certain DNA repair systems, loss of ROS and RNS defenses have been shown to attenuate bacterial pathogens [16], [20].
Studies supporting the importance of ROS/RNS defenses and DNA repair pathways in bacterial pathogenesis often focus on intracellular pathogens [1], [2], [4]. To survive, intracellular pathogens engulfed by phagocytic cells are either able to escape the phagosome or have mechanisms to survive within it. Within the phagosome, captured bacteria may be exposed to host production of ROS and RNS in a host defense response called the oxidative burst. It is hypothesized that the oxidative burst is responsible for the DNA damage experienced by engulfed bacteria [4], [16], [17].
Vibrio cholerae is the causative agent of the severe human diarrheal disease cholera. V. cholerae is a non-invasive pathogen that colonizes the small intestine of its host [21], [22]. As a non-invasive pathogen, V. cholerae is not expected to experience the same types of stresses as intracellular pathogens, such as an oxidative burst. However V. cholerae does pass through several hostile environments as the disease progresses. Immediately following ingestion, V. cholerae is exposed to the exceptionally antagonistic environment of the stomach where the pH of gastric acid can reach as low as 1 [23], [24]. Furthermore, nitrite from both food sources and the salivary nitrite cycle can enter the stomach creating acidified nitrite [25], [26], [27]. Acidified nitrite has potent antimicrobial effects on gut pathogens [28], [29], [30], [31]. These studies show that the viability of several pathogenic bacteria decreases rapidly under acidified nitrite conditions. Furthermore, nitrates, which can also be found in the stomach, have been shown to modify of gene expression reducing acid tolerance [32]. The antimicrobial effects of acidified nitrite are thought to be due to the generation of deleterious RNS [33]. However, with the exception of a few studies [34], [35], [36], the points of action of these RNS as well as the bacterial determinants required for protection against them have remained largely unexamined. After traversing the stomach V. cholerae faces several innate host defenses in the intestine including bile, lysozyme, small antimicrobial peptides and complement [37]. Thus, V. cholerae must overcome several barriers during infection that have the potential to cause DNA damage through a direct or indirect mechanism.
We report here that V. cholerae strain C6706 experiences increased DNA damage during passage through the murine gastrointestinal track. We demonstrate that increased genomic stress is a potential barrier to host colonization by V. cholerae. We found that two important DNA repair pathways are necessary for V. cholerae to efficiently colonize the infant mouse intestine. Furthermore, we show that defense against RNS is also necessary for V. cholerae to colonize the infant mouse. In doing so we identify a novel protein required for defense against RNS in pathogenic bacteria. In vitro we show that all our colonization defective DNA repair and RNS defense mutants share a common sensitivity to acidified nitrite and we further show that neutralizing stomach acid rescues intestinal colonization defect of these mutants.
To determine if V. cholerae requires defenses against DNA damage during colonization, we tested a series of transposon mutants that contained insertions in different steps in three important DNA repair pathways for their ability to colonize the infant mouse intestine in competition with the wild type strain. These pathways were nucleotide excision repair (NER), base excision repair (BER) and mismatch repair (MMR) (Table 1). While the SOS response is an important contributor to genomic integrity we did not test a requirement for SOS since Quiones et al. previously showed that SOS activation is not required for intestinal cholera toxin production or colonization [38]. We used uvrA as a representative gene required for NER since uvrA is obligatory for NER. We found no difference in the ability of the uvrA::Tn containing strain to colonize the infant mouse relative to the parental strain suggesting that NER is dispensable for V. cholerae pathogeneis (Table 1).
Apurinic/apyrimidininc (AP) endonucleases are critical in BER. BER has been most well studied in E. coli. E. coli encodes two class II AP endonucleases, Xth [endo II (endo VI)] and Nfo (endo IV). In E. coli Xth is responsible for ∼90% of the AP endonuclease activity in the cell [39], [40]. Few phenotypes have been attributed solely to Nfo activity but Nfo is known to contribute to BER [41]. V. cholerae carries close homologs of both Xth and Nfo (VC1860 and VC2360 respectively). Interestingly we found that the xth::Tn mutant was not defective in intestinal colonization however the nfo::Tn mutant showed a defect in colonization compared to the parental strains (Table 1). We created a clean deletion of nfo (Δnfo) in V. cholerae and found this mutant also had a colonization defect. In E. coli deletion of xth and nfo leads to a more profound defect in DNA repair than either single mutant. Consistent with this observation, we found that an xth::Tn Δnfo double mutant showed a ∼10-fold defect in colonization that appears slightly greater than the ∼5-fold defect in colonization of the Δnfo mutant alone (Table 1), although this difference is not statistically significant for this number of replicates tested (p>0.05). Thus, these results suggest that BER is important for V. cholerae to colonize the infant mouse intestine when in competition. These results also show a critical function for Nfo in survival, which has not been apparent under laboratory conditions.
Loss of mismatch repair function has been shown to be either beneficial or detrimental depending on the pathogen studied [12], [13], [14], [15]. We found that a transposon mutant in mutS, which encodes the gene product that initially binds to a mismatch, resulted in a decrease in colonization efficiency (Table 1). We constructed a clean deletion of mutS (ΔmutS) to ensure the defect was not due to the transposon. We found that the clean deletion of mutS was also attenuated in its ability to colonize the intestine suggesting that mismatch repair or at least MutS is important for V. cholerae pathogenesis (Table 1). We also found that a second clean deletion of mutS showed a similar competitive index defect (CI = 0.18±0.03) suggesting that the colonization defect was not due to mutations in the first mutS clean deletion strain. We noted that the mutS transposon mutant was more defective than its clean deletion counterpart (Table 1). This difference may be due to a polar effect of the transposon or mutations acquired by the mutS::Tn strain during outgrowth of the original isolate. We also tested the colonization proficiency of a Δnfo ΔmutS double mutant and found that the colonization defect of this double mutant appears slightly greater than either the Δnfo or ΔmutS mutant alone (Table 1) although this difference is not statistically significant for this number of replicates tested (p>0.05).
The requirement of BER and MMR for V. cholerae to efficiently colonize the infant mouse intestine suggests that V. cholerae experiences DNA damage in the mouse, and that a reduced ability to repair such damage is detrimental for V. cholerae pathogenesis.
V. cholerae genes encoding Xth, Nfo and MutS were identified based on sequence similarity with their well-studied E. coli homologs. To ensure the V. cholerae homologs possessed their predicted functions we tested our mutant strains for the well characterized phenotypes described in other bacterial systems. Loss of mismatch repair causes an increase in mutation rate often referred to as a mutator phenotype [42]. We found that our ΔmutS mutant had a significantly increased mutation frequency compared with the wild type control (Figure 1A). The wild type phenotype could be restored by expression of mutS from a plasmid but not by the plasmid itself (Figure S1B). This result indicates that MutS in V. cholerae shares the same activity as its other well studied bacterial homologs in the repair of DNA replication errors.
Loss of Xth activity in E. coli renders the strain sensitive to hydrogen peroxide (H2O2) [43]. We found that our xth::Tn strain was also sensitive to H2O2 (Figure 1B). Loss of nfo activity alone does not greatly sensitize E. coli to H2O2 but loss of xth and nfo creates a strain with increased sensitivity to H2O2 [43]. We found a similar effect in V. cholerae where the xth::Tn Δnfo strain was much more sensitive to H2O2 then the xth::Tn mutant alone (Figure 1B). Furthermore, high level expression of nfo from a plasmid complemented the H2O2 sensitivity of the xth::Tn Δnfo mutant (Figure S1C). These results suggest that V. cholerae Nfo acts like its E. coli homolog.
The requirement of BER and mismatch repair (MMR) systems for V. cholerae to efficiently colonize the mouse intestine suggests that V. cholerae experiences increased DNA damage while in the mouse. To address this possibility we measured the mutation frequency of V. cholerae following passage though the mouse as compared to passage in liquid culture. We inoculated five mice and five liquid cultures with the same size inoculums of V. cholerae. The following day we purified bacteria from the mouse intestine (see Materials and Methods). We plated both V. cholerae passaged through the mouse and grown in liquid cultures followed by selection for resistance to two antibiotics we used as an indicator for measuring mutation frequency. The first was a gain of function mutation in rpoB conferring resistance to rifampicin; the second was a loss of function of thyA conferring resistance to trimethoprim. Mutations in rpoB and thyA are well characterized markers for increases in mutation frequency [44], [45], [46]. We found that following passage of V. cholerae through the mouse there was an ∼2 fold increase in rifampicin resistance and ∼2.5 fold increase in trimethoprim resistance compared to the liquid culture grown strains (Figure 2A, B). We sequenced 19 trimethoprim resistance isolates that were passed through the mouse and 20 isolates obtained following growth in liquid culture. We identified 39 unique mutations in thyA (data not shown) suggesting that our results were not influenced by a mutation acquired early on in the procedure. We did not observe a bias in the types of mutation from the two conditions. These results suggest that passage through the mouse results in an increase in mutation rate for V. cholerae suggestive of an increase in DNA damage and the need for repair mechanisms.
We have identified two DNA repair mechanisms required by V. cholerae to efficiently colonize the infant mouse, and have shown that V. cholerae passaged through a mouse has an increased mutation frequency. Thus, we sought to identify potential causes of DNA damage for V. cholerae while in the mouse to understand the requirement for BER and MMR in the mouse. A major source of DNA damage for intracellular pathogens is from host produced ROS and RNS. While V. cholerae is a non-invasive pathogen we considered that it still may experience ROS and RNS at some point during infection. We used a genetic approach to determine if ROS/RNS affected V. cholerae colonization and if so what type(s) of ROS/RNS were most important during this encounter.
Bacteria have several enzymes to detoxify ROS/RNS. Each enzyme detoxifies a specific type of ROS or RNS [for review see [18], [19]. For example catalases/peroxidases decompose H2O2, superoxide dismutases remove superoxide and ferrisiderophore reductases remove nitric oxide. Bacteria can contain multiple proteins capable of dealing with one type of stress. V. cholerae possesses two catalases/peroxidases (KatB/PerA) and one alkyl hydroperoxide reductase (AhpC), three superoxide dismutases (SodA/B/C) but only one ferrisiderophore reductase (HmpA). We tested mutants defective for each of these different types of defense enzymes to identify the type(s) of radicals that may be damaging V. cholerae in the mouse (Table 2).
RNS, including nitric oxide, have been shown to be powerful antimicrobial agents. The most well studied RNS defense enzyme in bacteria is Hmp, a ferrisiderophore reductase that destroys nitric oxide [47]. V. cholerae carries an hmp homolog, hmpA. Both an hmpA::Tn mutant and a ΔhmpA deletion mutant showed a defect in colonizing the infant mouse intestine (Table 2). Deletion of hmpA delayed V. cholerae growth in the presence of a nitric oxide donor but not in the absence (Figure 3A) consistent with previous observations in other bacteria [20], [48]. This suggests that V.cholerae may encounter deleterious RNS during passage in the mouse. The growth defect of the ΔhmpA mutant in the presence of a nitric oxide donor could be complemented by ectopic expression of hmp from the arabinose inducible plasmid pBAD18 (Figure S1D). In fact, expression of hmp from pBAD18 allowed the ΔhmpA mutant to recover growth more rapidly than the parental strain in the presence of a nitric oxide donor.
After testing all previously predicted antioxidant enzymes we began to mine the V. cholerae genome for additional putative antioxidant enzymes. We began by searching for putative proteins that belonged to large antioxidant families. Enzymes, such as AhpC, belong to the Peroxiredoxin (PRX) family. Searching for peroxiredoxin family proteins yielded a putative defense enzyme we have called PrxA (VC2637). PrxA, classified is a peroxiredoxin-5 family protein, is found in several pathogenic bacteria and is a distant homolog of a macrophage peroxynitrite detoxification protein [49].
Deletion of prxA did not effect V. cholerae growth in LB alone but significantly delayed V. cholerae growth in the presence of a nitric oxide donor (Figure 3A). Furthermore, both the prxA::Tn mutant and the ΔprxA allele we constructed caused a decrease in the ability of V. cholerae to colonize the infant mouse in competition assays (Table 2). The growth defect of the ΔprxA mutant in the presence of a nitric oxide donor could be complemented by ectopic expression of prxA from the arabinose inducible plasmid pBAD18 (Figure S1E). The discovery of a new gene required for both defense against RNS and efficient colonization of the infant mouse further supports our findings that V. cholerae may be exposed to RNS during passage though the mouse.
We tested the sensitivity of a ΔprxA ΔhmpA double mutant and found that the growth of the double mutant in the presence of a nitric oxide donor was even more delayed than either the ΔprxA or ΔhmpA single mutant alone (Figure 3A). We also tested the colonization efficiency of a ΔprxA ΔhmpA (Table 2) and found that it was not significantly less than the ΔprxA mutant alone (p>0.05). Thus hmpA and prxA are both important for colonization but the effects were not additive.
We asked if defects in ROS defense also affect V. cholerae colonization. Disruption of ahpC, katB, perA, sodA or sodC did not affect the ability of V. cholerae to colonize the infant mouse and these deficiencies did not affect the ability of V. cholerae to colonize the infant mouse in competition experiments (Table 2). We did not test the sodB::Tn mutant since both it and a ΔsodB deletion strain we constructed had a decreased plating efficiency and grew very poorly compared to the parental strain (Figure S1F). Thus, while SodB appears to be important for growth of V. cholerae under laboratory conditions we did not pursue the sodB mutant in mouse experiments.
Interestingly, of ahpC, katB and perA only disruption of perA sensitized V. cholerae to H2O2 in vitro (Figure 3B). Furthermore, of strains disrupted individually for sodA, sodB and sodC only disruption of sodB sensitized V. cholerae to the superoxide generating compound plumbagin (Figure 3C). We also tested the ΔprxA mutant but found that it did not show increased sensitivity to either H2O2 or plumbagin (Figure S2A and data not shown). It is possible that some of these known ROS defense enzymes overlap in function masking the effects of a deficiency in any one gene in vitro or in mouse studies. For other bacterial pathogens and symbionts deletion of several or all catalases and superoxide dismutases has been required before a strong effect on virulence or symbiosis was observed [50], [51], [52]. Currently, the results from our analysis suggest RNS may pose a significant barrier to V. cholerae in colonizing the infant mouse. ROS may also play a role, however their effect is not immediately evident in our analysis.
XthA appears to be more important than Nfo in protecting V. cholerae against environmental stress in vitro (Figure 1B), yet Nfo appears to be more important for colonization of the small intestine (Table 1). The requirement for ΔhmpA and ΔprxA for efficient intestinal colonization lead us to ask if Nfo was required for defense against nitric oxide. We monitored the growth of the xth::Tn, Δnfo and xth::Tn Δnfo mutants in the presence of a nitric oxide donor. We found that, at least in vitro, xth:Tn was more important than Nfo for protection against nitric oxide (Figure 3D). We also found that the double mutant was again more sensitive to the stress than either single mutant alone (Figure 3D).
Our results led us to ask if our DNA repair and RNS defense defective V. cholerae mutants were sensitive to any host defenses. The intestine has several innate defenses [37]. We tested many of these defenses including lysozyme, phospholipase, antimicrobial peptides, complement, bile, changes in osmolarity and pH, however, we did not observe any difference in sensitivity between the parental and the mutant strains (data not shown). RNS have been shown to be generated by macrophages to kill phagocytised bacteria [reviewed in [53]]. The RNS from macrophages is generated by an inducible nitric oxide synthase (iNOS). Inhibition of iNOS activity has been shown to rescue the virulence defects in hmp mutant strains of Salmonella enterica serovar typhimurium [20]. However, our hmpA::Tn V. cholerae mutant showed no difference in it ability to colonize the intestine of a wild type or isogenic iNOS−/−infant mouse (Table S1).
Thus, our results suggest that the colonization defect of the DNA repair and RNS defense mutants may occur before V. cholerae is exposed to the host defenses found in the small intestine. In the stomach V. cholerae is exposed to low pH in combination with µM amounts of nitrite from ingested food and the salivary nitrite cycle [25], [26], [27]. Acidified nitrite produces a variety of toxic RNS. We quantified the amount of nitrite in the infant mouse stomach using the Griess reaction and found that it was 20.0±0.7 µM, which is similar to that of humans [26]. The pH range of human gastric juice is reported as 1–3 [23], [24]. We determined the pH of the infant mouse stomach to be 4.5±0.1 using a fluorescent pH sensitive dye. This measurement is conservative and the pH of the infant mouse gastric juice may be even less (see Materials and Methods). Thus, the infant mouse stomach is sufficiently acidic to promote the formation of acidified nitrite.
At pH 3 in rich medium we found that V. cholerae had a greater than 99.9% decrease in survival in less than 1 minute (data not shown) agreeing with similar work examining V. cholerae acid tolerance [54]. We did not find a difference in survival between the parental and mutant strains at low pH (1–4) levels (data not shown). We gradually increased pH to identify the lowest level at which V. cholerae could grow. At pH 5.5 V. cholerae and the DNA repair and RNS defense mutants grew with identical kinetics (Figure 4A). We titrated nitrite into the growth medium and found that nearly all the mutant strains showed a growth defect compared to the wild type at pH 5.5 in the presence of 400 µM nitrite (Figure 4B). No differences in growth between wild type and mutant strains were observed at pH 7.0 with or without 400 µM nitrite (Figure S3A, B). Not only did low pH and nitrite slow the growth of our mutants but the ΔhmpA, ΔprxA, Δnfo and xth::Tn Δnfo mutants began to show a decrease in optical density after longer exposure (Figure 4B) suggesting the cells were lysing.
Only the growth of the ΔmutS strain was unaffected at 400 µM. We considered that while MutS may not be required for survival of acidified nitrite during this time course it may be required to prevent acidified nitrite induced mutations in V. cholerae that are detrimental for colonization. We grew V. cholerae in LB at pH 5.5 over night in the presence or absence of 600 µM nitrite and then plated for rifampicin resistant colonies. We found that V. cholerae grown in the presence of nitrite had a greater than 10-fold increase in mutation frequency compared to the media only control (Figure 4C). Loss of MutS then increased the mutation frequency of V. cholerae in nitrite at pH 5.5 ∼ an additional 5-fold (Figure 4C). Thus, MutS may be important to prevent acidified nitrite induced mutations that could impair the ability of V. cholerae to colonize the infant mouse. To further test this possibility we created a ΔhmpA ΔmutS double mutant and tested its colonization proficiency (Table 2). Interestingly, the colonization defect of the ΔhmpA ΔmutS double mutant was not significantly different than either of the single mutants alone (p>0.05). This result may suggest that HmpA and MutS may share a similar defense pathway in the infant mouse.
Additionally, E. coli MutS can recognize an O6-methyl-dG:dC base pair, a mutation which can occur by alkylation of G bases [55]. Therefore it is possible that MutS may also be important for protection again some type of alkylation that occurs in the mouse stomach.
If acidified nitrite produces RNS that damaged V. cholerae DNA, we reasoned that we should be able to detect increased intracellular radical formation in V. cholerae following nitrite treatment. We grew V. cholerae at pH 5.5 with increasing amounts of nitrite and assayed for radical formation using 2′,7′-dichlorodihydrofluorescein diacetate (H2DCFDA). H2DCFDA is a cell permeable dye that fluoresces after reacting with RNS and ROS species. H2DCFDA reacts with several ROS and RNS including hydrogen peroxide, nitric oxide and peroxynitrite [56], [57], [58]. We found that H2DCFDA fluorescence correlated with increasing nitrite concentration at pH 5.5 indicating an increase in intracellular radicals (Figure 4D). For comparison, we found that H2DCFDA fluorescence did not increase over the concentrations of nitrate examined at pH 7.0. These results further support the requirement for low pH to induce radical formation from nitrite sources (Figure 4D).
Our results suggested that V. cholerae may experience DNA damage as it passes through the stomach. If so, we were curious if a colonization defect could be observed at an early time point after inoculation. We repeated the competitive colonization assays testing the ΔmutS, Δnfo, ΔprxA and ΔhmpA mutants. We found that at 3 hours post inoculation each mutant strain already showed a 50–60% colonization defect when co-inoculated with the wild type (Table S2) though this defect was not as great as the 5–20% defects reported in Table 1. This suggests that a defect of the mutant strains is detrimental for colonization early after inoculation. Angelichio et al. [59] reported that V. cholerae populations in the small intestine do not show a significant increase in number until between 10–24 h post inoculations. These results suggest that the effects of the DNA damage are ∼50% detrimental at the earliest stages of infection and become more apparent as the bacteria replicate to high numbers in the intestine. Such a conclusion is consistent with the concept of damage occurring primarily in the stomach.
Our results support a relationship between acidified nitrite sensitivity and colonization defects in our RNS and DNA repair. We reasoned that if acidified nitrite in the stomach was responsible for the colonization defects of our mutants then neutralizing stomach acid in the mouse should relieve, at least in part, the observed colonization defects. We used sodium bicarbonate to neutralize the mouse stomach acid. When we inoculated infant mice with our DNA repair and RNS defense defective mutants in the presence of sodium bicarbonate all four mutant strains (Δnfo, ΔmutS, ΔprxA and ΔhmpA) showed significant improvement in their ability to colonize the intestine in competition with the parental strain (Table 3). In fact the colonization defect of the ΔmutS and Δnfo mutant was completely rescued. The colonization defect of the ΔprxA was restored to near wild type levels. The colonization defect of the ΔhmpA mutant was partially rescued although this difference is not statistically significant for the number of replicates tested (p>0.05).
The nitrite concentration of the infant mouse stomach after sodium bicarbonate treatment remained nearly unchanged at 20.3±0.8 µM.
We have shown that V. cholerae must defend against DNA damage to efficiently colonize the infant mouse intestine and that such damage likely occurs early during infection as V. cholerae enters the stomach. We have demonstrated that V. cholerae specifically requires BER and MMR pathways to efficiently colonize the infant mouse intestine. Furthermore, we have identified one previously known and one novel RNS defense protein that facilitates intestinal colonization of the infant mouse. These DNA repair and RNS defense proteins were also required for V. cholerae to grow or maintain genomic fidelity in the presence of acidified nitrite. Furthermore the colonization defects of each mutant could be partially or fully complemented by neutralizing stomach acid suggesting that RNS defense and DNA repair share a common defensive role in the mouse.
V. cholerae has been shown to be very sensitive to low pH [54]. For this reason, human volunteers have their stomach contents neutralized to promote experimental V. cholerae infection as is done with live attenuated vaccine studies [60]. In the recently developed infant rabbit model for cholera, stomach acid is also neutralized and cimetidine is administered to prevent re-acidification in order for V. cholerae to colonize the infant rabbit intestine [61]. However, our DNA repair and RNS defense mutants did not show increased sensitivity to low pH compared to the parent strain. Agreeing with our observations a large screen used to identify genes necessary for colonization and tolerating low pH in V. cholerae did not identify any of the genes we reported here for influencing colonization [62]. While low pH of the stomach alone is undoubtedly detrimental towards V. cholerae our results suggest that neutralizing stomach acid may also be important to prevent RNS formation by acidified nitrite. We propose that the defect in colonization of the DNA repair and RNS defense mutants is due to RNS formation.
Acidified nitrite is present in the stomach where gastric juice interacts from nitrite sources from the diet or salivary nitrite pathway. The chemistry of acidified nitrite is known to produce several potentially deadly radicals (1).(1)Both NO• and NO2• can directly attack cellular macromolecules, but they can also interact with other radicals to form further species such as peroxynitrite and hydroxyl acids. From equation (1) we can see that a key factor in this process is the pH of the solution. In the human stomach, normal nitrite concentrations range from 10–50 µM [24] but at a pH of 1–3 acidification and radical production can happen rapidly. The low pKa of HNO2 may explain in part why we required much higher concentrations of nitrite to observe detrimental effects on the mutants we studied. pH 5.5 was the lowest pH level we could successfully grow V. cholerae, a value well above the pKa of HNO2. Thus at pH 5.5 the acidification of nitrite would occur more slowly and higher concentrations of nitrite would be necessary for mass action to drive the acidification and radicalization of nitrite. Agreeing with this we did not observe any effect of nitrite on the growth of our mutants or V. cholerae at pH 7.0. While acidified nitrite has been shown to effectively kill several bacterial pathogens [28], [29], [30], [31], the mechanism of its action and bacterial defenses to protect against it have remained unknown. We have now shown that MutS, Nfo, HmpA and PrxA are required for protection of acidified nitrite in V. cholerae.
While our results support a role for acidified nitrite in the stomach acting as a major DNA damaging agent, we have not identified the exact location in the gastrointestinal track where the damage occurs. The most apparent location is the stomach where ingested V. cholerae mixes with gastric juice. However it is possible that DNA damage induced by acidified nitrite radicals occurs, or continues to occur, in the upper intestinal tract. As gastric juice exits the stomach it is neutralized by bile salt, etc. and can reach a pH close to 8 [24]. Since the stability of at least some RNS, such as peroxynitrite, increases with increasing pH [63] the upper intestinal tract may provide a more favorable environment for RNS to reach V. cholerae and induce DNA damage.
Bicarbonate has been shown to induce V. cholerae virulence genes in a ToxT dependent fashion [64]. Abuaita and Withey show that significant upregulation of both cholera toxin and tcpA gene expression are observed 3–4 h after addition of bicarbonate [64]. While our inoculation of V. cholerae occurs on a much shorter time scale (∼5–15 min after exposure of bicarbonate) and the majority of V. cholerae has passed into the small intestine before 3 h, it is possible that some bicarbonate induced gene regulation may also aid in the bicarbonate rescue of the colonization defect of our DNA repair mutants.
While the debate over the benefits and detriments of increased mutation frequency for pathogenesis continues, we have shown that increased mutation frequency is detrimental to V. cholerae pathogenesis, at least for the short-term colonization of the infant mouse intestine. However, we cannot exclude the possibility that increased mutation frequency affects long-term survival of V. cholerae in the host. After the initial decrease in competitiveness it is possible that increased mutation frequency in V. cholerae could make it more competitive in later stages of colonization or during release into the environment. It would be interesting to test multiple clinical isolates for a mutator phenotype to address this question.
The Xth and Nfo homologs of V. cholerae have strong sequence similarity to their E. coli counterparts. We have shown that V. cholerae and E. coli deletion mutants of xth and nfo also share a similar pattern of sensitivity to hydrogen peroxide. Nfo and Xth have been most extensively studied in E. coli. In E. coli Xth is responsible for greater than ∼90% of all AP endonuclease activity in the cell [39], [40]. In E. coli, an xth mutant is very sensitive to a variety of DNA damaging agents whereas nfo mutants generally show milder effects [41]. Interestingly, we have shown that in V. cholerae Nfo is more important for colonization of the infant mouse than Xth. Our nfo xth double mutant suggests that Xth may play a role in colonization when Nfo is absent. However, for whatever damage is occurring, Nfo appears to play a more important role in the mouse.
It is possible that Nfo and Xth are also used differentially for repair of specific types of lesions. Preferences for specific types of damaged bases between Nfo and Xth from E. coli have been previously reported [65]. If RNS are responsible for DNA damage in the mouse we suggest that Nfo may have enhanced ability to aid in the repair of nitrosylative base damage. Additionally, there may be differential expression of xth and nfo or their preferential glycosylase partners in the host.
In our efforts to identify ROS and RNS defense enzymes required for intestinal colonization, we identified a new protein we have called PrxA that was required for RNS defense. Until now Hmp has been the only bacterial protein identified to detoxify RNS, specifically nitric oxide. We have shown that like HmpA, PrxA protects V. cholerae against the nitric oxide donor spermine NONOate. Similarly, HmpA and PrxA both protect V. cholerae against acidified nitrite. This agrees with previous work showing that Hmp protects Salmonella against nitric oxide and acidified nitrite [34]. While the species(s) produced by acidified nitrite that HmpA and PrxA defend against is not clear, we presume that it is a RNS. PrxA homologs are not as prevalent in bacteria as HmpA homologs, but they are found in several pathogens including Yersina pestis, Haemphilus influenza and Neisseria gonorrhoeae. It will be interesting to determine if PrxA homologs share a similar RNS defense role in other bacteria.
Our observation that HmpA and PrxA are required for colonization lead us to suggest that V. cholerae encounters RNS stress during infection. When studying ROS defense genes we found that deletion of SodB was detrimental for normal V. cholerae growth. This indicated that normal growth of V. cholerae must generate a significant amount of superoxide managed by SodB. The growth defect of the ΔsodB mutant prevented us from analyzing it by competition in the mouse model. While we did not identify any other single ROS defense enzyme that affected intestinal colonization it is possible that construction of various double mutants may show that V. cholerae must also deal with ROS during disease progression. In bacterial pathogens where SODs have been shown to be necessary for virulence, it is generally the periplasmic SOD that is required as this SOD encounters superoxide entering the cells from the environment [16], [66]. However, the V. cholerae periplasmic SOD, SodC, was not required for intestinal colonization suggesting V. cholerae does not experience superoxide stress from the host.
The animal experiments were performed with protocols approved by Harvard Medical School Office for Research Protection Standing Committee on Animals. The Harvard Medical School animal management program is accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care, International (AAALAC), and meets National Institutes of Health standards as set forth in the Guide for the Care and Use of Laboratory Animals (DHHS Publication No. (NIH) 85-23 Revised 1996). The institution also accepts as mandatory the PHS Policy on Humane Care and Use of Laboratory Animals by Awardee Institutions and NIH Principles for the Utilization and Care of Vertebrate Animals Used in Testing, Research, and Training. There is on file with the Office of Laboratory Animal Welfare (OLAW) an approved Assurance of Compliance (A3431-01).
Strains and plasmids are listed in Supporting Table S3. V. cholerae El Tor biotype strain C6706 and a spontaneous lacZ− derivative of C6706, were used as parental (wild type - WT) strains. E. coli DH5α λpir and Sm10 λpir were used for cloning and conjugation, respectively. Antibiotic concentrations used were streptomycin (Sm: 100 µg/ml or 500 µg/ml), kanamycin (Kan: 50 µg/ml), carbenicillin (Carb: 75 µg/ml) and chloramphenicol (Cm: 2.5 µg/ml for C6706 and 10 µg/ml for E. coli DH5α λpir). LB was used for all growth conditions [10 g/liter of tryptone (Bacto), 5 g/liter of yeast extract (Bacto), and 5 g/liter of NaCl] and was supplemented with 16 g/liter of agar (Bacto) for growth on plates. Arabinose was used at 0.1% for complementation assays. All ID numbers/ Accession numbers/for genes highlighted in this study are shown in Table S5.
The genomic sequence of C6706 has not been completed. We used the sequence of the close relative, N16961, for clone construction. For in-frame gene deletions of nfo, mutS, hmpA and prxA, genomic DNA surrounding the respective gene was amplified by crossover PCR and cloned into pWM91 for subsequent sacB-mediated allelic exchange in V. cholerae, as described [67], [68]. For complementation constructs, the respective gene was amplified from chromosomal DNA and cloned into plasmid pBAD18 after digestion with KpnI and SalI. The respective gene was induced by adding arabinose to the growth medium. All cloning products were sequence-verified, and the nucleotide sequence of all primers used is listed in Table S4.
A modified version of the protocol of Baselski and Parker [69] was performed for infection and recovery of C6706 derived strains. C6706 or C6706 lac− and mutant strains were grown on LB-agar plates with Sm overnight at 37°C. Wild type and mutant strains were mixed together in LB. 50 µl of this competition mixture (∼50 000 bacteria) was inoculated into a 5-day-old CD1 mouse pup (Charles River Company). Serial dilutions of the competition mixture were plated in LB+Sm100 and enumerated to determine the input ratio of wild type and mutant strain. After incubation at 30°C for 3 h or 18 h the mouse pups were sacrificed and small intestines were removed and homogenized in 10 ml of LB. Serial dilutions were plated in LB+Sm100 and enumerated to determine the output ratio of wild type and mutant strain. The competitive index for each mutant is defined as the input ratio of mutant/wild type strain divided by the output ratio of mutant/wild type strain. A minimum of four mice were assayed for each mutant strain. The in vivo experiments for the transposon and clean deletion strains were the accumulation of results performed on different days. For ease of communication we reported the average competitive index.
For NaHCO3 recue experiments, mice pups were first inoculated with 50 µl of 2.5 g/100 mL NaHCO3. After 3 h the pups were inoculated with 50 µl of the competition mixture in 2.5g/ 100 mL NaHCO3. iNOS−/− (#002609) and control C57BL/6J (#000664) mice were purchased from The Jackson Laboratory.
We have developed a fluorescence based assay to determine the pH of the infant mouse stomach. We first determined a standard curve using the fluorescent pH indicator Yellow/Blue DND-160 (Invitrogen) over a range from pH 3–8. We then extracted the gastric juice from 5 individual mice, diluted the sample 1∶2 with ddH2O (pH 7), added Yellow/Blue DND-160 and determined the fluorescence of the solution. Comparing these fluorescent values to our standard curve we determined the pH of the infant mouse stomach to be 4.5±0.1. We also note that this is a conservative measurement. In order to obtain enough liquid we diluted the gastric sample ∼1∶2 with ddH20 that was at ∼pH 7. Thus while water is not a buffer, the dilution of the gastric juice likely raised the final pH of our measurements.
Nitrite concentration was determined using the Griess Reagent System (Promega TB229). The concentration shown is the average of 10 mice treated with or without sodium bicarbonate.
Strains were grown to exponential phase in LB with Cm when required. Strains were serial diluted and spotted on LB plates containing increasing concentrations of hydrogen peroxide and incubated at 37°C overnight. For complementation Cm and arabinose were added while strains were growing in liquid, as well as in the LB agar plates.
Strains were grown to exponential phase in LB. Strains were then diluted to OD600 0.01 in LB±1.0 mM spermine NONOate and grown at 37°C in a 96 well plate with aeration (SpectraMax Plus 384, Molecular Devices). OD600 readings were taken every 15 min.
Overnight cultures were diluted into LB and grown to log phase at 37°C with aeration. Cultures were diluted to OD600 0.05 in 25 mM MES buffered LB of pH 7.0 or 5.5 with or without the addition of 400 µM sodium nitrite (Sigma-Aldrich). The LB media and MES were adjusted to a pH of 7.0 and 5.5 (Corning pH meter 240) with additions of HCl, and filter sterilized (0.22 µm, Corning) prior to use. The growth of strains under various treatments were determined by OD600 measurement using a 96 well format spectrophotometer (SpectraMax Plus 384, Molecular Devices). Environmental parameters were set to 37°C with shaking and readings were taken every 15 minutes for 16 hours. Studies were conducted in quadruplicate.
Overnight cultures were diluted into 100 mL LB with Sm100 and grown to OD600∼0.8 (37°C, aeration). 10 mL of culture was dispensed into 15 mL conical and centrifugated at 5,000 RPM for 5 minutes. The supernatant discarded and cells resuspended in an equal volume of 25 mM MES buffered LB of pH 7.0 or 5.5 with or without the addition of sodium nitrite (500 µM, 1 mM, 5mM, or 10 mM). Cells were treated for 1.5 hours at 37°C with aeration then centrifugated at 5,000 RPM for 5 minutes at 4°C. The supernatant was discarded, cells resuspended in 1 mL PBS (LONZA), and transferred to a 1.5 mL eppendorf tube. The cells were centrifugated and washed an additional two times in 1× PBS before being resuspended in 1 ml of PBS with 10 µM 2′,7′-dichlorodihydrofluorescein diacetate (Molecular Probes, Invitrogen). The cells were incubated at room temperature for 30 minutes then centrifugated and washed three times to remove all free, extracellular dye. The cells were lysed in 225 µL of lysis buffer (MilliQ water with 0.1M EDTA) via sonication. Cell lysates were centrifugated at 15,000 RPM for 5 minutes, supernatant transferred to another 1.5 mL eppendorf tube and centrifugated again. Fluorescence was measured at 490 nm / 519 nm (excitation/emission) (SpectraMax Gemini XS). Fluorescence was normalized against protein concentrations, as determined by Bradford assay. Studies were conducted in triplicate.
Statistical significance was assessed for mouse colonization assays and ΔmutS mutation frequency assays using a one-way analysis of variance (ANOVA) using a Bonferroni post test to determine significant differences in competitive index between all pairs of V. cholerae mutants used in our study. Statistical significance of acidified nitrite, nitric oxide and H2O2 sensitivities was assessed using a mixed model, repeated measures two-way analysis of variance (ANOVA), generating a p value for each pair wise curves over the concentration range of H2O2 to determine the significance of our results. Statistical significance of rifampicin and trimethoprim resistant mutants from LB vs. mouse passaged samples were assessed using a paired t-test. Differences were considered significant at p<0.05. All calculations were performed using Graphpad Prisim version 5.
|
10.1371/journal.pntd.0001435 | Identification of Loci Controlling Restriction of Parasite Growth in Experimental Taenia crassiceps Cysticercosis | Human neurocysticercosis (NC) caused by Taenia solium is a parasitic disease of the central nervous system that is endemic in many developing countries. In this study, a genetic approach using the murine intraperitoneal cysticercosis caused by the related cestode Taenia crassiceps was employed to identify host factors that regulate the establishment and proliferation of the parasite. A/J mice are permissive to T. crassiceps infection while C57BL/6J mice (B6) are comparatively restrictive, with a 10-fold difference in numbers of peritoneal cysticerci recovered 30 days after infection. The genetic basis of this inter-strain difference was explored using 34 AcB/BcA recombinant congenic strains derived from A/J and B6 progenitors, that were phenotyped for T. crassiceps replication. In agreement with their genetic background, most AcB strains (A/J-derived) were found to be permissive to infection while most BcA strains (B6-derived) were restrictive with the exception of a few discordant strains, together suggesting a possible simple genetic control. Initial haplotype association mapping using >1200 informative SNPs pointed to linkages on chromosomes 2 (proximal) and 6 as controlling parasite replication in the AcB/BcA panel. Additional linkage analysis by genome scan in informative [AcB55xDBA/2]F1 and F2 mice (derived from the discordant AcB55 strain), confirmed the effect of chromosome 2 on parasite replication, and further delineated a major locus (LOD = 4.76, p<0.01; peak marker D2Mit295, 29.7 Mb) that we designate Tccr1 (T. crassiceps cysticercosis restrictive locus 1). Resistance alleles at Tccr1 are derived from AcB55 and are inherited in a dominant fashion. Scrutiny of the minimal genetic interval reveals overlap of Tccr1 with other host resistance loci mapped to this region, most notably the defective Hc/C5 allele which segregates both in the AcB/BcA set and in the AcB55xDBA/2 cross. These results strongly suggest that the complement component 5 (C5) plays a critical role in early protective inflammatory response to infection with T. crassiceps.
| Infection with the cestode Taenia solium causes cysticercosis in humans and pigs. Neurocysticercosis is a severe manifestation of T. solium infection that constitutes an important health concern in developing countries. Studies in humans living in areas of endemic disease and in pigs experimentally infected have suggested a large spectrum of permissiveness to T. solium multiplication, with the possible contribution of genetic factors. In the present report, we have used an experimental mouse model of intraperitoneal infection with Taenia crassiceps to study the potential role of genetic factors in regulating replication of this parasite. Our study focused on two inbred mouse strains A/J and C57BL/6J that are respectively permissive and non-permissive to intraperitoneal multiplication of T. crassiceps. We have used a set of AcB/BcA recombinant congenic strains of mice along with standard F2 crosses to decipher the complexity and nature of the genetic component of the A/J vs. C57BL/6J interstrain difference in permissiveness. Our results point to a major role of the complement component 5 (C5) in early response and protection against T. crassiceps infection.
| Taenia solium seriously affects human health in many countries of Latin America, Asia and Africa [1]. The life cycle of T. solium includes a larval phase (cysticercus), which develops in both pigs and humans from ingested eggs contaminating the environment. When humans ingest improperly cooked pork meat infected with live cysticerci, the cysticerci develop to the stage of an adult intestinal tapeworm, which produces millions of eggs that are then shed to the environment in human faeces [2]. In rural communities where the disease is endemic, unsanitary conditions and presence of free-roaming pigs result in up to 9% of the human open population of these areas to be infected. Despite this high infection rate, only a small fraction of carriers become symptomatic and develop NC, suggesting intrinsic differences in host susceptibility to infection and pathogenesis of the disease [3]. Indeed, several reports have pointed at possible genetic effects in response to cysticercosis in human and pigs. In humans, multi-case families were identified in areas of highly endemic disease, favoring the idea of the participation of multiple genes in NC causality [3]. In a case-control study, resistance to NC was found associated to HLA [4]. Also, a three to five fold difference in parasite load was detected in a genetically heterogeneous pig cohort experimentally challenged with T. solium eggs [5].
Taenia crassiceps is a tapeworm of wild and domestic animals, which does not cause clinical disease in non-immunocompromised humans [6]. T. crassiceps has been used as an experimental model for cysticercosis due to its ability to proliferate by budding [7] and colonize the peritoneal cavity of the murine host [7], where its replication can be measured over time by enumeration of recovered metacestodes. Although the T. crassiceps ORF strain is unable to develop into adult tapeworms [8], its property to rapidly multiply in the peritoneal cavity of infected mice has been extensively used to explore the relevance of biological factors in host-parasite interactions [9], and to identify protective antigens of interest for vaccine development [9], [10]. The mechanisms involved in the protective immunity against T. crassiceps cysticercosis have been extensively studied, but are not fully understood. Studies in inbred mouse strains (growth permissive H2d-bearing BALB/c; growth restrictive H2b-bearing C57BL/6J) initially pointed at the importance of the major histocompatibility locus (MHC) and MHC-linked genes in regulating intraperitoneal growth of the parasite [11]. This was confirmed by additional studies of H2 congenic BALB/c substrains, where BALB/cJ mice express the Qa2 protein and are significantly more resistant than the BALB/cAnN mice [12], [13]. This differential susceptibility may be explained in part by activation of antigen presenting cells, and production of pro-inflammatory cytokine and modulatory chemokines both early and late during T. crassiceps infection [14]. Furthermore, phenotyping of different inbred strains has suggested that an additional, non-MHC linked genetic component may contribute to regulation of T. crassiceps replication [15]. Finally, clear differences between the parasite load of male and female have been noted in inbred mouse strains [16]. Females show higher numbers of cysticerci compared to males due to a significant effect of sex hormones on response to infection [17], [18].
With the aim of further characterizing the host genetic factors that affect host response to T. crassiceps cysticercosis, the differential susceptibility of A/J (permissive) and C57BL/6J (restrictive) mouse strains was studied. For this, a set of 34 reciprocal AcB/BcA recombinant congenic strains (RCS) derived by systematic inbreeding from a double backcross (N3) between A/J and C57BL/6J parents [19] was phenotyped for response to T. crassiceps infection. In the breeding scheme used to derive the AcB/BcA strains set, each of the strains harbors 12.5% of its genome from either A/J or B6, fixed as a set of discrete congenic segments onto 87.5% of the reciprocal parental background. The vast range of permissiveness to T. crassiceps growth in 34 RCS, as measured by the parasite load 30 days post-infection, along with haplotype association mapping suggested that response to T. crassiceps cestode is under complex genetic control, with identifiable contributions of chromosomes 2 and 6. Subsequent genetic linkage analysis in informative crosses validated the chromosome 2 locus, and established the regional position of the regulating locus.
The AcB/BcA set of recombinant congenic strains (RCS) were derived from a double backcross (N3) between A/J and C57BL/6J parents at McGill University and were provided by Emerillon Therapeutics. The breeding, genetic characteristics and genotype of RCS have been described earlier [19]. Inbred strains A/J, B6, and DBA/2 were obtained as pathogen-free mice at 7–8 weeks of age from the Jackson Laboratory (Bar Harbor, ME) and maintained as breeding colonies at UNAM. [AcB55xDBA/2] F2 progeny were bred by systematic brother-sister mating of [AcB55xDBA/2] F1 mice.
The study protocol (register number 021) was approved by the ethics committee of the Instituto de Investigaciones Biomédicas, Universidad Nacional Autónoma de México (UNAM). All housing and experimental procedures were performed according to the principles set forth in the Guide for the Care and Use of Laboratory Animals, Institute of Laboratory Animal Resources, National Council, Washington, D.C. 1996.
The fast growing ORF strain of T. crassiceps, originally isolated by R. S. Freeman [7], was maintained by serial intraperitoneal (i. p.) passage in female BALB/cAnN mice, as previously described [13]. All experimental mice were inoculated intraperitoneally with 10 small (<2 mm) non-budding T. crassiceps larvae, re-suspended in sterile isotonic saline. Thirty days following infection, parasites were harvested from the peritoneal cavity and counted using a stereoscopic microscope [20] to determine the parasite burden. Organs inside the abdominal cavity were removed and carefully inspected for any remaining T. crassiceps larvae.
Genomic DNA was isolated from tail clips of individual F2 mice collected at the time of sacrifice, as previously described [19]. A total of 185 female [AcB55xDBA/2]F2 mice were genotyped at the Centre for Applied Genomics (The Hospital for Sick Children, Toronto, Canada) using the Illumina Mouse Low Density Linkage panel containing 377 SNPs distributed across the genome, out of which 161 were polymorphic between AcB55 and DBA/2 strains. Additional microsatellite markers were obtained from the Mouse Genome Informatics Database (www.informatics.jax.org) and used for gap filling and fine mapping by a standard PCR-based method employing (α-32P) dATP labeling and separation on denaturing 6% polyacrylamide gels.
C5 status in the F2 mice was confirmed by RFLP analysis, as previously described [21]. Briefly, C5 fragment was amplified by PCR and digested with Bsg I, which recognizes a novel restriction site introduced by the 2-bp deletion in exon 6 of the Hc gene [21], [22]. The fragments were resolved on 2% agarose gel; the expected size for wild-type C5 was 446 bp, while the sizes for the samples containing the deletion were 318 and 126 bp.
QTL mapping was performed using Haley-Knott multiple regression analysis [23] and the two-dimensional scan was performed using the two-QTL model. Empirical genome-wide significance was calculated by permutation testing (1000 tests). All linkage analysis was performed using R/qtl [24]. The detailed algorithm underlying the efficient mixed-model for association mapping has been previously published [25]. The EMMA algorithm is based on the mixed-model association where a kinship matrix accounting for genetic relatedness between inbred mouse strains is estimated and then fitted to the vector of the phenotype, thereby reducing false positive rates. EMMA is publically available as an R package implementation.
An unpaired, two-tailed Student's t-test was used to establish significance of differences in mean parasite burden between mouse Tccr1 and C5 genotypes. These data were analyzed using GraphPad Prism 4.0 statistical software. P-values<0.05 were considered significant.
A/J and C57BL/6J (B6) mice show differential permissiveness to cysticercosis [16], following the intra-peritoneal inoculation of 10 small (<2 mm diameter) non-budding T. crassiceps larvae (Figure 1). In A/J mice, there is rapid parasite reproduction in the peritoneal cavity, which is detectable by visual and histological examination of the mice (Figure 1A, 1B), and by quantification of parasite load (Figure 1C, magnification in 1D). Enumeration of the parasites recovered 30 days following infection (Figure 2) indicates a 10-fold difference in parasite replication between A/J (200–250) and C57BL/6J (15–30). To study the genetic control of differential replication of T. crassiceps in restrictive B6 and permissive A/J strains, we phenotyped a set of 34 AcB/BcA recombinant congenic strains [19]. The breeding scheme used to generate the reciprocal AcB/BcA strains set results in individual strains harboring 12.5% of its genome donated from either A/J (in BcA strains) or B6 (in AcB strains), fixed as a set of discrete congenic segments onto 87.5% of the reciprocal parental background [19]. The AcB/BcA strain set has previously been used to characterize different genetic traits that control phenotypic differences between B6 versus A/J, including mapping of major monogenic trait [19], and dissection of complex genetic traits into several simple genetic effects [26], [27]. Between 5–10 animals from each strain were infected with T. crassiceps and the total parasite burden was determined 30 days later (Figure 2A and 2B). We segregated AcB/BcA strains according to the predominant genetic background (left and right panels in Figure 2), and further grouped them into permissive or restrictive categories based on the overall mean parasitic load, whereby strains harboring an average of >66 parasites were deemed permissive while those showing <66 were termed restrictive (determined as two standard deviations from the parasite load of restrictive B6 parental group). According to this arbitrary segregation, the majority of BcA strains were parasite growth restrictive, with the notable exception of strains BcA73, 70, 72 and 83 that showed parasite loads similar to those detected in the A/J controls (Figure 2B). Conversely, AcB strains were found to be generally permissive for parasite growth, with the notable exception of strains AcB55 and AcB60 that displayed an average of 12 and 31 cysticerci, respectively. The presence of such discordant strains in both sets of RCS suggests the possibility that the restrictiveness/permissiveness trait is under simple genetic control, and that transfer of a single congenic fragments onto the opposite strain background strongly influences the phenotype of the recipient strain. Such a situation would be similar to the segregation of the Ccs3 (colorectal cancer) [28], Ity (susceptibility to Salmonella) [29] and Lgn1 loci (susceptibility to Legionella) [19] we previously reported in these strains.
To explore the nature and complexity of the genetic control of parasite replication in the AcB/BcA strains set, we performed haplotype association mapping using parasite load as the primary phenotype and 1200 informative polymorphic genetic markers. We applied a statistical model EMMA [25] that corrects for genetic relatedness and population structure of the RCS by computing a kinship matrix in a manner analogous to an inbred mouse strain analysis, as we previously described [30]. Using this approach, we detected suggestive association of chromosome 2 (proximal region) and chromosome 6 alleles with T. crassiceps permissiveness (Figure 3). In the case of chromosome 6, both proximal (weaker) and distal (stronger) portions of the chromosome showed association. Also, for both chromosomes 2 and 6, A/J alleles are associated with permissiveness while B6 alleles are associated with restrictiveness, as expected. Additional strength for these associations is provided by some of the phenodeviant strains; for example, for the proximal part of chromosome 2 (∼35 Mb), susceptible BcA70, 72, and 73 harbor A/J haplotypes, while resistant AcB55 and AcB60 harbor B6-derived haplotypes (Figure S1). Likewise, for distal chromosome 6, BcA72, 73 and 83 have A/J alleles, while AcB55 and 60 have B6 alleles (Figure S1). Nevertheless, the imperfect correlation for both chromosomes requires validation in secondary crosses. It also suggests presence of additional genetic effects controlling T. crassiceps permissiveness in the strain set.
The genetic control of host response to T. crassiceps was further investigated in strain AcB55. This strain consistently showed the lowest parasite burden in the AcB set (Figure 2, mean parasite load = 11.7), despite ∼87.5% of its background being inherited from the highly susceptible A/J parent (mean parasite load = 256). Therefore, we hypothesized that AcB55 is likely to carry B6-derived chromosomal segments responsible for restrictiveness in both AcB55 and B6. To map such segments, AcB55 was crossed to the permissive strain DBA/2 (Figure 4A, mean parasite load = 108) to produce an [AcB55xDBA/2] F2 population in which individual animals would be informative for the entire genome in linkage analyses. [AcB55xDBA/2] F1 hybrids and 379 [AcB55xDBA/2] F2 animals were infected intraperitoneally with 10 non-budding T. crassiceps larvae in three separate infections, and parasite burden was measured 30 days later (Figure 4A). Due to the previously reported gender-associated differential permissiveness to T. crassiceps, where higher concentrations of estrogen and estradiol are concomitant with increased parasite burdens [17], [18] typically occurring in female mice, we segregated males and females in the analysis (Figure 4A). Both male and female [AcB55xDBA/2] F1 hybrids were fully resistant with parasite load similar to the AcB55 controls (Figure 4A), suggesting that resistance to T. crassiceps is inherited in a dominant fashion. The parasite load of [AcB55xDBA/2] F2 mice followed a continuous distribution between highly permissive and highly restrictive animals, with a clustering of F2 animals in the resistant range. This suggests both a complex genetic control of permissiveness to infection, with a dominant pattern of inheritance of restrictiveness more apparent in the female population (Figure 4A). Interestingly, we observed an identical pattern of inheritance of restrictiveness (dominant) in a distinct F2 population, where the AcB55 strain was crossed to the permissive A/J founder (Figure S2). However, due to limited genetic diversity in the [AcB55xA/J] F2 progeny (∼12.5% due to B6 genomic segments), we conducted genetic linkage analysis by whole genome scanning in the [AcB55xDBA/2] F2 population. Because the frequency distribution of parasite load in F2 females was skewed, we applied a log2 correction to normalize the distribution, followed by regression to an experiment-specific mean (set at 0) (Figure 4B).
A total of 185 [AcB55xDBA/2] F2 female mice were genotyped with the Illumina Mouse Low Density Linkage Panel consisting of 161 informative polymorphic markers with 10 additional microsatellite markers to complete genome coverage. Whole-genome multiple regression linkage analysis in R/qtl (Figure 4C) identified a highly significant locus associated with parasite burden on chromosome 2 (LOD = 4.76, P<0.01) and an additional suggestive linkage on chromosome 19 (LOD = 4.03, P<0.05) (Figure 4C and S3). These loci contributing restrictiveness to T. crassiceps in the AcB55 strain were given a temporary appellation Tccr1 (Taenia crassiceps cysticercosis restrictiveness 1) for chromosome 2 QTL and Tccr2, for chromosome 19 QTL. The Bayesian 95% credible intervals were determined to be 13.1–44.1 Mb for Tccr1 (Figure 5A) and the entire chromosome 19 for Tccr2 (Figure S3), whereas the peak LOD scores were identified at 29.7 Mb (peak marker: D2Mit295) and 46.7 Mb (peak SNP: rs13483650), respectively.
To examine the effect of the most significant Tccr1 locus on parasite load, F2 mice were segregated according to genotype at the most significant chromosome 2 marker D2Mit295 (Figure 5B). Mice carrying the DBA/2 alleles at Tccr1 have significantly higher number of cysticerci (P = 0.0009; Student's t-test) than those harboring one or two AcB55 alleles. This indicates that restrictiveness alleles at Tccr1 are inherited in a strictly dominant fashion. We also examined the LOD score trace for chromosome 19 and the inheritance of parental alleles underlying Tccr2 at the peak marker (rs13483650) (Figure S3). We observed two prominent peaks, suggesting that Tccr2 may be due to multiple genetic effects contributing to lower parasite loads in the AcB55 strain. However, once we segregated the female F2 mice according to their genotype (Figure S3B), we observed that the Tccr2 QTL is driven by heterozygosity, with homozygosity for neither the AcB55- nor the DBA/2-derived alleles having a significant effect on the parasite burden. To determine whether Tccr1 and Tccr2 acted in an additive or epistatic manner, a two-dimensional Haley-Knott multiple regression analysis was carried out, followed by simulation of an overall QTL model using R/qtl. This analysis revealed that, although Tccr1 and Tccr2 are non-interacting loci, their individual and joint contribution to the full QTL model raises the LOD score to 11.26 and explains 25% of observed phenotypic variance.
Interestingly, linkage analysis in the AcB55xDBA/2 cross was successful in validating the chromosome 2 association initially detected by EMMA analysis in 34 RCS (Figure 3), but not the more significant chromosome 6 association. In fact, detailed examination of the region underlying chromosome 6 indicated that both AcB55 and DBA/2 strains harbor similar haplotype blocks at the proximal and distal regions of chromosome 6 (data not shown) and would therefore not segregate in the analyzed F2 cross. Together, haplotype association mapping in 34 RCS along with linkage analysis in an informative F2 cross strongly suggest that T. crassiceps replication in the murine host is controlled by multiple genetic factors, amongst which Tccr1 strongly contributes to the noted resistance of AcB55 mice.
Maximum linkage for Tccr1 locus is coupled to the D2Mit295 marker, which lies at 29.7 Mb and was previously associated to the gene coding for hemolytic complement (Hc/C5) [21] located approximately 5 Mb further downstream (34.8–34.9 Mb). Since C5 was mapped in an F2 cross derived from A/J and B6 strains and its deficiency correlated to high susceptibility to the fungal pathogen Candida albicans in the majority of RCS [21], we examined the involvement of C5 in the context of T. crassiceps infection. We genotyped the parental AcB55 and DBA/2 strains for the deficiency-causing 2-bp deletion in the C5 gene and confirmed that DBA/2 is C5-deficient [22], while AcB55 is wild type for C5 and does not harbor the deletion [21]. C5 status was also determined in 185 female [AcB55xDBA/2] F2 mice and the parasite replication permissiveness was associated with C5-deficiency (Figure 5C) in a recessive manner, where mice harboring at least one functional copy of the gene are fully resistant. In addition, classifying the set of 34 RCS according to permissiveness to infection and C5 status further corroborates the association of wild-type C5 alleles with increased protection against T. crassiceps (Table 1).
Taken together our results suggest a critical role for the complement component 5 in restricting proliferation of T. crassiceps in mice.
The panel of 34 reciprocal AcB/BcA RCS has been used to map major monogenic traits [19], [28] or to facilitate identification of multiple loci involved in complex trait diseases [26], [27]. This approach is based on the premise that unique small congenic fragments derived from the donor strain are fixed and delineated for each strain, which may allow for detection of causative haplotype by the sole study of the strain distribution pattern in relation to the phenotype of interest [19]. Here, we have phenotyped the set of 34 RCS to study the genetic control of susceptibility to T. crassiceps cysticercosis (Figure 2), where the majority of AcB strains were found to be permissive for parasite replication and conversely, most of the BcA strains were deemed restrictive, similarly to progenitor strains A/J and B6, respectively. The presence of phenodeviant strains allowed us to conduct haplotype association mapping and identify significant genomic regions associated with response to infection on chromosomes 2 and 6 (Figure 3). Subsequent genome scan in informative F2 mice generated between resistant AcB55 and susceptible DBA/2 progenitors (fixed for chromosome 6 locus) identified a highly significant linkage on chromosome 2 (Tccr1, LOD score = 4.76, P value<0.01) as conferring resistance to AcB55 in a dominant fashion along with an additional heterozygous-driven effect observed on chromosome 19 (Tccr2, LOD score = 4.03, P value = 0.05) (Figure 4 and Figure 5).
Interestingly, the Tccr1 locus was analogous to a previously identified susceptibility locus for the fungal pathogen Candida albicans attributable to the complement component 5 gene (Hc/C5) [21]. The complement pathway represents the initial line of defense of the innate immune system and elicits an inflammatory response to the site of infection [31], [32]. Activation of the complement cascade is triggered by microbial products via several pathways, which ultimately results in the activation of C3 convertase, cleavage of C5, release of chemotactic factors (C3a and C5a), and generation of the membrane attack complex (MAC) [31]. Similarly to T. crassiceps infection, inbred mouse strains display various degrees of susceptibility to C. albicans, where A/J is highly susceptible and B6 resistant [21]. This differential susceptibility was attributed to the major gene C5, where a single allele of the wild-type C5 confers complete resistance to infection [21]. As previously described [22], C5-deficiency is caused by a 2-bp deletion in the exon 6, leading to a premature stop codon and a product that is not secreted in the serum [33]. We have confirmed that AcB55 carries the B6 allele at C5 and is therefore C5-suficient and resistant to T. crassiceps, whereas DBA/2 is C5-deficient and susceptible. C5 status was also determined in 185 [AcB55xDBA/2]F2 progeny (Figure 5C), where resistance segregated with one or two copies of the wild-type C5 indicating that C5 exerts an early dominant protective effect upon T. crassiceps infection.
In the mouse model of T. crassiceps infection, although the extent of parasite replication depends on the orchestration of immune and hormonal responses, the parasite restriction is initiated by the early immunological response, which was shown to destroy the peritoneal larvae [34] and involve the complement system in innate resistance to T. taeniaeformis [35]. Earlier studies demonstrated that C3 along with IgG is deposited on larval T. taeniaeformis [35], [36] as early as 2 days post infection, but that it is not directly involved in lytic activity through formation of the MAC. Rather, the complement system indirectly triggers host cell recognition and/or activation [35], leading to parasite elimination. Inhibition of complement by cobra venom factor administration to the resistant BALB/cByJ mice prior to T. taeniaformis infection results in decreased parasite mortality, indicating that complement component deposition and activation is necessary in host defense against taenid infection [35]. More precisely, the role of complement component 5 (C5) was established in the mouse model of hyatid disease, caused by the Echinococcus granulosus cestode. C5-deficiency in B10.D2 o/SnJ mice infected with E. granulosus was associated with poor eosinophil infiltration and increased growth of established cysts [37], indicating that C5-mediated mechanisms are detrimental for parasite growth.
The immunopathology of T. crassiceps cysticercosis in mice is that of an initial non-permissive Th1 type, which shifts to a parasite permissive Th2 type during chronic stage of infection and is accompanied by an increase in IL-4, IL-6 and IL-10 cytokines [38]–[40]. This transition was also reported in individuals with T. solium neurocysticercosis [41] and more precisely, in the brain granulomas surrounding T. solium parasite [42]. Although A/J mice succumb to C. albicans infection early after infection (48 h), they mount a similar Th2-like cytokine storm consisting of high levels of IL-6, IL-10 and TNFα [43], [44], suggesting a common role for C5 in both pathologies. In addition, animals deficient for the Th1 hallmark cytokine IFNγ or upon its neutralization exhibit increased susceptibility to C. albicans [45] and T. crassiceps [39], respectively. Regarding the role of complement in parasite damage, it is known that cysticerci are able to inhibit both classical and alternative pathways of the complement cascade, through C1q inhibition by parasite's paramyosins [46], but the relevance of the complement component 5 (C5) in the restrictiveness to parasite growth remains to be elucidated, especially considering that C5 is cleaved in two different active forms: C5a, which is a potent anaphylatoxin and chemotactic protein, and C5b, which has a binding site for C6 and is the molecule responsible for promoting the MAC (membrane attack complex) assembling in cell membrane. To assess the importance of C5 and possible involvement of additional genetic effects in the A/J versus B6 differential susceptibility to T. crassiceps, we classified the RCS according to susceptibility and C5 status (Table 1). In the majority of RCS, C5 status was a strong predictor of response to T. crassiceps infection with C5 deficiency causing increased parasite replication permissiveness, even when present on the B6 background as seen in BcA73, BcA70, BcA72 and BcA83 strains. Conversely, the presence of wild-type alleles at C5 on an otherwise permissive A/J background, as seen in AcB55 strain, conferred restrictiveness to infection. We noted the presence of several discordant strains (Table 1, denoted by an asterisk), such as BcA76, AcB60 and AcB63. Although C5 explains 25% of phenotypic variance, discordant strains suggest that additional susceptibility and resistance genetic factors, distinct from the dominant C5 effect, modulate response to T. crassiceps infection. In fact, STAT4 and STAT6 transcription factors were recently involved in modulation of the immune response to T. crassiceps in mice. Permissiveness of BALB/c mice was abrogated in STAT6−/− mice of the same background, as they were able to efficiently mount a strong Th1 response and control the infection [47]. Conversely, Th1 response induced via STAT4-dependent signaling pathway is essential for development of immunity against cysticercosis [48]. Also, macrophage activation and subsequent production of nitric oxide represent additional mechanisms essential for host resistance to T. crassiceps infection [49]. Interestingly, we did not uncover any associations of the MHC with susceptibility in our AcB55xDBA/2 cross where DBA/2 mice carry parasite permissive H2-d [11]. This further suggests that the differential permissiveness observed in BALB/c susbstrains and attributed to H2-d haplotype is in fact, BALB-background specific. Similarly, the Tccr2 locus represents a novel effect that arose upon combination of the AcB55 and DBA/2 genetic backgrounds and further emphasizes the importance of genetic modifiers in T. crassiceps permissiveness. Finally, generation of additional informative crosses will be necessary to validate and study the effect contributed by the chromosome 6 underlying genetic effect(s) in T. crassiceps cysticercosis, independently of C5 or in conjunction.
In conclusion, we have combined haplotype association mapping in 34 recombinant congenic strains and linkage analysis to identify and validate a novel locus that modulates the outcome to T. crassiceps infection in mice. We have demonstrated that C5 underlies the Tccr1 locus and uncovered an important role of the complement pathway in susceptibility to T. crassiceps cysticercosis.
|
10.1371/journal.pcbi.1006928 | Stability of working memory in continuous attractor networks under the control of short-term plasticity | Continuous attractor models of working-memory store continuous-valued information in continuous state-spaces, but are sensitive to noise processes that degrade memory retention. Short-term synaptic plasticity of recurrent synapses has previously been shown to affect continuous attractor systems: short-term facilitation can stabilize memory retention, while short-term depression possibly increases continuous attractor volatility. Here, we present a comprehensive description of the combined effect of both short-term facilitation and depression on noise-induced memory degradation in one-dimensional continuous attractor models. Our theoretical description, applicable to rate models as well as spiking networks close to a stationary state, accurately describes the slow dynamics of stored memory positions as a combination of two processes: (i) diffusion due to variability caused by spikes; and (ii) drift due to random connectivity and neuronal heterogeneity. We find that facilitation decreases both diffusion and directed drifts, while short-term depression tends to increase both. Using mutual information, we evaluate the combined impact of short-term facilitation and depression on the ability of networks to retain stable working memory. Finally, our theory predicts the sensitivity of continuous working memory to distractor inputs and provides conditions for stability of memory.
| The ability to transiently memorize positions in the visual field is crucial for behavior. Models and experiments have shown that such memories can be maintained in networks of cortical neurons with a continuum of possible activity states, that reflects the continuum of positions in the environment. However, the accuracy of positions stored in such networks will degrade over time due to the noisiness of neuronal signaling and imperfections of the biological substrate. Previous work in simplified models has shown that synaptic short-term plasticity could stabilize this degradation by dynamically up- or down-regulating the strength of synaptic connections, thereby “pinning down” memorized positions. Here, we present a general theory that accurately predicts the extent of this “pinning down” by short-term plasticity in a broad class of biologically plausible network models, thereby untangling the interplay of varying biological sources of noise with short-term plasticity. Importantly, our work provides a novel theoretical link from the microscopic substrate of working memory—neurons and synaptic connections—to observable behavioral correlates, for example the susceptibility to distracting stimuli.
| Information about past environmental stimuli can be stored and retrieved seconds later from working memory [1, 2]. Strikingly, this transient storage is achieved for timescales of seconds with neurons and synapse transmission operating mostly on time scales of tens of milliseconds and shorter [3]. An influential hypothesis of neuroscience is that working memory emerges from recurrently connected cortical neuronal networks: memories are retained by self-generating cortical activity through positive feedback [4–7], thereby bridging the time scales from milliseconds (neuronal dynamics) to seconds (behavior).
Sensory stimuli are often embedded in a physical continuum: for example, positions of objects in the visual field are continuous, as are frequencies of auditory stimuli, or the position of somatosensory stimuli on the body. Ideally, the organization of cortical working memory circuits should reflect the continuous nature of sensory information [3]. A class of cortical working memory models able to store continuously structured information is that of continuous attractors, characterized by a continuum of meta-stable states, which can be used to retain memories over delay periods much longer than those of the single network constituents [8]. Continuous attractors were proposed as theoretical models for cortical working memory [9–11], path integration [12–14], and other cortical functions [15–17] (see e.g. [3, 18–21] for recent reviews), well before experimental evidence was found in cortical networks [22] and the limbic system [18, 23]. The one-dimensional ring-attractor in the fly responsible for self-orientation [24, 25] is a particularly intriguing example.
Continuous attractor models have been successfully employed in the context of visuospatial working memory to explain behavioral performance [26–29], to predict the effects of neuromodulation [30, 31], or the implications of cognitive impairment [32, 33]. However, in networks with heterogeneities, the continuum of memory states quickly breaks down, since noise and heterogeneities break, transiently or permanently, the crucial symmetry necessary for continuous attractors [10, 11, 13, 16, 34–40]. For example, the stochasticity of neuronal spiking (“fast noise”) leads to transient asymmetries that randomly displace encoded memories along the continuum of states [10, 11, 35, 37, 39, 40], leading, averaged over many trials, to diffusion of encoded information. More drastically, introducing fixed asymmetries (“frozen noise”) due to network heterogeneities causes a directed drift of memories and a collapse of the continuum of attractive states to a set of discrete states. Examples of heterogeneities in biological scenarios include the sparsity of recurrent connections [13, 36], or randomness in neuronal parameters [36] and values of recurrent weights [16, 34, 38]. Since both (fast) noise and heterogeneities are expected in cortical settings, the feasibility of continuous attractors as computational systems of the brain has been called into question [3, 6, 41].
The question then arises, whether short-term plasticity of recurrent synaptic connections can rescue the feasibility of continuous attractor models. In particular, short-term depression has a strong effects on the directed drift of attractor states in rate models [42, 43], but no strong conclusions were drawn in a spiking network implementation [44]. Short-term facilitation, on the other hand, increases the retention time of memories in continuous attractor networks with noise-free [38] and, as shown parallel to this study, noisy [45] rate neurons. In simulations of continuous attractors implemented with spiking neurons for a fixed set of parameters, facilitation was reported to cause slow drift [46, 47] and a reduced amount of diffusion [47]. However, despite the large number of existing studies, several fundamental questions remain unanswered. What are the quantitative effects of short-term facilitation in more complex neuronal models and across facilitation parameters? How does short-term depression influence the strength of diffusion and drift, and how does it interplay with facilitation? Do phenomena reported in rate networks persist in spiking networks? Finally, can a single theory be used to predict all of the effects observed in simulations?
Here, we present a comprehensive description of the effects of short-term facilitation and depression on noise-induced displacement of one-dimensional continuous attractor models. Extending earlier theories for diffusion [39, 40, 45] and drift [38], we derive predictions of the amount of diffusion and drift in ring-attractor models of randomly firing neurons with short-term plasticity, providing, for the first time, a general description of bump displacement in the presence of both short-term facilitation and depression. Our theory is formulated as a rate model with noise, but since the gain-function of the rate model can be chosen to match that of integrate-and-fire models, our theory is also a good approximation for a large class of heterogeneous networks of integrate-and-fire models as long as the network as a whole is close to a stationary state. The theoretical predictions of the noisy rate model are validated against simulations of ring-attractor networks realized with spiking integrate-and-fire neurons. In both theory and simulation, we find that facilitation and depression play antagonistic roles: facilitation tends to decrease both diffusion and drift while depression increases both. We show that these combined effects can still yield reduced diffusion and drift, which increases the retention time of memories. Importantly, since our theory is, to a large degree, independent of the microscopic network configurations, it can be related to experimentally observable quantities. In particular, our theory predicts the sensitivity of networks with short-term plasticity to distractor stimuli.
We investigated, in theory and simulations, the effects of short-term synaptic plasticity (STP) on the dynamics of ring-attractor models consisting of N excitatory neurons with distance-dependent and symmetric excitation, and global (uniform) inhibition provided by a population of inhibitory neurons (Fig 1A). For simplicity, we describe neurons in terms of firing rates, but our theory can be mapped to more complex neurons with spiking dynamics. An excitatory neuron i with 0 ≤ i < N is assigned an angular position θ i = 2 π N i − π ∈ [ − π , π ), where we identify the bounds of the interval to form a ring topology (Fig 1A). The firing rate ϕi (in units of Hz) for each excitatory neuron i (0 ≤ i < N − 1) is given as a function of the neuronal input:
ϕ i ( t ) = F ( J i ( t ) + J inh ) . (1)
Here, the input-output relation F relates the dimensionless excitatory Ji and inhibitory Jinh inputs of neuron i to its firing rate. This represents a rate-based simplification of the possibly complex underlying neuronal dynamics [48]. We assume that the excitatory input Ji(t) to neuron i at time t is given by a sum over all presynaptic neurons
J i ( t ) = ∑ j = 0 N − 1 w i j s j ( t ) , (2)
where wijsj(t) describes the total activation of synaptic input from the presynaptic neuron j onto neurons i. The maximal strength wij of recurrent excitatory-to-excitatory connections is chosen to be local in the angular arrangement of neurons, such that connections are strongest to nearby excitatory neurons (Fig 1A, red lines). The momentary input depends also on the synaptic activation variables sj, to be defined below. Finally, connections to and from inhibitory neurons are assumed to be uniform and global (all-to-all) (Fig 1A, blue lines), thereby providing non-selective inhibitory input Jinh to excitatory neurons.
As a model of STP, we assume that excitatory-to-excitatory connections are subject to short-term facilitation and depression, which we implemented using a widely adopted model of short-term synaptic plasticity [49]. The outgoing synaptic activations sj of neuron j are modeled by the following system of ordinary differential equations:
s ˙ j = − s j τ s + u j x j ϕ j , u ˙ j = − u j − U τ u + U ( 1 − u j ) ϕ j , x ˙ j = − x j − 1 τ x − u j x j ϕ j . (3)
The synaptic time scale τs governs the decay of the synaptic activations. The timescale of recovery τx is the main parameter of depression. While the recovery from facilitation is controlled by the timescale τu, the parameter 0 < U ≤ 1 controls the baseline strength of unfacilitated synapses as well as the timescale of their strengthening. For fixed τu, we consider smaller values of U to lead to a “stronger” effect of facilitation, and take U = 1 as the limit of non-facilitating synapses.
As a reference implementation of this model, we simulated networks of spiking conductance-based leaky-integrate-and-fire (LIF) neurons with (spike-based) short-term plastic synaptic transmission (Fig 1B1, see Spiking network model in Materials and methods for details). For these networks, under the assumption that neurons fire with Poisson statistics and the network is in a stationary state, neuronal firing can be approximated by the input-output relation F of Eq (1) [50, 51] (see Firing rate approximation in Materials and methods), which allows us to map the network into the general framework of Eqs (1) and (2). In the stationary state, synaptic depression will lead to a saturation of the synaptic activation variables sj at a constant value as firing rates increase. This nonlinear behavior enables spiking networks to implement bi-stable attractor dynamics with relatively low firing rates [46, 52] similar to saturating NMDA synapses [11, 47]. Since we found that without depression (for τx → 0) the bump state was not stable at low firing rates (in agreement with [52]), we always keep the depression timescale τx at positive values.
Particular care was taken to ensure that networks display nearly identical bump shapes (similar to Fig 1B1, inset; see also S1 Fig), which required the re-tuning of network parameters (recurrent conductance parameters and the width of distance-dependent connections; see Optimization of network parameters in Materials and methods) for each combination of the STP parameters above.
Simulations with spiking integrate-and-fire neurons generally show a bi-stability between a non-selective state and a bump state. In the non-selective state, all excitatory neurons emit action potentials asynchronously and irregularly at roughly identical and low firing rates (Fig 1B1, left of dotted line). The bump state can be evoked by stimulating excitatory neurons localized around a given position by additional external input (Fig 1B1, red dots). After the external cue is turned off, a self-sustained firing rate profile (“bump”) emerges (Fig 1B1, right of dashed line, and inset) that persists until the network state is again changed by external input. For example, a short and strong uniform excitatory input to all excitatory neurons causes a transient increase in inhibitory feedback that is strong enough to return the network to the uniform state [11].
During the bump state, fast fluctuations in the firing of single neurons transiently break the perfect symmetry of the firing rate profile and introduce small random displacements along the attractor manifold, which become apparent as a random walk of the center position. If the simulation is repeated for several trials, the bump has the same shape in each trial, but information on the center position is lost in a diffusion-like process. We additionally included varying levels of biologically plausible sources of heterogeneity (frozen noise) in our networks: random connectivity between excitatory neurons (E-E) and heterogeneity of the single neuron properties of the excitatory population [36], realized as a random distribution of leak reversal potentials. Heterogeneities makes the bump drift away from its initial position in a directed manner. For example, the bump position in the randomly connected (p = 0.5) network of Fig 1B1 shows a clear upwards drift towards center positions around 0. Repeated simulations of the same attractor network with bumps initialized at different positions provide a more detailed picture of the combined drift and diffusion dynamics: bump center trajectories systematically are biased towards a few stable fixed points (Fig 1B2) around which they are distributed for longer simulation times (histogram in Fig 1B2, t = 13.5s). The theory developed in this paper aims at analyzing the above phenomena of drift and diffusion of the bump center.
To untangle the observed interplay between diffusion and drift and investigate the effects of short-term plasticity, we derived a theory that reduces the microscopic network dynamics to a simple one-dimensional stochastic differential equation for the bump state. The theory yields analytical expressions for diffusion coefficients and drift fields, that depend on short-term plasticity parameters, the shape of the firing rate profile of the bump, as well as the neuron model chosen to implement the attractor.
First, we assume that the system of Eq (3) together with the network Eqs (1) and (2) has a 1-dimensional manifold of meta-stable states, i.e. the network is a ring-attractor network as described in the introduction. This entails, that the network dynamics permit the existence of a family of solutions that can be described as a self-sustained and symmetric bump of firing rates ϕ0,i(φ) = F(J0,i(φ)) with corresponding inputs J0,i(φ) (for 0 ≤ i < N). Importantly, the center φ of the bump can be located at any arbitrary position φ ∈ { j N 2 π − π | 0 ≤ j < N }. For example, if ϕ0,i(0) is a solution with input J0,i(0), then ϕ 0 , i + 1 ( 2 π N ) is also a solution with input J 0 , i + 1 ( 2 π N ). This solution is illustrated in Fig 1C for a bump centered at φ = 0. Second, we assume that the number N of excitatory neurons is large (N → ∞), such that we can think of the possible positions φ as a continuum. Third, we assume that network heterogeneities are small enough to capture their effect as a linear (first order) perturbation to the stable bump state. Our final assumption is that neuronal firing is noisy, with spike counts distributed as Poisson processes, and that we are able to replace the shot-noise of Poisson spiking by white Gaussian noise with the same mean and autocorrelation, similar to earlier work [39, 53]; see Diffusion in Materials and methods, and Discussion. Under these assumptions, we are able to reduce the network dynamics to a one-dimensional Langevin equation, describing the dynamics of the center φ(t) of the firing rate profile (see Analysis of drift and diffusion with STP in Materials and methods):
φ ˙= B η ( t ) + A ( φ ) . (4)
Here, η(t) is white Gaussian noise with zero mean and correlation function 〈η(t), η(t′)〉 = δ(t − t′).
The first term is diffusion characterized by a diffusion strength B1, which describes the random displacement of bump center positions due to fluctuations in neuronal firing. For A(φ) = 0 this term causes diffusive displacement of the center φ(t) from its initial position φ(t0), with a mean (over realizations) squared displacement of positions 〈[φ(t) − φ(t0)]2〉 = B ⋅ (t − t0) that, during an initial phase, increases linearly with time [14, 54, 55], before saturating due to the circular domain of possible center positions [39]. Our theory shows (see Diffusion in Materials and methods) that the coefficient B can be calculated as a weighted sum over the neuronal firing rates (Fig 1D)
B = ∑ i ( C i S) 2 ( d J 0 , i d φ) 2 ϕ 0 , i , (5)
where d J 0 , i d φ is the change of the input to neuron i under shifts of the center position (Fig 1C, orange line), and S is a normalizing constant that tends to increase additionally with the synaptic time constant τs.
The analytical factors Ci express the spatial dependence of the diffusion coefficient on the short-term plasticity parameters through
C i = U ( 1 + 2 τ u ϕ 0 , i + U τ u 2 ϕ 0 , i 2 ) ( 1 + U ϕ 0 , i ( τ u + τ x ) + U τ u τ x ϕ 0 , i 2 ) 2 . (6)
1In Brownian motion, the diffusion constant is usually defined as D = B/2. The dependence of the single summands in Eq (5) on short-term plasticity parameters is visualized in Fig 1D, where we see that: a) due to the squared spatial derivative d J 0 , i d φ of the bump shape and the squared factors Ci/S, the important contributions to the sum arise primarily from the flanks of the bump; b) for a fixed bump shape, summands increase with stronger short-term depression (larger τx) and decrease with stronger short-term facilitation (smaller U, larger τu).
The second term in Eq (4) is the drift field A(φ), which describes deterministic drifts due to the inclusion of heterogeneities. For heterogeneity caused by variations in neuronal reversal potentials and random network connectivity, we calculate (see Frozen noise in Materials and methods) systematic deviations Δϕi(φ) of the single neuronal firing rates from the steady-state bump shape that depend on the current position φ of the bump center. In Drift in Materials and Methods, we show that the drift field is then given by a weighted sum over the firing rate deviations:
A ( φ ) = ∑ i C i S d J 0 , i d φ Δ ϕ i ( φ ) , (7)
with weighing factors depending on the spatial derivative of the bump shape d J 0 , i d φ and the parameters of the synaptic dynamics through the same factors Ci/S. This is illustrated in Fig 1E: in contrast to Eq (5) summands are now asymmetric with respect to the bump center, since the spatial derivative is not squared.
To calculate the diffusion and drift terms of the last section, we assume the number of neurons N to be large enough to treat the center position φ as continuous: this allows us (similar to [39]) to derive projection vectors (see Projection of dynamics onto the attractor manifold in Materials and methods) that yield the dynamics of the center position. However, the actual projection yields sums over the system size N, whose scaling we made explicit (see System size scaling in Materials and methods). For the diffusion strength B (cf. Eq (5)) we find a scaling as as 1/N, in agreement with earlier work [11, 14, 36, 39, 46]. For drift fields caused by random connectivity, we find a scaling with the connectivity parameter p and the system size N to leading order as 1/( p N ), whereas drift fields due to heterogeneity of leak potentials (and other heterogeneous single-neuron parameters) will scale as 1/N, both in accordance with earlier results [16, 36, 38, 46].
In addition to reproducing the previously known scaling with the system size N, our theory exposes the scaling of both drift and diffusion with the parameters τx, τu, and U of short-term depression and facilitation via the analytical pre-factors Ci/S appearing in Eqs (5) and (7). Our result extends the calculation of the diffusion constant [39] to synaptic dynamics with short-term plasticity: In the limiting case of no facilitation and depression (U → 1, τx → 0ms), the pre-factor reduces to Ci = 1 and the normalization factor simplifies to S static = τ s ∑ i ( d J 0 , i d φ) 2 ϕ 0 , i ′, where ϕ 0 , i ′ =d ϕ i d J i |J 0 , i is the derivative of the firing rate of neuron i at its steady-state input J0,i. For static synapses we thereby recover the known result for diffusion [39, Eq. S18], but also add an analogous relation for the drift A static ( φ ) = ( ∑ i d J 0 , i d φ Δ ϕ i ( φ ) ) / ( τ s ∑ i d J 0 , i d φ 2 ϕ 0 , i ′ ). Our approach relies on the existence of a stationary bump state (which is stable for large noise-free homogeneous networks), around which we calculate drift and diffusion as perturbations. Following earlier work [11, 50, 52], we use in our simulations with spiking integrate-and-fire neurons a slow synaptic time constant (τs = 100ms) as an approximation of recurrent (NMDA mediated) excitation. While our theory captures the effects of changing this time constant τs in the pre-factors Ci/S, we did not check in simulations whether the bump state remains stable and whether our theory remains valid for very short time constants for τs.
Finally, two limiting cases are worth highlighting. First, for strong facilitation (U → 0) we obtain pre-factors C i/S = ( 1 + 2 τ u ϕ 0 , i )( ∑ i ( d J 0 , i d φ) 2 ϕ 0 , i ′ [ τ s ( 1 + 2 τ u ϕ 0 , i ) + τ u 2 ϕ 0 , i ] ) − 1, indicating that (i) this limit will leave residual drift and diffusion which (ii) will both be controlled by the time constants for facilitation (τu) and synaptic transmission (τs), with no dependence upon depression. Second, for vanishing facilitation (U → 1 and τu → 0) we find that the normalization factor S will tend to zero if the depression time constant τx is increased to a finite value τx,c. Through the pre-factors Ci/S this, in turn, yields exploding diffusion and drift terms (see S8 Fig). While this is a general feature of bump systems with short-term depression, the exact value of the critical time constant τx,c depends on the firing rates and neural implementation of the bump state (see section 6 in S1 Text): for the spiking network investigated here, we find a critical time constant τx,c = 223.9ms (see S8 Fig). In networks with both facilitation and depression, the critical τx,c increases as facilitation becomes stronger (see S8 Fig).
To demonstrate the accuracy of our theory, we chose random connectivity as a first source of frozen variability. Random connectivity was realized in simulations by retaining only a random fraction 0 < p ≤ 1 (connection probability) of excitatory-to-excitatory (EE) connections. The uniform connections from and to inhibitory neurons are taken as all-to-all, since the effects of making these random or sparse would have only indirect effects on the dynamics of the bump center positions.
Our theory accurately predicts the drift-fields A(φ) (see Eq (7)) induced by frozen variability in networks with short-term plasticity (Fig 2). Briefly, for each neuron 0 ≤ i < N, we treat each realization of frozen variability as a perturbation Δi around the perfectly symmetric system and use an expansion to first order of the input-output relation F to calculate the resulting changes in firing rates (see Frozen noise for details):
Δ ϕ i ( φ ) = d F d Δ i Δ i . (8)
The resulting terms are then used in Eq (7) to predict the magnitude of the drift field A(φ) for any center position φ, which will, importantly, depend on STP parameters. The same approach can be used to predict drift fields induced by heterogeneous single neuron parameters [36] (see next sections) and additive noise on the E-E connection weights [16, 38].
We first simulated spiking networks with only short-term depression and without facilitation (Fig 2A, left, same network as in Fig 1B1), for one instantiation of random (p = 0.5) connectivity. Numerical estimates of the drift in spiking simulations (by measuring the displacement of bumps over time as a function of their position, see Spiking simulations in Materials and methods for details) yielded drift-fields in good agreement with the theoretical prediction (Fig 2B, left). At points where the drift field prediction crosses from positive to negative values (e.g. Fig 2B, left, φ = π 2), we expect stable fixed points of the center position dynamics in agreement with simulation results, which show trajectories converging to these points. Similarly, unstable fixed points (negative-to-positive crossings) can be seen to lead to a separation of trajectories (e.g. Fig 2A, left, φ = − π 2). In regions where the positional drifts are predicted to lie close to zero (e.g. Fig 2A, left φ = 0) the effects of diffusive dynamics are more pronounced. Finally, numerical integration of the full 1-dimensional Langevin equation Eq (4) with coefficients predicted by Eqs (5)–(7), produces trajectories with dynamics very similar to the full spiking network (Fig 2C, left). When comparing the center positions after 13.5s of delay activity between the full spiking simulation and the simple 1-dimensional Langevin system, we found very similar distributions of final positions (Fig 2D, left, compare to Fig 1B1, histogram). Our theory thus produces an accurate approximation of the dynamics of center positions in networks of spiking neurons with STP, thereby reducing the complex dynamics of the whole network to a simple equation. It should be noted that, in regions with strong drift or steep negative-to-positive crossings, the numerically estimated drift-fields deviate from the theory due to under-sampling of these regions as trajectories move quickly through them, yielding fewer data points. In Short-term plasticity controls drift we additionally show that the theory, as it relies on a linear expansion of the effects of heterogeneities on the neuronal firing rates, tends to generally over-predict drift-fields as heterogeneities become stronger.
Introducing strong short-term facilitation (U = 0.1) reduces the predicted drift fields (Fig 2B, left, dashed line), which resemble a scaled-down version of the drift-field for the unfacilitated case. We confirmed this theoretical prediction by simulations including facilitation (Fig 2A, right): the resulting drift fields show significant reduction of speeds (Fig 2B, right) while zero crossings remained similar to the unfacilitated network, similar to the results in [38]. Theoretical predictions of the drift fields with bump shapes extracted from these simulations again show an accurate prediction of the dynamics (Fig 2B, right). Thus, as before, forward integrating the simple 1-dimensional Langevin-dynamics yields trajectories (Fig 2C, right) highly similar to those of the full spiking network, with closely matching distributions of final positions (Fig 2D, right), indicative of a matching strength of diffusion. In summary, our theory predicts the effects of STP on the joint dynamics of diffusion and drift due to network heterogeneities, which we will show in detail in the next sections.
To isolate the effects of STP on diffusion, we simulated networks without frozen noise for various STP parameters. For each combination of parameters, we simulated 1000 repetitions of 13.5s delay activity (after cue offset) distributed across 20 uniformly spaced initial cue positions (see Fig 3A for an example). From these simulations, the strength of diffusion was estimated by measuring the growth of variance (over repetitions) of the distance of the center position from its initial position as a function of time (see Spiking simulations in Materials and methods for details). For all parameters considered, this growth was well fit by a linear function (e.g. Fig 3A, inset), the slope of which we compared to the theoretical prediction obtained from the diffusion strength B (Eq (5)).
We find that facilitation and depression control the amount of diffusion along the attractor manifold in an antagonistic fashion (Fig 3B and 3C). First, increasing facilitation by lowering the facilitation parameter U from its baseline U = 1 (no facilitation) towards U = 0, while keeping the depression time constant τx = 150ms fixed, decreases the measured diffusion strength over an order of magnitude (Fig 3B, dots). On the other hand, increasing the facilitation time constant τu from τu = 650ms to τu = 1000ms (Fig 3B, orange and blue dots, respectively) only slightly reduces diffusion. Our theory further predicts that increasing the facilitation time constants above τu = 1s will not lead to large reductions in the magnitude of diffusion (see S2 Fig). Second, we find that increasing the depression time constant τx for fixed U, thereby slowing down recovery from depression, leads to an increase of the measured diffusion (Fig 3C). More precisely, increasing the depression time constant from τx = 120ms to τx = 200ms leads only to slight increases in diffusion for strong facilitation (U = 0.1), but to a much larger increase for weak facilitation (U = 0.8).
For a comparison of these simulations with our theory, we used two different approaches. First, we estimated the diffusion strength by using the precise shape of the stable firing rate profile extracted separately for each network with different sets of parameters. This first comparison with simulations confirms that the theory closely describes the dependence of diffusion on short-term plasticity for each parameter set (Fig 3B, crosses). The observed effects could arise directly from changes in STP parameters for a fixed bump shape, or indirectly since STP parameters also influence the shape of the bump. To separate such direct and indirect effects, we used for a second comparison a theory with fixed bump shape, i.e. the bump shape measured in a “reference network” (U = 1, τx = 150ms) and extrapolated curves by changing only STP parameters in Eq (5). This leads to very similar predictions (Fig 3B, dashed lines) and supports the following conclusions: a) the diffusion to be expected in attractor networks with similar observable quantities (mainly, the bump shape) depends only on the short-term plasticity parameters; b) the bump shapes in the family of networks we have investigated are sufficiently similar to be approximated by measurement in a single reference network. It should be noted that the theory tends to slightly over-estimate the amount of diffusion, especially for small facilitation U (see Fig 3B and 3C left). This may be because slower bump movement decreases the firing irregularity of flank neurons, which deviates from the Poisson firing assumption of our theory (see Discussion). However, given the simplifying assumptions needed to derive the theory, the match to the spiking network is surprisingly accurate.
Having established that our theory is able to predict the effect of STP on diffusion, as well as drift for a single instantiation of random connectivity, we wondered how different sources of heterogeneity (frozen noise) would influence the drift of the bump. We considered two sources of heterogeneity: First, random connectivity as introduced above, and second, heterogeneity of the leak reversal potential parameters of excitatory neurons: leak reversal potentials of excitatory neurons are given by VL + ΔL, where ΔL is normally distributed with zero mean and standard deviation σL [36]. The resulting fields can be calculated by calculating the resulting perturbations to the firing rates of neurons by Eq (8) (see Frozen noise in Materials and methods for details).
The theory developed so far allowed us to predict drift-fields for a given realization of frozen noise, controlled by the noise parameters p (for random connectivity) and σL (for heterogeneous leak reversal-potentials) (see S3 Fig for a comparison of predicted drift fields to those measured in simulations for varying STP parameters and varying strengths of frozen noises). We wondered, whether we could take the level of abstraction of our theory one step further, by predicting the magnitude of drift fields from the frozen noise parameters only, independently of a specific realization. First, the expectation of drift fields under the distributions of the frozen noises vanishes for any given position: 〈A(φ)〉frozen = 0, where the expectation 〈.〉frozen is taken over both noise parameters. We thus turned to the expected squared magnitude of drift fields under the distributions of these parameters (see Squared field magnitude in Materials and methods for the derivation):
〈 A2 〉frozen=1S2∑iCi2((ϕ0,i′)2NE2(1p−1)∑j(s0,j)2(wijEE)2+(dϕ0,idΔiL)2σL2), (9)
where s0,j is the steady-state synaptic activation. Here, we introduced the derivatives of the input-output relation with respect to the noise sources that appear in Eq (8): ϕ 0 , i ′ = d F d J ( J 0 , i ( φ ) ) is the derivative with respect to the steady state synaptic input, and d ϕ 0 , i d Δ i L is the derivative with respect to the perturbation in the leak potential. In Squared field magnitude in Materials and Methods, we show that Eq (9) is independent of the center position φ, and can be estimated from simulations as the variance of the drift field across positions, averaged over an ensemble of network instantiations.
We defined the root of the expected squared magnitude of Eq (9) as the expected field magnitude:
⟨ A 2 ⟩ frozen . (10)
This quantity predicts the magnitude of the deviations of drift-fields from zero that are expected from the parameters that control the frozen noise—in analogy to the standard deviation for random variables, it predicts the standard deviation of the fields. To compare this quantity to simulations, we varied both heterogeneity parameters. First, the connectivity parameter p was varied between 0.25 and 1. Second, for heterogeneities in leak reversal-potentials, we chose values for the standard deviation σL of leak-reversal potentials between 0mV and 1.5mV, which lead to a similar range of drift magnitudes as those of randomly connected networks. For each combination of heterogeneities and STP parameters (networks had either random connections or heterogeneous leaks) we then realized 18–20 networks, for which we simulated 400 repetitions of 6.5s of delay activity each (20 uniformly spaced positions of the initial cue). We then estimated the drift-field numerically by recording displacements of bump centers along their trajectories (as in Fig 2A and 2B) and measured the standard deviation of the resulting fields across all positions.
Similar to the analysis of diffusion above, we find that facilitation and depression elicit antagonistic control over the magnitude of drift fields. In both simulations and theory, we find (Fig 4A and 4B) that the expected field magnitude decreases as the effect of facilitation is increased from unfacilitated networks (U = 1) through intermediate levels of facilitation (U = 0.4) to strongly facilitating networks (U = 0.1). Our theory predicts this effect surprisingly well, which we validated twofold (as for the diffusion magnitude). First, we used Eq (10) with all parameters and coefficients estimated from each spiking simulation separately (Fig 4A and 4B, plus-signs and crosses). Second, we extrapolated the theoretical prediction by using coefficients in Eq (9) from the unfacilitated reference network only (U = 1, τx = 150ms) but changed the facilitation and heterogeneity parameters (Fig 4A and 4B, dashed lines). The largest differences between the extrapolated and full theory are seen for U < 1 and randomly connected networks (p < 1), which we found to result from the fact that bump shapes for these networks tended to be slightly reduced under random and sparse connectivity (e.g. the top firing rate is reduced to ∼ 35Hz for U = 0.1, p = 0.25). Generally, as noise levels increase, our theory tends to over-estimate the squared magnitude of fields, since we rely on a linear expansion of perturbations to the firing rates to calculate fields (Eq (8)). Such deviations are expected as the magnitude of firing rate perturbations increases, and could be counter-acted by including higher-order terms. Since in the theory facilitation (and depression) only scales the firing rate perturbations (Eq (7)), these deviations can also be observed across facilitation parameters. Finally, we performed a similar analysis to investigate the effect of short-term depression on drift fields. Here, we varied the depression time constant τx for randomly connected networks with p = 0.6, by simulating networks with combinations of short-term plasticity parameters from U ∈ {0.1, 0.4, 0.8} and τx ∈ {120ms, 160ms, 200ms} (Fig 4C). We find that an increase of the depression time constant leads to increased magnitude of drift fields, which again is well predicted by our theory.
The theory developed in previous sections shows that diffusion and drift of the bump center φ are controlled antagonistically by short-term depression and facilitation. In a working memory setup, we can view the attractor dynamics as a noisy communication channel [56] that maps a set of initial positions φ(t = 0s) (time of the cue offset in the attractor network) to associated final positions φ(t = 6.5s), after a memory retention delay of 6.5s. We used the distributions of initial and (associated) final positions to investigate the combined impact of diffusion and drift on the retention of memories (Fig 5A). Because of diffusion, distributions of positions will widen over time, which degrades the ability to distinguish different initial positions of the bump center (Fig 5A, top). Additionally, directed drift of the dynamics will contract distributions of different initial positions around the same fixed points, making them essentially indistinguishable when read out (Fig 5A, bottom).
As a numerical measure of this ability of such systems to retain memories over the delay period, we turned to mutual information (MI), which provides a measure of the amount of information contained in the readout position about the initially encoded position [57, 58]. To measure MI from simulations (see Mutual information measure in Materials and methods), we analyzed network simulations for varying short-term facilitation parameters (U) and magnitudes of frozen noises (p and σL) (same data set as Fig 4A and 4B). We recorded the center positions encoded in the network at the time of cue-offset (t = 0) and after 6.5s of delay activity, and used binned histograms (100 bins) to calculate discrete probability distributions of initial (t = 0) and final positions (t = 6.5). For each trajectory simulated in networks of spiking integrate-and-fire neurons, we then generated a trajectory starting at the same initial position by using the Langevin equation Eq (4) that describes the drift and diffusion dynamics of center positions. The MI calculated from the resulting distributions of final positions (again at t = 6.5) for each network serve as the theoretical prediction for each network. As a reference, we used the spiking network without facilitation (U = 1, τu = 650ms, τx = 150ms) and no frozen noises (p = 1, σL = 0mV) and normalized the MI of all other networks (both for spiking simulations and theoretical predictions) with respect to the reference, yielding the measure of relative MI presented in Fig 5B–5E.
We found that the relative MI decreased compared to the reference network as network heterogeneities were introduced (Fig 5B, green). This was expected, since directed drift caused by heterogeneities leads to a loss of information about initial positions. There were two effects of increased short-term facilitation (by decreasing the parameter U). First, diffusion was reduced, which was visible in a vertical shift of the relative MI for facilitated networks (Fig 5A, orange and blue, at 0 heterogeneity). Second, the effects of frozen noise decreased with increasing facilitation, which was visible in the slopes of the MI decrease (see also S4 Fig). The MI obtained by integration of the Langevin equations (see above) matched those of the simulations well (Fig 5A, lines). From earlier results, we expected the drift-fields to be slightly over-estimated by the theory as the heterogeneity parameters increase (Fig 4), which would lead to an under-estimation of MI. We did observe this here, although for U = 1 the effect was slightly counter-balanced by the under-estimated level of diffusion (cf. Fig 3A, right), which we expected to increase the MI. For networks with stronger facilitation (U = 0.1), we systematically over-estimated diffusion (cf. Fig 3, left), and therefore under-estimated MI.
Using our theory, we were able to simplify the functional dependence between MI, short-term plasticity, and frozen noise. Combining the effects of both diffusion and drift into a single quantity for each network, we replaced the field A(φ) by our theoretical prediction 〈 A 2 〉 frozen in Eq (4) and forward integrated the differential equation for a time interval Δt = 1s, to arrive at the expected displacement in 1s:
| Δ φ | ( 1 s ) = ⟨ A 2 ⟩ frozen · 1 s + B · 1 s . (11)
This quantity describes the expected absolute value of displacement of center positions during 1s: it increases as a function of the frozen noise distribution parameters (Fig 5C), but even in the absence of frozen noise it is nonzero due to diffusion. Plotting the MI data in dependence of the first term only (〈 A 2 〉 frozen), shows that the MI curves collapse onto a single curve for each facilitation parameter (Fig 5D). Finally, plotting the MI data against |Δφ|(1s) we find that all data collapse on to nearly a single curve (Fig 5E). Thus, the effects of the two sources of frozen noise (corresponding to 〈A2〉frozen) and diffusion (corresponding to B) are unified into a single quantity |Δφ|(1s).
We performed the same analyses on a large set of network simulations with fixed random connectivity (p = 0.6) and varying STP parameters for both depression (τx) and facilitation (U) (same data set as in Fig 4C). Increasing the short-term depression time constant τx leads to decreased relative MI with a positive offset induced through stronger facilitation (Fig 6A, blue line). Calculating the expected displacement for these network configurations collapsed the data points mostly onto the same curve as earlier (Fig 6B). For strong depression combined with weak facilitation (τx = 200ms, U = 0.8), the drop-off of the relative MI saturates earlier, indicating that for these strongly diffusive networks the effect on MI may not be sufficiently captured by its relationship to |Δφ|(1s).
The abstraction of our theory condenses the complex dynamics of bump attractors in spiking integrate-and-fire networks into a high-level description of a few macroscopic features, which in turn allows matching the theory to behavioral experiments. Here, we demonstrate how such quantitative links could be established using two different features: 1) the sensitivity of the working memory circuit to distractors, and 2) the stability of working memory expressed by the expected displacement. We stress that our model is a simplified description of biological circuits, in which several further sources of variability and also dynamical processes influencing displacement should be expected (see Discussion). Thus, at the current level of simplification, the results presented in this section should be seen as proofs of principle rather than quantitative predictions for a cortical setting.
We presented a theory of drift and diffusion in continuous working memory models, exemplified on a one-dimensional ring attractor model. Our framework generalizes earlier approaches calculating the effects of fast noise by projection onto the attractor manifold [37, 39, 40] by including the effects of short-term plasticity (see [45] for a similar analysis for facilitation only). Our approach further extends earlier work on drift in continuous attractors with short-term plasticity [38] to include diffusion and the dynamics of short-term depression. Our theory predicts that facilitation makes continuous attractors robust against the influences of both dynamic noise (introduced by spiking variability) and frozen noise (introduced by biological variability) whereas depression has the opposite effect. We use this theory to provide, together with simulations, a novel quantitative analysis of the interaction of facilitation and depression with dynamic and frozen noise. We have confirmed the quantitative predictions of our theory in simulations of a ring-attractor implemented in a network model of spiking integrate-and-fire neurons with synaptic facilitation and depression, and found theory and simulation to be in good quantitative agreement.
In Section Short-term plasticity controls memory retention, we demonstrated the effects of STP on the information retained in continuous working memory. Using our theoretical predictions of drift and diffusion we were able to derive the expected displacement |Δφ| as a function of STP parameters and the frozen noise parameters, which provides a simple link between the resulting Langevin dynamics of bump centers and mutual information (MI) as a measure of working memory retention. Our results can be generalized in several directions. First, the choice of 1s of forward integrated time for |Δφ| (Eq (11)) was arbitrary. While a choice of ∼ 2s lets the curves in Fig 5E collapse slightly better, we chose 1s to avoid further heuristics. Second, we expect values of MI to decrease as the length of the delay period is increased. Our choice of 6.5s is comparable to delay periods often considered in behavioral experiments (usually 3-6s) [61, 66, 67]. However, a more rigorous link between the MI measure and the underlying attractor dynamics would be desirable. Indeed, for noisy channels governed by Fokker-Planck equations, this might be feasible [68], but goes beyond the scope of this work.
In Section Linking theory to experiments: Distractors and network size, we demonstrated that the high-level description of the microscopic dynamics obtained by our theory allows its parameters to be constrained by experiments. Considering that our model is a simplified description of its biological counterparts (see next paragraph), these demonstrations are to be seen as a proof of principle as opposed to quantitative predictions. However, since distractor inputs can be implemented in silico as well as in behavioral experiments (see e.g. [59]), they could eventually provide a quantitative link between continuous attractor models and working memory systems, by matching the resulting distraction curves. Our theory goes beyond previous models in which these distraction curves had to be extracted through repeated microscopic simulations for single parameter settings [47]. We further used our theory to derive bounds on network parameters, in particular the size of networks, that lead to “tolerable” levels of drift and diffusion in the simplified model. For large magnitudes of frozen noise our theory tends to over-estimate the expected magnitude of drift-fields slightly (cf. Fig 4). Thus, we expect the predictions made here to be upper bounds on network parameters needed to achieve a certain expected displacement. Finally, while the predictions of our theory might deviate from biological networks, they could be applied to accurately characterize the stability of, and the effects of inputs to, bump attractor networks implemented in neuromorphic hardware for robotics applications [69].
Our results show, that strong facilitation (small values of U) does not only slow down directed drift [38], but also efficiently suppresses diffusion in spiking continuous attractor models. However, in delayed response tasks involving saccades, that presumably involve continuous attractors in the prefrontal cortex [11, 22], one does observe an increase of variability in time [66]: quickly accumulating systematic errors (alike drift) [61] as well as more slowly increasing variable errors (with variability growing linear in time, alike diffusion) have been reported [60]. Indeed, there are several other possible sources of variability in cortical working memory circuits, which we did not consider here. In particular, we expect that heterogeneous STP parameters [62], noisy synaptic transmission and STP [70] or noisy recurrent weights [38] (see Random and heterogeneous connectivity in Materials and methods), for example, will induce further drift and diffusion beyond the effects discussed in this paper. Additionally, variable errors might be introduced elsewhere in the pathway between visual input and motor output (but see [71]) or by input from other noisy local circuits during the delay period [72]. Note that we excluded AMPA currents from the recurrent excitatory interactions [11]. However, since STP acts by presynaptic scaling of neurotransmitter release, it will act symmetrically on both AMPA and NMDA receptors so that an analytical approach similar to the one presented here is expected to work.
Several additional dynamical mechanisms might also influence the stability of continuous attractor working memory circuits. For example, intrinsic neuronal currents that modulate the neuronal excitability [47] or firing-rate adaptation [73] affect bump stability. These and other effects could be accommodated in our theoretical approach by including their linearized dynamics in the calculation of the projection vector (cf. Projection of dynamics onto the attractor manifold in Materials and methods). Fast corrective inhibitory feedback has also been shown to stabilize spatial working memory systems in balanced networks [74]. On the timescale of hours to days, homeostatic processes counteract the drift introduced by frozen noise [36]. Finally, inhibitory connections that are distance-dependent [11] and show short-term plasticity [75] could also influence bump dynamics.
We have focused here on ring-attractor models that obtain their stable firing-rate profile due to perfectly symmetric connectivity. Our approach can also be employed to analyze ring-attractor networks with short-term plasticity in which weights show (deterministic or stochastic) deviations from symmetry (see Frozen noise in Materials and methods for stochastic deviations). Although not investigated here, continuous line-attractors arising through a different weight-symmetry should be amenable to similar analyses [39]. Finally, it should be noted that adequate structuring of the recurrent connectivity can also positively affect the stability of continuous attractors [14]. For example, translational asymmetries included in the structured heterogeneity can break the continuous attractor into several isolated fixed points, which can lead to decreased diffusion along the attractor manifold [58].
We provided evidence that short-term synaptic plasticity controls the sensitivity of attractor networks to both fast diffusive and frozen noise. Control of short-term plasticity via neuromodulation [76] would thus represent an efficient “crank” for adapting the time scales of computations in such networks. For example, while cortical areas might be specialized to operate in certain temporal domains [7, 77], we show that increasing the strength of facilitation in a task-dependent fashion could yield slower and more stable dynamics, without changing the network connectivity. On the other hand, modulating the time scales of STP could provide higher flexibility in resetting facilitation-stabilized working memory systems to prepare them for new inputs [47], although there might be evidence for residual effects of facilitation between trials [45, 78]. By changing the properties of presynaptic calcium entry [79], inhibitory modulation mediated via GABAB and adenosine A1 receptors can lead to increased facilitatory components in rodent cerebellar [80] and avian auditory synapses [81]. Dopamine, serotonin and noradrenaline have all been shown to differentially modulate short-term depression (and facilitation when blocking GABA receptors) at sensorimotor synapses [82]. Interestingly, next to short-term facilitation on the timescale of seconds, other dynamic processes up-regulate recurrent excitatory synaptic connections in prefrontal cortex [62]: synaptic augmentation and post-tetanic potentiation operate on longer time scales (up to tens of seconds), and might be able to support working memory function [83]. While the long time scales of these processes might again render putative short-term memory networks inflexible, there is evidence that they might also be under tight neuromodulatory control [84]. Finally, any changes in recurrent STP properties of continuous attractors (without retuning networks as done here) will also lead to changes in the stable firing rate profiles, with further effects on their dynamical stability (see final section of the Discussion). This interplay of effects remains to be investigated in more detail.
Similar to an earlier theoretical approach using a simplified rate model [38], we find that the slowing of drift by facilitation depends mainly on the facilitation parameter U, while the time constant τu has a less pronounced effect. While the approach of [38] relied on the projection of frozen noise onto the derivative of the first spatial Fourier mode of the bump shape along the ring, here we reproduce and extend this result (1) for arbitrary neuronal input-output relations and (2) a more detailed spatial projection that involves the full synaptic dynamics and the bump shape. While, our theory can also accommodate noisy recurrent connection weights as frozen noise, as used in in [38] (see Frozen noise in Materials and methods for derivations), the drifts generated by these heterogeneities were generally small compared to diffusion and the other sources of heterogeneity.
A second study investigated short-term facilitation and showed that it reduces drift and diffusion in a spiking network, for a fixed setting of U (although the model of short-term facilitation differs slightly from the one employed here) [47]. Contrary to what we find here, these authors find that an increase in τu leads to increased diffusion, while we find that an increase over the range they investigated (∼ 0.5s − 4s) would decrease the diffusion by a factor of nearly two. More precisely, for our shape of the bump state (which we keep fixed) we predict a reduction from ∼ 26 to ∼ 16 deg2/s for a similar setting of facilitation U. These differences might arise from an increasing width of the bump attractor profile for growing facilitation time constants in [47], which would then lead to increased diffusion in our model. Whether this effect persists under the two-equation model of saturating NMDA synapses used there remains to be investigated. Finally, increasing the time constant of recurrent NMDA conductances has been shown to also reduce diffusion [47], in agreement with our theory, according to which the normalization constant S increases with τs [39].
A study performed in parallel to ours [45] used a similar theoretical approach to calculate diffusion with short-term facilitation in a rate-based model with external additive noise, but did not compare the results for varying facilitation parameters. The authors report a short initial transient of stronger diffusion as synapses facilitate, followed by weaker diffusion that is dictated by the fully facilitated synapses. Our theory, by assuming all synaptic variables to be at steady-state, disregards the initial strong phase of diffusion. We also disregarded such initial transients when comparing to simulations (see Numerical methods).
In a study that investigated only a single parameter value for depression (τx = 160ms, no facilitation) in a network of spiking integrate-and-fire neurons similar to the one investigated here, the authors observed no apparent effect of short-term depression on the stability of the bump [44]. In contrast, we find that stronger short-term depression will indeed increase both diffusion and directed drift along the attractor. Our result agrees qualitatively with earlier studies in rate models, which showed that synaptic depression, similar to neuronal adaptation [10, 85], can induce movement of bump attractors [42, 43, 86, 87]. In particular, simple rate models exhibit a regime where the bump state moves with constant speed along the attractor manifold [42]. We did not find any such directed movement in our networks, which could be due to fast spiking noise which is able to cancel directed bump movement [85].
The coefficients of Eq (4) give clear predictions as to how drift and diffusion will depend on the shape of the bump state and the neural transfer function F. The relation is not trivial, since the pre-factors Ci and the normalization constant S also depend on the bump shape. For the diffusion strength Eq (5), we explored this relation numerically, by artificially varying the shape of the firing rate profile (while extrapolating other quantities). Although a more thorough analysis remains to be performed, a preliminary analysis shows (see S6 Fig) that diffusion increases both with bump width and top firing rate, consistent with earlier findings [11, 32].
Our theory can be used to predict the shape and effect of drift fields that are generated by localized external inputs due to distractor inputs; see Section Linking theory to experiments: Distractors and network size. Any localized external input (excitatory or inhibitory) will cause a deviation Δϕi from the steady-state firing rates, which, in turn, generates a drift field by Eq (7). This could predict the strength and location of external inputs that are needed to induce continuous shifts of the bump center at given speeds, for example when these attractor networks are designed to track external inputs (see e.g. [10, 88]). It should be noted that in our simple approximation of this distractor scheme, we assume the system to remain at approximately steady-state, i.e. that the bump shape is unaffected by the additional external input, except for a shift of the center position. For example, we expect additional feedback inhibition (through the increased firing of excitatory neurons caused by the distractor input) to decrease bump firing rates. A more in depth study and comparison to simulations will be left for further work.
Our networks of spiking integrate-and-fire neurons are tuned to display balanced inhibition and excitation in the inhibition dominated uniform state [53, 89], while the bump state relies on positive currents, mediated through strong recurrent excitatory connections (cf. [44] for an analysis). Similar to other spiking network models of this class, this mean–driven bump state shows relatively low variability of neuronal inter-spike-intervals of neurons in the bump center [90, 91] (see also next paragraph). Nevertheless, neurons at the flanks of the bump still display variable firing, with statistics close to that expected of spike trains with Poisson statistics (see S7 Fig), which may be because the flank’s position slightly jitters. Since the non-zero contributions to the diffusion strength are constrained to these flanks (cf. Fig 1D), the simple theoretical assumption of Poisson statistics of neuronal firing still matches the spiking network quite well. As discussed in Short-term plasticity controls diffusion, we find that our theory over-estimates the diffusion as bump movement slows down for small values of U—this may be due to a decrease in firing irregularity in stable bumps in particular in the flank neurons, at which the Poisson assumption becomes inaccurate.
More recent bump attractor approaches allow networks to perform working memory function with a high firing variability also during the delay period [3], in better agreement with experimental evidence [92]. These networks show bi-stability, where both stable states show balanced excitation and inhibition [90] and the higher self-sustained activity in the delay activity is evoked by an increase in fluctuations of the input currents (noise-driven) rather than an increase in the mean input [93]. This was also reported for a ring-attractor network (with distance-dependent connections between all populations), where facilitation and depression are crucial for irregularity of neuronal activity in the self-sustained state [46]. Application of our approach to these setups is left for future work.
For the following, we define a concatenated 3 ⋅ N dimensional column vector of state variables y = (sT, uT, xT)T of the system Eq (3). Given a (numerical) solution of the stable firing rate profile ϕ → 0 we can calculate the stable fixed point of this system by setting the l.h.s. of Eq (3) to zero. This yields steady-state solutions for the synaptic activations, facilitation and depression variables y0 = (s0, u0, x0):
s 0 , i = τ s u 0 , i x 0 , i ϕ 0 , i , u 0 , i = U 1 + τ u ϕ 0 , i 1 + U τ u ϕ 0 , i , x 0 , i = 1 + U τ u ϕ 0 , i 1 + U ( τ u ϕ 0 , i + τ u τ x ϕ 0 , i 2 + τ x ϕ 0 , i ) . (12)
We then linearize the system Eq (3) at the fixed point y0, introducing a change of variables consisting of perturbations around the fixed point: y = y0 + δ y = y0 + (δ sT, δ uT, δ xT) and ϕi = ϕ0,i + δϕi. To reach a self-consistent linear system, we further assume a separation of time scales between the neuronal dynamics and the synaptic variables, in that the neuronal firing rate changes as an immediate function of the (slow) input. This allows replacing δϕi=dϕidJi|J0,i∑jdJidsjδsj=ϕ0,i′∑jwijδsj, where we introduce the shorthand ϕ0,i′≡dϕidJi|J0,i. Finally, keeping only linear orders in all perturbations, we arrive at the linearized system equivalent of Eq (3):
δ y ˙ =( − 1 τ s I + D ( u 0 · x 0 · ϕ → 0 ′ ) W D ( ϕ → 0 · x 0 ) D ( ϕ → 0 · u 0 ) U D ( ( 1 − u 0 ) · ϕ → 0 ′ ) W − 1 τ u I − U D ( ϕ → 0 ) 0 − D ( u 0 · x 0 · ϕ → 0 ′ ) W − D ( x 0 · ϕ → 0 ) − 1 τ x I − D ( ϕ → 0 · u 0 ) ) δ y ≡K δ y (13)
Here, dots between vectors indicate element-wise multiplication, the operator D : R n → R n × n creates diagonal matrices from vectors, and W = (wij) is the synaptic weight matrix of the network.
Spiking simulations are based on a variation of a popular ring-attractor model of visuospatial working memory of [11] (and used with variations in [27, 29, 32, 36, 47]). The recurrent excitatory connections of the original network model have been simplified, to allow for faster simulation as well as analytical derivations of the recurrent synaptic activation. The implementation details are given below, however the major changes are: 1) all recurrent excitatory conductances are voltage independent; 2) a model of synaptic short-term plasticity via facilitation and depression [49, 94, 95] is used to dynamically regulate the weights of the incoming spike-trains 3) recurrent excitatory conductances are computed as linear filters of the weighted incoming spike trains instead of the second-order kinetics for NMDA saturation used in [11].
|
10.1371/journal.pcbi.1006769 | Correlation Structure in Micro-ECoG Recordings is Described by Spatially Coherent Components | Electrocorticography (ECoG) is becoming more prevalent due to improvements in fabrication and recording technology as well as its ease of implantation compared to intracortical electrophysiology, larger cortical coverage, and potential advantages for use in long term chronic implantation. Given the flexibility in the design of ECoG grids, which is only increasing, it remains an open question what geometry of the electrodes is optimal for an application. Conductive polymer, PEDOT:PSS, coated microelectrodes have an advantage that they can be made very small without losing low impedance. This makes them suitable for evaluating the required granularity of ECoG recording in humans and experimental animals. We used two-dimensional (2D) micro-ECoG grids to record intra-operatively in humans and during acute implantations in mouse with separation distance between neighboring electrodes (i.e., pitch) of 0.4 mm and 0.2/0.25 mm respectively. To assess the spatial properties of the signals, we used the average correlation between electrodes as a function of the pitch. In agreement with prior studies, we find a strong frequency dependence in the spatial scale of correlation. By applying independent component analysis (ICA), we find that the spatial pattern of correlation is largely due to contributions from multiple spatially extended, time-locked sources present at any given time. Our analysis indicates the presence of spatially structured activity down to the sub-millimeter spatial scale in ECoG despite the effects of volume conduction, justifying the use of dense micro-ECoG grids.
| Electrocorticography (ECoG) is a type of electrophysiological monitoring that uses electrodes placed directly on the exposed surface of the brain. ECoG is a promising technique for studying the brain, and EcoG signals can be used to control brain-computer interfaces. Advances have made it possible to record simultaneously with an increasing number of smaller, and more closely spaced electrodes. However, a property of electrical recording from outside the brain is that common signals appear on different electrodes at different locations, and this affects decisions about how to best distribute a limited number of electrodes to maximize the information that can be gathered. Large spacing of electrodes around one centimeter apart on the brain’s surface has proven useful for clinical and research use, but how much benefit there is to recording from more locations in a smaller area remains to be answered. We found that we can explain the commonality between the different locations as the combination of different patterns of brain activity that are present at multiple electrode locations, and that signals recorded from very closely spaced electrodes, around a millimeter or less apart, are able to identify patterns that are at this small scale.
| Electrical recording from the brain surface, known as electrocorticography (ECoG), is becoming more common due to technological advances that enable recording from large cortical surface area with high temporal resolution and far better spatial resolution than non-invasive EEG [1, 2]. ECoG has also been used as an alternative to penetrating intracortical recording electrodes in brain-computer-interface (BCI) applications [3–9] due to its less invasive nature and long-term stability that are important features for driving drive BCI’s [3]. Electrodes designed for BCI will typically have more closely spaced electrodes to target specific cortical regions compared to clinical ECoG, in which large cortical coverage is important. Recently, high density ECoG grids have become more common, and many questions on the properties, uses, and design of these grids, e.g., what is the optimal spacing for the electrodes [8–11], remain to be answered.
Recording hardware sets an upper limit on the number of channels that can be simultaneously recorded. This creates a tradeoff in designing ECoG electrode grids between coverage and resolution. Clinical grids are typically on the larger coverage side, with 1 centimeter being a typical pitch between electrodes. BCI and research applications have pushed for more resolution in order to place more electrodes near cortical regions of interest [3, 6, 8–10, 12–17]. A challenge of scaling down the size of ECoG grids is that low impedance electrodes improve signal quality, but electrode impedances increase as the contact area decreases [18]. Combining fabrication techniques that allow for smaller, more closely spaced ECoG contacts with novel materials that can significantly reduce impedance makes very small contact sizes feasible. In the present study, we used electrodes coated with PEDOT:PSS on gold traces embedded in a parylene-C substrate [19, 20] with contact diameter as small as 20 microns and pitches as low as 200 microns. Hereafter, we will refer to these ECoG electrode grids as micro-ECoG. Previous work has shown that recordings with micro-ECoG electrodes are more similar to intracortical recordings than to the recordings from larger clinical ECoG electrode grids [14].
The primary signal of interest in ECoG recordings is the lower frequency component (less than ~200–500 Hz) called the local field potential or LFP. LFP is an uncertain signal in that its precise physiological and spatial origins are poorly understood [21, 22]. Much of the difficulty both in studying and using LFP is due to its lack of spatial specificity, that is, the potentials are an aggregate of nearby activity—in contrast to single- or multi-unit electrophysiological signals which are indicative of action potentials very near the recording site [23]. The spatial extent of LFP is itself an area of study [24–26] with a dependence on the geometry and activity of the region generating the signal. The spread of the potentials manifests itself in ECoG as similar signals being recorded on different electrodes. This feature of the potentials at different electrodes can be used to interpret the signals [27, 28] or guide the design of electrode grids to optimally sample the cortical surface.
To quantify the similarity between electrodes, previous studies examined the correlation or coherence of EEG, ECoG, and intracortical electrodes as a function of inter-electrode distance by averaging the correlation or coherence across many pairs separated by the same distance [8, 13–15, 28–31]. Most of these studies are in human or nonhuman primate, with some early investigations on smaller mammals, reptiles, and invertebrates. In ECoG recordings these studies have shown a consistent nearly monotonic decrease in the correlation as the electrode separation increases that exhibits a roughly exponential shape. Also consistent across the studies is a dependence of the correlation/coherence on the frequency band examined.
It is expected that the correlation/coherence would tend to zero (or bias level) at large distances, and this can be seen in EEG and clinical ECoG [14, 30, 31]. On the other hand, the correlation should approach 1 as the separation approaches 0. This is the case because the brain is a conductive medium, and for finite sources distant from the electrode in a volume conductor the potential will be the sum of all of the sources with amplitudes attenuated with distance. The distance at which the similarity will effectively approach 1 will depend on both the geometry of the sources and the properties of the medium. An example of this is discussed in [29, 32], where ECoG correlation between submillimeter-spaced electrodes is mostly close to 1 while correlation between even more closely spaced intracortical electrode pairs is frequently much lower than 1. This sub-millimeter regime in ECoG is largely unexplored, and at the smallest distances in most previous studies the correlation or coherence is significantly below 1, meaning there is still room to explore smaller electrode spacing.
On the other hand, for practical purposes recording ECoG at the scale in which the neighboring pairs measure very close to the same signal is not optimal because this would mean a large amount of redundancy between channels. The spacing should be guided by the spatial extent of features of interest. It has been suggested that contacts should be less than 5 mm apart for adequate sampling of gamma band in human ECoG [12], that for BMI applications subdural electrodes in humans be spaced 1.7 mm apart and in rat 0.6 mm apart [9], and that by halving the spacing of electrodes from 3.5 to 1.68 mm implanted in minipig, more and separate response peaks could reliably be identified [10]. The optimal separation will depend on factors such as species, location, and the nature of the activity of interest, but in general it will be difficult experimentally to recognize the optimal spatial resolution for a specific application until it is exceeded. However, we expect there may be an approximate resolution to use as a rule of thumb for each species.
We analyze the similarity of micro-ECoG with inter-electrode spacings down to 0.4 mm in human recordings and 0.2 mm in mouse. In agreement with past studies, we found that on average the signals were more similar for more closely spaced electrodes. With exceptions, higher frequency components of the signal showed a larger decrease in similarity with distance.
We also investigated the nature of the pairwise correlation between electrodes across the electrode grid. For a group of closely spaced electrodes to be correlated to each other there must be parts of their signals that are common between each channel pair, and parts that are independent to each electrode. The relative size, distribution, and properties of these signals determines the correlation between each pair of electrodes across the whole array. There are several analyses that are tailored to finding common signals across multiple channels commonly used on electrophysiological data such as principal component analysis (PCA), factor analysis, or independent component analysis (ICA). We modeled the effect of the properties of the components on the correlation structure, and then used ICA on the data to identify and separate common signals and find how they are distributed across the grid. We found that there are smoothly distributed sources present in the data, and due to the linearity of the ICA decomposition, show that the spatially coherent ICA components account for much of the correlation structure in the data.
Human subjects (n = 2) were implanted with a grid of 56 electrodes with 0.4 mm center-to-center distance referenced to another larger (3 mm diameter) ECoG electrode a few centimeters away, and mice (n = 2) were implanted with 32-electrode square grids with 0.2 mm or 0.25 mm spacing with a subcutaneous reference near the skull. After removing poor channels and potential artifacts, the signals were bandpass filtered into 6 different bands, and separated into non-overlapping 2.0 second windows (527 windows for subject 1, 326 for subject 2, 1,486 for mouse 1, and 893 for mouse 2). The Pearson correlation coefficient (referred to as simply correlation) was calculated for each window between all pairs of channels on the ECoG grids for each filter. After averaging correlations across equidistant electrode pairs as in [26] which we will call distance-averaged correlation (DAC), we see that the correlation between pairs of channels decreases on average as the distance between the electrodes increases (Fig 1). The values of correlation in Fig 1 are averages of correlation calculated in 2 second segments of the data across all segments and channel pairs that share the same spacing. We find similar values to previous studies of the correlation as a function of electrode distance [14, 15, 30]. Correlation was analyzed separately for different frequency bands due to the 1/f nature of the signal power, and that the presence of distinct processes present in different bands are common in electrophysiology studies. Also, it is a well-known feature of ECoG that common activity in lower frequencies tends to appear over larger regions than high frequency activity. We find a similar trend in the correlation plots with some exceptions between adjacent frequency bands, and that between the two commonly used bands in electrophysiology studies, beta (15–30 Hz), and high gamma (70–110+ Hz), the difference is quite large as expected.
The differences between the results in human and mouse is also large (Fig 1B inset). In mouse the correlations fall below 0.5 within 1.5 mm while in human even the highest frequencies are correlated above 0.5 up to around 2 mm. The low frequencies in human are noticeably more correlated across distances of a few millimeters. The values within the 2.0 second time windows vary considerably but are concentrated near the mean values (Fig 1C and 1D).
Correlation is a measure of to what degree two signals are similar, but an alternative approach to view the similarity is to look for commonality among all the signals simultaneously rather than an aggregate of pair-wise comparisons. There are a few methods which are commonly used in electrophysiology for identifying common signals that are present on multiple channels: principal component analysis (PCA), factor analysis, and independent component analysis (ICA). All are built around the assumption that there exists a set of signals that are present in the data with various amplitudes across all of the channels. ICA and factor analysis were developed to find underlying signals while PCA was not. Factor analysis assumes the components are drawn from a Gaussian distribution (across time samples, not channels), which does not describe the data, especially the sinusoidal signals obtained after bandpass filtering. We used ICA because we expect it to best find the underlying signals, and it is commonly used for this purpose. An important point in using ICA (as well as PCA and FA) is that the geometry of the recordings is not an input to the algorithm. The inputs are a set of time series (in this case) with no particular ordering, arrangement, or any other information relating the channels to one another. Therefore, an orderly spatial arrangement of the ICA results has been taken as an indication of the efficacy of ICA in separating sources, and is compelling in many cases.
To explore the connection between the ICA/PCA decomposition of the data into components and the correlation we start with a model how the spatial extent of the components affects the DAC. Component weights are modeled as two-dimensional Gaussian distributions sampled on a square grid of “electrodes”. The resulting correlation vs. distance curves are well-approximated by Gaussians, and we find a direct correlation between the width of the component Gaussians and the standard deviation parameter of the fit of the correlation vs. distance curve as shown in Fig 2. The relationship between the two is linear in the limit that there are many components that are sufficiently sampled by the grid of electrodes. The addition of uncorrelated noise to all channels decreases all correlation values by a factor, and the effect of having a reference signal added to every channel is to increase all correlation values. The effect of the noise is more apparent at small distances where even with the Gaussian components, the apparent y-intercept of the DAC drops as noise is added. On the other hand, the reference has the effect of raising the asymptotic value of the DAC for large distances.
We also directly connected the components to the DAC through the weight matrices (mixing matrices in ICA). Applying the ICA unmixing matrix (the inverse of the mixing matrix) to the data will decorrelate the data, and the unmixing matrix can be arbitrarily rescaled, therefore it is always possible for all of the components to have unit variance. This makes the ICA unmixing matrix a whitening, or sphering, matrix which is straightforwardly connected to the covariance matrix of the data because when multiplied by its transpose it must equal the covariance matrix. This allows a reduced, single-component covariance matrix to be calculated for each component of the mixing matrix separately, and the contribution of each component to the DAC can then be calculated.
ICA is applied (using the EEGlab implementation, see Methods) to the same filtered 2 second windows as were used in the DAC calculation. The mixing matrices, when plotted in the arrangement of the electrodes, show readily apparent spatial patterns throughout the recordings as shown in Fig 3. As a way to quantify the spatial patterns in the component weights in the mixing matrices, we fit the weights as they are laid out on the brain to a circular two-dimensional Gaussian function. The goodness-of-fit gives a rough assessment of the smoothness of the mapped weights and their spatial gradients (see Fig 4). Of the parameters of the fit, the one with we expect to have the most relevance to the correlation is the standard deviation, or width, of the Gaussian. Larger widths would correspond to larger correlated areas, and as a result, a higher correlation at larger distances. This effect can be seen when comparing the median width values for each frequency band independently. As shown In Fig 5, the median Gaussian width decreases with frequency in agreement with the frequency dependence of the DAC, and that the components tend to be more Gaussian for lower frequencies. At the highest frequencies there is a marked decrease in the goodness of fit that may be due to lower signal-to-noise ratios expected as the 1/f decrease in the signal approaches the noise floor of the hardware.
By comparing the contributions of each ICA component to its Gaussian fit we find if and how the DAC curves are influenced by the spatial distribution of the component weights. The spatial distribution of the weights must explain the DAC curves, but to determine whether the Gaussian fits have any significance for the DAC in real data, the contribution of the components is plotted as a function of the R2 values of their fits in Fig 6. The higher R2 components account for a disproportionately large amount of the drop in the DAC with distance, and therefore the particular shape of the DAC is mostly attributable to the more Gaussian components. The correlation and variance are concentrated in more Gaussian components which shows that the larger, more significant components are generally roughly Gaussian (Supplement Fig 1).
As a control, PCA is substituted for ICA, and because it can also be rescaled into a whitening matrix, all of the analyses can be carried out in the same manner for PCA as for ICA. PCA is not a source separation algorithm like ICA, but in the case where there are sources that account for most of the variation these sources will show up in the first principal components. This can be seen in the R2 histograms (Supplement S2 Fig) in where the components are concentrated near 0, but there is also a smaller number of components very near 1.0 that account for much of the variance. These are the first few PCA components which are larger and more Gaussian. These more spatially Gaussian principal components account for much of the drop in correlation, so using PCA for comparison both shows the effectiveness of ICA in finding local sources, and that again, the more Gaussian components are more strongly tied to the drop in the DAC with distance.
The location of the reference electrode can have a large impact on the correlation values as shown in Fig 2D [31, 33–35]. The reference electrode subtracts the same signal from all of the recording channels and this will act to increase the correlation between any two channels if the reference is sufficiently uncorrelated with the signals. In mouse the reference electrode was placed subcutaneously and not on the skull, while in human the reference was multiple contacts within 10 cm of the of the micro-ECoG grid on the cortical surface. The latter are more likely to be active at the frequencies of interest, and even correlated with the unipolar signals measured at the grid. The reference electrode placement should always be taken into account when interpreting correlation or other measures of signal similarity, and that the relatively close reference used in the human subjects is not an ideal placement for studying DAC. Consequently, the DAC curves we obtained in Fig 1 should not be interpreted as the DAC corresponding to unipolar potentials (potentials measured against a theoretical reference potential of zero) which would be the ideal for studying spatial correlation across the brain. Using the methods in [36] we attempted to identify the reference signal in the human recordings, but a signal that matched the criteria was not found. Additionally, the reference will ideally be identified by ICA as a component with the same weight on every channel across the grid. In practice this is unlikely, but it may be identified in part and represented by components with relatively flat weights. In fact, this method was used in [37]. In our case it may be correlated with the unipolar surface potentials at the grid and could be mixed in with components of those.
A common method in ECoG to remove the true reference is to use the common average reference (CAR), at the expense of introducing a virtual reference which is also unknown. When ICA is applied after CAR, and fits are recalculated the distribution of R2 values is nearly unchanged (supplemental S4 Fig). While the mean Gaussian widths are significantly different after CAR (two-sample Kolmogorov-Smirnov test p > 0.05), they follow the unreferenced values closely but are slightly larger (19 +/- 9%). This shows that the references that were used did not have a large effect on the components and their spatial properties. The effect of CAR on the correlation is shown in supplemental S3 Fig, and the large difference between the original and CAR correlations can be understood through the effect of the CAR matrix discussed in the Methods. The amount subtracted from each component is uniform across all the channels and is equal to the average weight. Therefore, the shapes of the components are unchanged, but they are shifted such that they mean weight of each component across channels is zero. Reference effects are removed in this way due to their representation as a uniform component across all channels. This shifting of the weights can be seen in the data through the offset term of the Gaussian fit becoming strictly negative after CAR.
The results of our study indicate high degree of variability of the DAC, both within any set of data, as Fig 1 shows, and between datasets due to external factors such as where on the cortex the electrodes are placed. There is large variation across the 2.0 second windows as shown in Fig 1, that may reflect changes in ongoing activity. In fact, it has been shown that there are task-related changes in the DAC [13, 15, 38], however we did not find there to be task- or state-related changes in the DAC in the human recordings.
The particular curve of the DAC may change between time windows, recording epochs, subjects, and species, but a robust feature in our recordings and previous studies is the frequency dependence of the spatial correlation. This agrees with past studies that the responses in lower frequency bands are more spread out than in higher bands [13, 17, 26, 32, 39] and is evident in similar studies that used coherence instead of correlation, which is an inherently frequency dependent similarity metric. The coherence is plotted as a function of frequency, and in for ECoG data almost completely monotonically decreases with frequency [13–15].
The geometry of the ICA component weights offers a possible explanation of this frequency dependence. Two ways by which neighboring electrodes can be correlated are by volume conduction and coactivation of populations close to each electrode which produce distinct, but correlated potentials [40, 41]. In many cases there is a degree of both which contributes to the correlation, but for large distances where volume conduction is assumed to be negligible the presence of correlation is used as an indication of connectivity [28]. On the other hand, at the submillimeter scale we expect volume conduction may play a larger role. The presence of a single-peaked, radially symmetric, and smooth ICA weights map is consistent with volume conduction of the potentials, and the Gaussian fits show that many of the components fit this description. Coactivating regions could also be described by this shape but are not limited to it; there could be distinct, separated peaks, plateaus, or checkerboard-patterned regions.
The large and consistent difference in the DAC between human and mouse can also be explained by either larger coactivated areas of cortex or a larger effect of volume conduction in human cortex. The spread of a signal due to volume conduction in brain tissue would be the same in either species assuming they have similar conductivities, but human cortex has neurons with larger lateral spread of their dendritic and axonal trees and is roughly twice as thick as mouse cortex. The effect of volume conduction on deeper sources will spread the potentials they cause more widely across the cortical surface. Additionally, the size of functionally distinct cortical regions is larger in humans, and we are again left with the ambiguity between the two possible factors: the size of the correlated activity, and its spatial spread as in [26].
In our analysis the choice of ICA as the particular form of whitening and 2 dimensional Gaussians as the function used for fitting are not the only choices that could have been used, but they were chosen for simplicity and applicability to this analysis. As a fitting function, a 2-dimensional Gaussian was chosen for its simplicity and flexibility, and not due to any assumption that the components would take this particular form. The purpose of the fit is to identify smoothly varying component weights. The function is smooth on a small scale, with the only peak being the center, so that neighbor-to-neighbor oscillations in the weights will degrade the fit. It is also able to describe radially symmetric peaked distributions as well as flat linear gradients by being fit to a very wide Gaussian with its center far from the electrode grid. Alternatives more tailored to quantify the smoothness could be used such as taking the second spatial derivative of the component weights and finding smooth gradients and peaks by taking first spatial derivatives. These are harder to implement and interpret, and the high R2 values when using ICA and low ones from PCA show the fitting approach is able to describe the components while not being so flexible that it can be fit to any component weights.
Another reason for using the Gaussian fits is that they provide a single parameter that characterizes the size of the regions that contain the components. The mean width of the fits of each frequency band decreases with increasing frequency, and this may be due to the effect that frequency has on the spatial spread of LFP. Additionally, the fits provide another method of removing distant volume conducted activity similar to the ICA approach used in [37] by using both the center location of the Gaussian along with the width to identify components of the signals that are far from the grid location. This kind of ICA-based method as an alternative to standard re-referencing schemes has been proposed in [42].
On the other hand, the DAC curves are not fit to Gaussians for the data despite the modeling results that showed that Gaussian components have Gaussian DACs. We expect that using the same model, but with other peaked, but not necessarily radially symmetric distributions, will still result in monotonically decreasing DAC curves with a different shape. Additionally, the effect of noise and reference will add a predictable modification to the curve but adds additional unknown parameters. These may be estimated but this is confounded by the unknown effect of the actual non-Gaussian shape of the components, and the fit becomes more difficult to implement and interpret.
The component mixing matrix weights analysis requires any whitening matrix to separate the components, but ICA was chosen for this purpose. Commonly used whitening transformations are not intended to perform source separation, but ICA can be both a whitening and a blind source separation algorithm. PCA and factor analysis have been used for finding common sources in the data, but the assumptions about the data of ICA are more well suited to finding sources in electrophysiology, hence its popularity. Factor analysis is designed to find similar localized sources but is not easily modified to be a whitening transformation and its assumption of normally distributed components is incompatible with the sinusoidal nature of narrow bandpass filtered signals.
There are drawbacks to calculating ICA in separate frequency bands, as any broadband processes or ones that span frequency ranges between or across multiple bands will not be as accurately identified or be recognized as part of the same component. Still, ICA was calculated by frequency band so as to be calculated on the exact same windows and signals as the correlation, and due to the 1/f power spectrum typical of electrophysiology. The first consideration is necessary specifically for linking the correlation and the component mixing matrix through the covariance matrix, while the second is a general problem in applying ICA to LFP. ICA may less accurately separate sources whose power is concentrated in higher frequencies due to the much larger power present in lower frequencies biasing ICA towards identifying sources concentrated in those. In addition, if PCA is applied as a pre-processing step, the components that contain some high frequency sources may even be discarded.
ICA was applied only to small time windows in addition to narrow frequency bands. This has similar drawbacks in terms of the effectiveness of ICA because it limits the number of observations which ICA can use to identify source. For the same reason as before, consistency with the segments analyzed for correlation, the 2 second windows are needed. Also, this length of time may be appropriate because the components were found to vary even between adjacent windows. However, there is some consistency in the ICA mixing matrices across time—that is very similar components show up repeatedly, but not consistently. This suggests that the time scale of the duration of stable components may be 2 seconds or less, and perhaps ICA would be better suited to even shorter windows for this data in future work to avoid temporal fluctuations in source strengths which does not fit the assumption in ICA of time-invariant mixing matrices.
The curve generated by averaging the correlation over many contacts offers some guidance as to how the signals will be related for a given electrode spacing, but it is more straightforward to choose a spacing when given a measure of the spatial extent of the activity. The two are linked, and as has been shown previously with the correlation, the frequency has a strong effect on the spread of potentials measured at the surface of the brain. Electrodes spaced less than a millimeter apart are more suited to higher frequencies or to smaller animals than humans, but even with very limited cortical coverage volume conduction still allows activity that is not directly under the grid to be recorded.
All human subject research was conducted in accordance with a study protocol that was reviewed and approved by the UC San Diego Health Institutional Review Board (protocol #121090). All animal work procedures were in accordance with a protocol approved by the Institutional Animal Care and Use Committees of UC San Diego (protocol S07360). Anesthetics used in mice were isoflurane and alpha-chloralose. Pentobarbital injection was used as the method of euthanasia.
The details of the electrodes, their preparation, their implantation, and the recording setup are given in [18–20, 43]. Subjects who were undergoing awake craniotomy surgeries were chosen for recording. The entire section of hardware up to the amplifiers had to be sterilized due to its proximity to the surgical field. The electronics underwent STERRAD sterilization, and it was important to ensure that the devices would remain intact after autoclave sterilization [43]. The electrode grid was placed over STG, with larger electrodes within a few centimeters of the grid used as reference electrodes. The ground electrode was placed in the scalp. Recording was sampled at 20 kS/s, with a built-in high pass filter at 0.1 Hz and low pass filter at 7500 Hz.
ICR mice weighing 25–35 g were used in the experiments. The mice were placed on a heating pad and anesthetized with isoflurane. A femoral artery was catheterized for monitoring and injection, and a tracheotomy was performed for ventilation of the mice. After fixing the skull to a holder using dental acrylic, a craniotomy and durotomy were performed over the right whisker barrel and surrounding cortex. A well was formed around the craniotomy using dental acrylic, and the exposure was kept filled with artificial CSF until the electrode array was placed. Prior to recording the mice were administered pancuronium and artificially ventilated, and prior to stimulus trials the mice were switched to alpha-chloralose anesthesia. The exposure was dried prior to the electrode placement, and then covered with 0.7% agarose made with artificial CSF. The electrodes arrays used were arranged in square grids with either 0.2 or 0.25 mm spacing, and either 50 micron or variable diameter contact sizes. The reference electrode was silver-chloride ball placed between muscle tissue exposed for the craniotomy. Whisker flick stimuli were presented every 2 seconds, and recordings included both spontaneous epochs as well as periods with stimulation.
All data was recorded with an Intan RHD2000 system, and the recordings were sampled at 20 kS/s with a high pass filter at 0.1 Hz and low pass filter at 7500 Hz. Channels that by visual inspection were highly contaminated with noise were assumed to be from damaged electrodes removed from further analysis.
Data was then downsampled to 4 kS/s, and 6 FIR bandpass filters were applied, chosen such that they span about 0.6 octaves, have no overlap, not include 60 Hz, and roughly correspond to physiological bands (theta, alpha, beta, gamma, high gamma): 6–9 Hz, 10–15 Hz, 20–30 Hz, 35–50 Hz, 70–110 Hz, and 130–200 Hz.
Windows of time where the reference signal was more than 35 uV from zero, any one channel was outside +/- 4 mV, or the signal in the highest band was more than 20 times the RMS in that band were marked as potential artifacts and excluded along with 750 ms prior and 1.25 s after. Regions that were not removed in this way but were shorter than 6 seconds were also excluded.
The data was segmented into continuous 2 s windows. For each window correlation was calculated using Pearson’s correlation coefficient for every possible pair of electrodes on the grid. Each channel pair also has an associated inter-electrode distance, and the correlation vs. distance plots are the result of averaging the pairwise correlations with all equally spaced pairs. For a subject the average correlation is calculated by pooling all correlations across time and averaging the values by distance.
PCA and ICA decompose the data matrix, s, into linear combinations of components, z, with the transformation, mixing, or weight matrix, W,
s=Wz
The components are all uncorrelated with every other component. Therefore, the mixing matrix obtained from either ICA or PCA can be used to whiten the data—which is to linearly transform the data such that the covariance matrix of the transformed data is the identity matrix. PCA is commonly used as for whitening data, and for our purpose ICA can be defined such that it is a whitening transform due to the ambiguity in the scaling of the components. The components can be arbitrarily scaled so long as the weights are scaled inversely such that the original data is unchanged. We make use of this choice to conveniently express the covariance of the data as a function of the weights
cov(s)=cov(Wz)=Wcov(z)WT=WIWT=WWT
Using this result, the correlation matrix can be computed directly from the mixing matrix. For the model we can start with mixing matrices with each component’s weights being drawn from a two-dimensional Gaussian
Wi,j=Aje−(xj−xi)2+(yj−yi)22σj2
where i is the channel with position (xi, yi), and j is the component with location (xj, yj) and standard deviation 𝜎j. For our model we “sample” the components on a 10x15 square grid with 150 components, one per channel, whose center positions are drawn from a uniform random distribution on the 2D space covered by the grid plus one fifth of the standard deviation of the component for which the center is being determined on either side. The amplitudes of each component are drawn from a uniform random distribution between 0.5 and 1.5, sorted in descending order and then scaled by e-0.1 k, where k is 1,2,3… corresponding to the first, second, third, etc. amplitude. This is to mimic the trend of decreasing variance for components typically obtained from PCA and ICA. Once the mixing matrix is determined the covariance, Σ, is given as before by
Σ=WWT
The properties of products of Gaussians allows the covariance elements to be rewritten as
Σi,j=∑k=1NWi,kWj,k=∑k=1N e−(xk−xi)2+(xk−xj)2+(yk−yi)2+(yk−yj)22σ2
=∑k=1N e−2(xk−12(xi+xj))22σ2e−2(yk−12(yi+yj))22σ2e−12(xi−xj)2+12(yi−yj)22σ2
Separating the terms that involve the distance between channels i and j gives
Σi,j=e−dij22(2σ)2∑k=1Ne−2(xk−12(xi+xj))22σ2e−2(yk−12(yi+yj))22σ2=Fi,je−dij22(2σ)2
Each element is the product of a Gaussian function of the distance between channels and Fi,j, a sum over 2D Gaussian functions of the component positions centered at the average location of two electrode positions, which given a fixed set of component locations, is a function of the two electrodes’ positions. For uniformly distributed component positions xk, Fi,j looks like a discrete approximation of the integral over xk of the 2D Gaussian. In the limit of an infinite number of components uniformly distributed across a sufficiently large area it becomes proportional to the integral
Fi,j∝∫−∞∞∫−∞∞e−2(xk−12(xi+xj))22σ2e−2(yk−12(yi+yj))22σ2dxkdyk=F0
and given that this integral is not a function of the center position of the Gaussian, in this limit F is a constant regardless of the positions of channels i and j. In this limit the correlation matrix is exactly a Gaussian function of the distance between the channel pairs with a standard deviation square-root of 2 larger than the standard deviation of the components that generated it:
ri,j=F0e−dij22(2σ)2(F0)(F0)=e−dij22(2σ)2
For the above form of the correlation to be valid doesn’t require that F is near the limit of infinite components, rather that F is not a function of the electrode pair i and j. For the correlation to take the form above on average is an even weaker condition that F can be a weak function of the electrode pair in relation to the distance term so that the small factors multiplying the distance term will tend to cancel when averaged over equidistant pairs and multiple DAC’s.
To include the effect of noise specific to each channel, and uncorrelated from the activity or noise on the other channels, a new component has to be created for each noise element because every component is required to be uncorrelated to all other components. This results in a diagonal matrix of weights which we model as each having the same amplitude, ϵ, across channels
Wnoisei,j=ϵδij
Where ẟij is the Kronecker delta, not the distance used previously. The effect of a reference electrode which is added to all channels can be modeled as a single component with a constant weight vector across all channels
Wref=ρ
so that the modified mixing matrix is the original mixing matrix with additional columns for the noise and reference
W′=(WWnoiseWref)=(ϵ0ρW0ϵρ⋱⋮)
The modified covariance matrix is now given by
Σi,j′=Σi,j+ϵδij+ ρ
and its modified correlation matrix r’ is given by
ri,j′=Σi,j+ϵδij+ρ(Σi,i+ϵ+ρ)(Σj,j+ϵ+ρ)
ICA was applied using the runica() function from EEGLab [44] to the same filtered and segmented 2 second windows that were used to compute the DAC. The built-in PCA option was used to apply PCA prior to ICA as a dimensionality reduction technique, and the ‘extended-ICA’ option was used. For human data with 56 channel grids the first 30 components were kept, and for mice with 32 channel grids the first 20 components were kept. In both cases the excluded components accounted for less than 5% of the variance in the data, and usually were close to 1%.
Component mixing matrices were fit using least squares fitting function lsqcruvefit() in MATLAB to a two-dimensional circular Gaussian function with 5 parameters
Ae(x−B)2+(y−C)22D2+E
with lower and upper bounds such that A be positive, B and C to force the center to lie within 40 grid pitches on either side, D to fit to Gaussians that have standard deviations larger than a single grid pitch.
The median width parameter D was chosen instead of the mean due to the distribution of values being skewed towards zero. In order to calculate the 95% confidence interval of the median a bootstrap with 5,000 resamples was used. Only widths corresponding to components with R2 over 0.7 to exclude components for which a Gaussian is not a good representation and the value of D may not be meaningful.
The individual contribution of each component to the overall covariance matrix can be found by re-calculating the covariance using only the desired component. Any entry in the covariance matrix is a sum over weights corresponding to all components.
Therefore, we define the reduced covariance corresponding to a single component, c, as just the terms involving that component. The sum of all of the reduced covariance matrices is therefore the full covariance matrix.
The corresponding reduced correlation matrix cannot be calculated as usual
ri,j=Σi,jΣi,iΣj,j
because all of the reduced correlation matrices would be identity matrices. Rather the reduced correlation is calculated using the reduced covariance in the numerator, and the full covariance in the denominator,
ri,jc=Σi,jcΣi,iΣj,j
so that the sum of all components of the reduced correlation matrices is equal to the full correlation matrix.
The reduced correlations can be averaged by distance in the same manner as the full one. The contribution of each component to the variance, correlation, and drop in the correlation can be calculated using the reduced covariance and correlation. The variances of the channels are the diagonal entries of the covariance matrix, and we will define the overall variance (across channels) as the trace of the covariance matrix. The variance across channels for a given component is then given by
Varc=∑i=1NWi,cWi,c
such that the overall variance is the sum over components, and the percentage of the variance explained by each component is the component variance divided by the overall variance.
We also want to know the contribution of each component to the DAC. The contribution to the DAC is a similar measure to the contribution to the variance, but the contribution to the drop in the DAC is more relevant for explaining the shape of the correlation vs. distance curve. To calculate these, the DAC curve for each reduced correlation matrix is calculated in the same manner as for correlation matrices. The drop in the DAC due to each component is calculated by subtracting the zero-distance value of the DAC from all the values, and it retains the desired property that the sum over all components is equal to the drop in the full DAC.
To summarize the amount each component contributes to the drop in the DAC into a single quantity, the percentage explained is calculated at each distance, and then averaged over all distances. With this, each component can be assigned a percentage of the total variance, total DAC, and total drop in the DAC.
The common average reference can be computed with the matrix
1NJ
Where N is the number of electrodes and J is the NxN matrix of ones. The product of this matrix with s computes the average signal and is subtracted from the original signal to yield the CAR version of the signal
sCAR=s−s¯=s−1NJs=[I−1NJ]s=Cs
The covariance of the signals after CAR is then
cov(sCAR)=Ccov(s)CT
The effect on the component-based representation is
sCAR=Cs=CWz
It is important to note that this modification of the weight matrix applies to the weight matrix obtained without the change of reference. When CAR is applied to the data the temporal components identified by a whitening algorithm such as PCA are not necessarily the same.
|
10.1371/journal.pcbi.1005590 | Dynamic metabolic modeling of heterotrophic and mixotrophic microalgal growth on fermentative wastes | Microalgae are promising microorganisms for the production of numerous molecules of interest, such as pigments, proteins or triglycerides that can be turned into biofuels. Heterotrophic or mixotrophic growth on fermentative wastes represents an interesting approach to achieving higher biomass concentrations, while reducing cost and improving the environmental footprint. Fermentative wastes generally consist of a blend of diverse molecules and it is thus crucial to understand microalgal metabolism in such conditions, where switching between substrates might occur. Metabolic modeling has proven to be an efficient tool for understanding metabolism and guiding the optimization of biomass or target molecule production. Here, we focused on the metabolism of Chlorella sorokiniana growing heterotrophically and mixotrophically on acetate and butyrate. The metabolism was represented by 172 metabolic reactions. The DRUM modeling framework with a mildly relaxed quasi-steady-state assumption was used to account for the switching between substrates and the presence of light. Nine experiments were used to calibrate the model and nine experiments for the validation. The model efficiently predicted the experimental data, including the transient behavior during heterotrophic, autotrophic, mixotrophic and diauxic growth. It shows that an accurate model of metabolism can now be constructed, even in dynamic conditions, with the presence of several carbon substrates. It also opens new perspectives for the heterotrophic and mixotrophic use of microalgae, especially for biofuel production from wastes.
| Most existing metabolic modeling tools are not suitable for studying diauxic growth with dynamic substrate shifts. This paper describes a successful modeling of Chlorella sorokiniana metabolism, based on 172 reactions and validated by nine independent dynamic experiments (nine experiments were used for its calibration), in which microalgae were grown heterotrophically or mixotrophically on acetate and/or butyrate, in the light or dark. Such an extensive validation has not been performed before for microalgae. It was demonstrated that the model could be used to assess the flux dynamics in the cell. This innovative simulation tool was also used to derive original strategies to bypass the toxicity of some substrates using mixotrophic regimes.
| Microalgae are unicellular eukaryote microorganisms that can grow autotrophically using light energy and CO2. Many species can also grow heterotrophically in darkness on various organic carbon sources, including glucose, or can combine heterotrophy and autotrophy for a mixotrophic growth [1]. Microalgae have been domesticated and used to synthesize many products with industrial applications, such as pharmaceutics or cosmetics (antioxidants, pigments, unsaturated long-chain fatty acids), agricultural products (food supplements, functional food, colorants) and animal feed (aquaculture, poultry or pig farming) [2]. They are also promising organisms for green chemistry (bioplastics), the environment (wastewater treatment, CO2 mitigation), and even energy production (biodiesel, bioethanol, hydrogen) [2].
Autotrophic growth of microalgae is limited by light distribution to all the cells, constraining the cell concentration to below 10 g/l (for the thinnest and most concentrated cultivation systems). Heterotrophic growth does not have this limiting factor and higher biomass density can be achieved [1], drastically reducing the harvesting costs. In addition, heterotrophic growth is usually faster, reducing the cultivation time [3]. However, industrial production of heterotrophic microalgae is hampered by the high economic and environmental costs of glucose, commonly used as a substrate. One solution is to use the waste from other processes, such as glycerol, acetate (ACE) or butyrate (BUTYR), which represent low cost carbon substrates. For instance, dark anaerobic fermentation produces an effluent mainly composed of acetate and butyrate [4]. However, some substrates in waste, such as butyrate, can be inhibitory [5]. Moreover, the successive metabolic switches between different substrates are not well understood and are likely to significantly affect growth. Therefore, this bioprocess still needs to be mastered and optimized to produce microalgae and extract the targeted byproducts on an industrial scale and at a competitive price, with consistent quality and in a sustainable way.
In this context, mathematical modeling of the metabolism has proven to be an efficient tool for optimizing growth and increasing the production of target molecules. To date, no models exist for heterotrophic microalgal metabolism dynamically switching between several substrates (S1 Table), including mixotrophic growth in light. So far, only static fluxes have been predicted under constant substrate consumption [6–9]. Representing the dynamic shifts for a blend of substrates typical of real wastewater is a major challenge, since some intracellular accumulation might occur, either during the transition between substrates, or due to the varying nature of the light. As a consequence, the quasi-steady state assumption (QSSA) required by most of the existing metabolic approaches may be an invalid hypothesis in this case [10]. The DRUM modeling framework recently proposed in [10] was used here to handle the non quasi-steady state (QSS). It allowed the development of a dynamic metabolic model for Chlorella sorokiniana grown on a single-substrate culture and a mixed-substrate (acetate and butyrate) culture, combined with various combinations of light. The model is thus designed to represent autotrophic, heterotrophic or mixotrophic modes under diauxic conditions. Our purpose is to propose a relatively generic model, instantiated and calibrated for C. sorokiniana. According to Baroukh et al. [11], such a generic model should be applicable to a wide range of microalga species.
The goal of the experiments was to grow Chlorella sorokiniana on a synthetic medium mimicking the digestate composition produced by a dark fermenter processing organic waste. At this stage, the composition of the medium was kept simple, with only the two main organic components—acetate and butyrate [4]–to gain a clear understanding of their effects on microalgae growth. Chlorella sorokiniana was grown both in the dark and in the light (136 μE.m-2.s-1), in axenic conditions at 25°C and constant pH (6.5) in triplicate batches with different initial concentrations of acetate and butyrate (Table 1). Nitrogen (ammonium) and phosphorus were provided in non-limiting concentrations in order to focus solely on carbon metabolism. To ensure that no substrate was favored because of acclimation, the inoculum was grown autotrophically beforehand. See Turon et al. [5,12] for more details of the experimental protocols.
A detailed description of the metabolic network reconstruction is provided in S1 File. Since Chlorella sorokiniana has not been sequenced yet, no genome-scale metabolic network (GSMN) reconstruction was possible. However, the core carbon and nitrogen metabolic networks in the GSMN of previously reconstructed microalgae species are relatively similar [13]. Thus, the conserved core metabolic network was used, containing the central metabolic pathways relevant to mixotrophy and heterotrophy: photosynthesis, glycolysis, pentose phosphate pathway, citric acid cycle, oxidative phosphorylation, and synthesis of chlorophyll, carbohydrates, amino acids and nucleotides. Species-specific pathways such as the synthesis of secondary metabolites were not represented, since these pathways were assumed to represent negligible fluxes compared to the main pathways and thus to have little impact on the metabolism. The reactions involved in macromolecule synthesis (proteins, lipids, DNA, RNA and biomass) were lumped into generic reactions. The growth-associated ATP maintenance (GAM) was replaced by the value experimentally measured by Boyle et al. (2009) for growth of Chlamydomonas reinhardtii on acetate. They observed 29.890 moles of ATP for 1000g of biomass. In our model, the biomass reaction yields 186g of biomass; the maintenance term is thus 5.56 moles of ATP per mol of biomass (details on how this value was computed are available in S1 File). A sensitivity analysis was carried out to assess the impact of this maintenance value on model accuracy (S1 Fig). Results showed that the optimal growth associated yield is 12.2mol ATP per mol of biomass, with [0–16.7 molATP/mol B] as the 10% interval confidence. The value of Boyle et al [6] falls in this interval, demonstrating that an error of this term in this range has a minor impact on model predictions. The non-growth ATP maintenance (NGAM) was assumed negligible. It may explain the slightly higher value of the optimal growth associated yield resulting from optimization, which might also compensate for NGAM.
Generally, metabolic modeling relies on the QSSA of the whole metabolic network, where intracellular metabolites cannot accumulate or be depleted [14]. The idea of the DRUM approach is to mildly relax this hypothesis, by splitting the metabolic network into a limited number of sub-networks [10], for each of which the QSS is assumed. The metabolites situated at the junction between the sub-networks, can therefore have dynamics of accumulation and depletion. The sub-networks are defined by metabolic functions and take into account cellular compartments. This assumption is supported by the idea that cell function and cell compartment are often associated with co-regulation and substrate channeling, which implies synchronicity of reactions and thus quasi-steady state for those reactions [10]. The idea is also to find a network splitting simple enough for explaining the experimental data, so as to avoid overfitting by postulating too many reactions kinetics [10]. Further details on the philosophy behind the network splitting and the DRUM framework are given in the discussion section of Baroukh et al. [10] (DRUM principles are also summarized in supplementary information (S1 File, section 8)). For representing the growth of Chlorella sorokiniana with different inorganic or organic carbon sources, the network was split taking into account the compartments of the cell with a global catabolic or anabolic function (Fig 1): i) the glyoxysome for acetate and butyrate assimilation, ii) the chloroplast for photosynthesis, and iii) the rest of the reactions for functional biomass production (synthesis of lipids, carbohydrates, proteins, DNA, RNA and chlorophyll). Several splitting were tested, particularly on the transported metabolites between the glyoxysome and the cytosol (which are not consistent in literature). The best fitting results were obtained with these three sub-networks. Apart from inorganic compounds, only succinate (SUC) and glyceraldehyde 3-phosphate (GAP) were intracellular metabolites (A) that potentially accumulate. The idea is that these metabolites, which shuttle between the compartments (respectively between the glyoxysome and the cytosol and between the chloroplast and the cytosol), and which act as intermediate between catabolism and anabolism, are the ones that could act as “buffers” inside the cell. Each sub-network was balanced in cofactors and in chemical elements (carbon, oxygen, nitrogen, phosphorus, sulfur).
Each sub-network was then reduced to macroscopic reactions (MRs) using elementary flux mode analysis [15]. To compute elementary flux modes (EFMs), the efmtool program was used [16]. For the three sub-networks, the EFM could be computed easily, since their total number was less than 1751 (it should be noted that an EFM analysis of the full network results in 5105 modes).
Glyoxysomes are specialized peroxisomes found in plants or microalgae [8], in which fatty acids (including acetate and butyrate) can be used as a source of energy and carbon for growth when photosynthesis is not active. Fatty acids are hydrolyzed to acetyl-CoA, and then transformed into succinate via the glyoxylate cycle. Succinate can then be transformed into a variety of macromolecules for biomass growth, through combinations of other metabolic processes taking part in other compartments of the cell.
Reduction of the glyoxysome sub-network yielded two MRs, one for each substrate (Table 2).
Photosynthesis supports the generation of cell energy in phototrophic organisms and contributes to the incorporation of inorganic carbon. The process takes place in the chloroplast and consists of two steps commonly known as the light and dark reactions. The light reaction consists of the generation of cell energy (ATP, NADPH) from water and photons, producing oxygen. Thanks to the energy of the light reaction, the dark step reactions incorporate carbon dioxide through the Calvin cycle producing a 3-carbon sugar (3-phosphoglycerate, or 3PG). Then, 3PG is transformed into glyceraldehyde 3-phosphate (GAP) and transported to the cell cytosol.
Elementary flux mode analysis of this sub-network yielded only one Elementary Flux Mode (EFM) (Table 2), associated with one macroscopic reaction (MR3). The stoichiometry of the derived macroscopic reaction is in agreement with the literature: a quota of 8 photons are needed per carbon incorporated [17].
The synthesis reactions of lipids, proteins, DNA, RNA, chlorophyll and carbohydrates were grouped into a functional biomass synthesis sub-network and assumed to be in QSS. This sub-network includes glycolysis, TCA cycle, oxidative phosphorylation, pentose phosphate pathway, nitrogen and sulfur assimilation, carbohydrate synthesis, lipid synthesis, amino acid synthesis and nucleotide synthesis.
The reduction of this sub-network yielded 1748 EFMs, which is reasonable given the number of involved reactions (143), and much lower than the number of modes for the full network (5105). Most of these EFMs (1618) yielded biomass, while the others correspond to futile cycles. In terms of carbon, once normalized by unit of biomass synthesis flux, the 1618 MRs deduced from the EFMs only differed in their consumption of SUC and GAP and their production of CO2 (S2 and S3 Figs). As in Flux Balance Analysis [18], we assumed that the cell maximized biomass growth, and hence minimized carbon loss when synthesizing biomass from each substrate. The elementary flux modes with the best SUC/CO2 yield and GAP/CO2 yield were thus chosen (Table 2). The resulting MR consumes SUC or GAP and NH4 for carbon and nitrogen sources, SO4 and Mg for protein and chlorophyll synthesis and O2 for ATP synthesis through oxidative phosphorylation.
At this stage, the macroscopic kinetics of the MRs must be determined in order to simulate the metabolic dynamics [10]. We assumed that only the carbon substrates of each MR were limiting, playing thus a role in the kinetics. Michaelis-Menten kinetics was used for acetate consumption (Table 2), since experimental data showed no growth inhibition on acetate. Haldane kinetics was chosen for the butyrate consumption reaction (Table 2), since experimental data showed that butyrate inhibited biomass growth (growth only possible with maximum 0.1 gC.L-1 (Fig 2C) in butyrate-only experiments).
Biomass growth clearly exhibited a diauxic growth for the mixed substrate conditions: acetate was entirely consumed before the butyrate concentration started to decrease (Fig 3). This diauxic growth is probably the result of transcriptional regulations taking place inside the cell. DRUM framework, can explicitly describe such regulations via an appropriate choice of the kinetics in connection with metabolite accumulation. However, to the best of our knowledge, the transcriptional regulations responsible for this diauxic growth are not known. It is thus premature to propose an explicit mathematical expression for representing this phenomenon; a general kinetic expression for diauxic growth, which implies that regulation is performed by acetate directly on the cell transporter of butyrate, was thus chosen. An inhibitory term of acetate concentration on butyrate consumption kinetics was included in the butyrate uptake kinetics (Table 2). Linear kinetics depending on the mean light intensity in the reactor was chosen to represent photosynthesis (Table 2). The mean light intensity was computed using a Beer-Lambert law (S1 File). Linear kinetics with respect to the carbon substrate were chosen for biomass synthesis (Table 2).
Finally, the dynamics of the 172 fluxes in the metabolism can be derived from a system of 14 differential equations comprising 14 metabolites and 5 macroscopic reactions representing 3 compartments:
dM′dt=d(SAB)dt=K′.α.B
where M’ is the vector of metabolites (14x1) composed of substrate S, metabolites susceptible to accumulate A (SUC and GAP) and functional biomass B; K’ is the reduced stoichiometric matrix (14x5) and α is the kinetics vector (5x1) (Table 2, S6 Fig). The way all the metabolic fluxes are computed from K’ and α is recalled in S1 File. Total biomass X (g.L-1) is computed thanks to a mass balance on the cell:
X(t)=∑AMA.A(t)+MB.B(t)
with MA and MB the molar masses of metabolites A and B (for further details, see S1 File section 5).
The dynamic model has 10 degrees of freedom, and each degree is represented by a parameter that needs to be calibrated. To estimate the parameters, we minimized the squared-error between simulation and experimental measurements using the Nelder-Mead algorithm [19] (function fminsearch in Scilab®). To reduce the risk of local minima, several optimizations were performed with random initial parameters. Nine experiments were used to estimate the parameters (Table 1); the nine remaining experiments were reserved to assess the validity of the model (Table 1). Results of the parameter identification are presented in Table 3.
The model simulation accurately reproduces experimental data, even for the validation data sets that were not used for calibration (Figs 2–4). The diauxic growth is particularly well represented (Figs 3 and 4D), and the transient behavior, together with the final biomass, is correctly predicted (Figs 2–4), showing that the biomass yields obtained from the metabolic network are accurate. Indeed, one of the advantages of metabolic modeling [20] is the prediction of biomass yields supported by the stoichiometry of the metabolic network. Here, the predicted conversion yield of acetate and butyrate to biomass is 0.514 grams of carbon biomass per gram of carbon in the incoming substrate. This yield contributes to correct prediction of the biomass for both acetate (Fig 2B) and butyrate (Fig 2D), thus validating the approach. Interestingly, the yields are identical between the two substrates. A possible explanation is the fact that more ATP is required for the transport of butyrate into the cell than for acetate, thus balancing the ATP created when converting butyrate and acetate to succinate.
The set of kinetic parameters matches both the single-substrate culture and the mixed-substrate culture. This implies that butyrate has no impact on acetate growth rate. However, the inverse is not true, since the acetate concentration at which butyrate consumption starts (kD) is very low (5.39*10−10 M), illustrating the strong diauxic growth that occurs. Even the smallest amount of acetate inhibited butyrate uptake. The maximum acetate uptake rate was higher than the maximum butyrate uptake rate by nearly 15 fold, reflecting the preference of Chlorella sorokiniana for acetate. The non-inhibiting butyrate concentration (SoptMR2) was very low (1.93*10−5 M), which highlights the strong inhibition of butyrate in the medium on its uptake. It also explains why, in the butyrate-only experiments, no biomass growth was observed for butyrate concentrations above 0.1 g.L-1 (Fig 2D).
In addition to substrates and biomass concentrations, light evolution inside the culture vessel was computed. During the first few days, the average light intensity decreases until equilibrium is reached around 16 μE.m-2.s-1 (S1 File section 4, S8 Fig). It represents 11.7% of the incident light and is in agreement with the literature [21]. Interestingly, equilibrium is reached faster for mixotrophic growth, particularly on acetate, which supports fast heterotrophic growth (S8 Fig). In addition, the photosynthetic quotient for autotrophic growth varies between 1.0 and 1.16, matching the typical range of 1.0–1.8 for microalgae [6].
The predicted metabolic fluxes (Fig 5, S9 Fig) are in accordance with previous studies [11]. Autotrophy (S9C Fig) is characterized by high fluxes in the photosynthetic pathways, which convert light and CO2 to GAP. Beyond these pathways, fluxes drop considerably in terms of absolute magnitude. Upper glycolysis is in the gluconeogenic direction to produce the carbohydrate and sugar precursor metabolites (Glucose 6-phosphate (G6P), Ribose 5-phosphate (R5P), Erythrose 4-phosphate (E4P)) necessary for growth. In the heterotrophic mode, fluxes are more homogenous among reactions (Fig 5B, S9A Fig). Acetate and butyrate are converted to acetyl-CoA in the glyoxysome (Fig 5B, S9D Fig). Acetyl-CoA is then converted into succinate by the glyoxylate cycle and injected in the TCA cycle. Upper glycolysis also goes in the gluconeogenic direction to produce carbohydrate and sugar precursors. This can be achieved thanks to the anaplerotic reactions that convert oxaloacetate to phosphoenolpyruvate (PEP). Mixotrophy is a mixed combination of the autotrophic and heterotrophic modes (Fig 5A, S9A Fig). For mixotrophic growth on acetate, heterotrophic metabolism is dominant, whereas autotrophic metabolism is dominant for mixotrophic growth on butyrate. This is due to the fact that autotrophic growth is slower than growth on acetate but faster than growth on butyrate.
Interestingly, in agreement with the data, the model did not predict any growth on butyrate above 0.1 gC.L-1, and at the same time successfully forecasted growth on 0.9 gC.L-1 butyrate in mixed substrate conditions (Fig 3E) and on 0.3 gC.L-1 butyrate in mixotrophic conditions. Indeed, in these conditions, the first-stage growth on acetate and/or light produces enough biomass to finally consume such an inhibiting quantity of butyrate. The substrate to biomass (S/X) ratio is known to be a key process parameter for overcoming the inhibitory effects of the substrate [22]. The model therefore represents a tool to compute and optimize the amount of co-substrate that must be added to overcome the inhibition and consume the butyrate. Different strategies could be tested to achieve a low S/X ratio and accelerate butyrate consumption. The simplest approach would involve adding a non-inhibiting substrate in order to reduce the amount of inhibitory substrate per unit biomass. For example, the addition of 0.5 gC.L-1 of acetate for a volume equal to half of the culture volume has been found to eventually lead to the consumption of 0.5 gC.L-1 of butyrate in 14 days (Fig 6B), which would not have been possible otherwise (Fig 6A). However, in general, such pure substrate is not available. We therefore simulated the addition of a mix of acetate and butyrate in proportions that are representative of fermentative digestate [4], for a volume equal to half of the culture volume. On the one hand, the acetate contained in the waste stimulated growth, but since it is associated with addition of butyrate, it also increased inhibition. Simulations show that the inhibition is overcome, but does not lead to the total consumption of butyrate within 15 days (Fig 6D). Furthermore, the mixotrophic potential can be exploited: autotrophic growth can be enhanced by illumination in order to ultimately dilute the inhibitory substrates. Illuminating the algae at an incident intensity of 136 μE.m-2.s-1 leads to the consumption of the same quantity of butyrate in 13 days, and this delay can be reduced to 9 days using a light intensity of 272 μE.m-2.s-1 (Fig 6C). Finally, if light is provided at the same time as the addition of fermentative digestate (for a volume equal to half of the culture volume), inhibition can be overcome after 10 days (Fig 6D).
The advantage of the DRUM approach is its ability to account for the accumulation of some intracellular metabolites and thus to characterize the time to reach steady state. It can also determine more quantitatively the time scales of flux variations in the cell than earlier frameworks. This analysis was applied to SUC and GAP, which are, in our model, the intermediate accumulating metabolites.
Interestingly, SUC actually hardly accumulates in the simulations and rapidly achieves a QSS (S10 Fig) where its concentration evolves slowly compared to the other variables in the system (substrate consumption, biomass formation). We developed an algorithm to automatically detect the time needed to reach QSS (tQSS). In the experimental conditions of this study, approximately 3 minutes were necessary for succinate to achieve QSS (S2 File) thanks to a higher biomass synthesis rate (via a high kMR4) compared to the substrate assimilation rate, implying that succinate is immediately consumed once it is synthesized from butyrate or acetate. A sensitivity analysis on the parameter kMR4 revealed that the confidence interval of tQSS was [0.6; 34] minutes (model error less than 5% of the minimal error) (S1 File section 8, S2 File). After the brief transient succinate step, the QSSA for heterotrophic growth on butyrate and acetate is valid. Therefore, the macroscopic model can be reduced further, by merging reaction MR4 with reactions MR1 and MR2 (S1 File section 6). The same kinetic parameters can be used for simulation, and the fit is nearly identical (increase of 0.6% of the error). As a consequence, results considering QSSA are very close to the ones based on DRUM.
GAP, in contrast to SUC, does not reach a QSS rapidly (S11 Fig). First, GAP accumulates at high light intensities, reaching a maximum when average light intensity is approximately 60 μE.m-2.s-1. Then, it is consumed at low light intensities, reaching a QSS when average light intensity reaches a steady state at 16 μE.m-2.s-1 (S1 File section 8). This suggests that microalgal metabolism in autotrophic and mixotrophic modes only reaches a QSS when average light is constant in the culture media, meaning that growth has ceased. This behavior is similar to that of microalgae grown in day/night cycles [10,11], involving accumulation of carbon-reserve metabolites (carbohydrates, lipids) during the day, when the light is intense enough, and re-consumption during the night or at the beginning and end of the day, when light intensity is low. Here, the carbon reserve metabolite is GAP, because only GAP accumulated in the model. Nevertheless, it is probable that carbohydrates and/or lipids also accumulate. Further experiments are required to validate these results more extensively and to determine which carbon-only metabolite is stored inside the cell.
To confirm these results, a Macroscopic Bioreaction Model of the system [23], relying on the QSSA assumption, was developed (see S1 File section 10 for details on the methodology). Without accumulation of SUC, the model error was almost unchanged (0.06% increase of the error). But without the possibility for GAP to accumulate, a 40% increase in the error is observed. This confirms our finding that GAP do accumulate inside the cell at high light intensities to be consumed later at lower light intensities. It is also interesting to note that the MBM approach is sufficient and produces accurate results, for applications in heterotrophy only cultures, without the need for accumulating metabolites.
The dynamic metabolic model developed for the heterotrophic, mixotrophic and autotrophic growth of Chlorella sorokiniana on acetate and butyrate achieved a so far unequalled accuracy. The model efficiently fits the dynamic experimental data and correctly predicts the biomass yields for a broad range of experimental conditions. This new powerful simulation tool provides new insight into the mixotrophic microalgal process, and allows us to explore the different possibilities to overcome the inhibition induced by some of the substrates, in particular by adjusting the mixotrophic regimes. The model also highlights the dynamics of some internal compounds, especially under an auto- or mixotrophic regime, while light intensity is slowly affected by an increase in self-shading. As a consequence, the model shows that QSSA is not valid for mixotrophic growth as long as the light is variable in the culture medium. In the future, the model should be extended further in order to handle mixotrophic behavior under periodic light/dark cycles.
|
10.1371/journal.pmed.1002789 | Pharmacokinetics, optimal dosing, and safety of linezolid in children with multidrug-resistant tuberculosis: Combined data from two prospective observational studies | Linezolid is increasingly important for multidrug-resistant tuberculosis (MDR-TB) treatment. However, among children with MDR-TB, there are no linezolid pharmacokinetic data, and its adverse effects have not yet been prospectively described. We characterised the pharmacokinetics, safety, and optimal dose of linezolid in children treated for MDR-TB.
Children routinely treated for MDR-TB in 2 observational studies (2011–2015, 2016–2018) conducted at a single site in Cape Town, South Africa, underwent intensive pharmacokinetic sampling after either a single dose or multiple doses of linezolid (at steady state). Linezolid pharmacokinetic parameters, and their relationships with covariates of interest, were described using nonlinear mixed-effects modelling. Children receiving long-term linezolid as a component of their routine treatment had regular clinical and laboratory monitoring. Adverse events were assessed for severity and attribution to linezolid. The final population pharmacokinetic model was used to derive optimal weight-banded doses resulting in exposures in children approximating those in adults receiving once-daily linezolid 600 mg. Forty-eight children were included (mean age 5.9 years; range 0.6 to 15.3); 31 received a single dose of linezolid, and 17 received multiple doses. The final pharmacokinetic model consisted of a one-compartment model characterised by clearance (CL) and volume (V) parameters that included allometric scaling to account for weight; no other evaluated covariates contributed to the model. Linezolid exposures in this population were higher compared to exposures in adults who had received a 600 mg once-daily dose. Consequently simulated, weight-banded once-daily optimal doses for children were lower than those currently used for most weight bands. Ten of 17 children who were followed long term had a linezolid-related adverse event, including 5 with a grade 3 or 4 event, all anaemia. Adverse events resulted in linezolid dose reductions in 4, temporary interruptions in 5, and permanent discontinuation in 4 children. Limitations of the study include the lack of very young children (none below 6 months of age), the limited number who were HIV infected, and the modest number of children contributing to long-term safety data.
Linezolid-related adverse effects were frequent and occasionally severe. Careful linezolid safety monitoring is required. Compared to doses currently used in children in many settings for MDR-TB treatment, lower doses may approximate current adult target exposures, might result in fewer adverse events, and should therefore be evaluated.
| Linezolid is an antibiotic that has recently been repurposed for the treatment of multidrug-resistant tuberculosis (MDR-TB).
Based on emerging data from clinical trials, it is now recommended for inclusion as a priority in treatment regimens for adults and children.
There is a lack of evidence-based dosing guidelines for linezolid in children for MDR-TB.
Linezolid has frequent adverse effects that are associated with higher doses and longer durations of treatment, but this has not been well described in children treated for MDR-TB.
We assessed linezolid pharmacokinetics (concentrations) and safety among children routinely treated for MDR-TB from 2 observational studies.
We modelled this pharmacokinetic data and simulated the doses in children that would be needed to achieve current targets.
Doses supported by our data for some weight bands were lower than currently in use, which may reduce the risk of adverse effects.
Linezolid-related adverse effects were frequent—with low haemoglobin being the most common—and occasionally severe (grade 3 or 4).
Revised linezolid doses supported by this study now provide an evidence base for international dosing recommendations.
All linezolid-treated children with MDR-TB should have careful safety monitoring to include at least regular haemoglobin monitoring.
Larger studies, evaluating the revised dosing, are needed to better characterise linezolid safety and to assess the risk of more rare linezolid-related events, especially given the small number of children in this study.
| Multidrug-resistant (MDR) tuberculosis (TB) continues to threaten global TB control, with an estimated 460,000 incident cases worldwide in 2017 [1]. Treatment options remain limited. Linezolid, an oxazolidinone antibiotic that binds to the 50S ribosomal subunit inhibiting protein synthesis [2], is increasingly being used for MDR-TB treatment. In routine use in adults with MDR-TB, linezolid has been associated with good outcomes, with 68% and 82% of patients successfully treated in 2 systematic reviews [3, 4]. In a recent systematic review and individual patient data meta-analysis of 12,030 patients with MDR-TB, treatment with linezolid was significantly associated with treatment success compared to failure or relapse, and also with reduced mortality [5]. The World Health Organisation (WHO) released updated guidance of MDR-TB treatment in 2018, which now classifies linezolid as a Group A drug, meaning that it is a priority to include in individually constructed MDR-TB regimens for adults and children [6].
Interest in linezolid has grown further based on the preliminary results of the Nix-TB trial (NCT02333799), a single-arm open-label phase III study that evaluated very difficult to treat adults with extensively drug-resistant (XDR)-TB or MDR-TB treatment intolerance, or failure, with a three-drug regimen of bedaquiline, pretomanid, and linezolid (1,200 mg given once daily) for 6 months [7]. In an interim analysis of the first 75 patients, 6 died, and all the remaining patients completed the 6 months of treatment; 66 (88%) had durable cure 6 months later, and there were only 2 relapses [7]. Linezolid is also a component of multiple other novel, shortened MDR-TB treatment regimens currently under evaluation in adults [8].
Used in short courses (<28 days), linezolid is safe and well tolerated, but with the longer treatment durations being used for MDR-TB treatment (typically 6 months or longer), it is associated with frequent serious, dose- and duration-dependent adverse effects, including anaemia, neutropaenia, thrombocytopaenia, peripheral neuropathy, and more rarely, optic neuropathy, lactic acidosis, pancreatitis, and rhabdomyolysis [9]. In the systematic reviews of linezolid-treated adults with MDR-TB, linezolid-related adverse effects were reported in 61% and 59% of patients, respectively [3, 4], with 69% of these requiring linezolid dose adjustment or discontinuation in one review [3]. A linezolid dose of >600 mg daily was associated with an increased risk of adverse effects in these reviews [3, 4]. In early interim analysis of the Nix-TB study, 27% of participants experienced a serious adverse event, and 71% of participants had at least one linezolid-related treatment interruption, due mostly to myelosuppression or peripheral neuropathy [10].
There is a substantial burden of MDR-TB in children, with an estimated 25,000 to 32,000 incident cases globally each year [11, 12], many of whom could benefit from linezolid treatment. The evidence base for linezolid use in children with MDR-TB is currently limited. A 2014 review of the literature identified only case reports and small case series that described 18 linezolid-treated children with MDR-TB [9]. As in adults, treatment outcomes were good, with 15 of 18 children (83%) successfully treated, but 9 (50%) experienced a linezolid-related adverse effect, 5 (28%) required dose adjustment, and 2 (11%) discontinued linezolid [9]. These limited early data were retrospective and had variable reporting of key information. High-quality prospective data are needed to better characterise the incidence, severity, and timing of adverse events among children with MDR-TB receiving long-term linezolid (>28 days of treatment).
The optimal dose of linezolid for treatment of adults with MDR-TB is not yet defined. However, 600 mg once daily is currently the most frequently used routine dose in adults. The dose of linezolid in children needed to approximate target exposures in adults with MDR-TB receiving a 600 mg daily dose has not yet been characterised. Linezolid is well absorbed after oral administration, with bioavailability of nearly 100% [2, 13]. Linezolid has a complex metabolism, with the primary metabolite formed through a nonenzymatic mechanism and the major pathway of elimination urinary excretion [13]. Linezolid has good penetration into tissues, including cerebrospinal fluid (CSF), where its exposure is 70% to 98% of plasma exposure [13–15]. Studies of linezolid pharmacokinetics in adults to date have shown substantial variability between patients, patient populations, and studies, with limited data in patients with TB [16–18].
Linezolid pharmacokinetics has been studied in children with gram-positive infections [19]; however, there are no data in children treated for TB. This gap is important, because dosing guidance should ideally be based on data from the target population, particularly as linezolid pharmacokinetics is known to vary considerably across disease states [18]. Understanding the optimal dose of linezolid for MDR-TB treatment in children is critical, because too high a dose may increase the risk of serious adverse effects, while underdosing may reduce the efficacy of this drug and may potentially lead to the development of linezolid resistance [20, 21]. Practical, weight-banded paediatric dosing guidance for linezolid based on its pharmacokinetics in children with TB is urgently needed as its use is likely to increase substantially based on the revised 2018 WHO guidance.
The objective of this study was to characterise the pharmacokinetics and safety of linezolid in children routinely treated for MDR-TB and to estimate optimal weight-banded doses that achieve adult target exposures for MDR-TB treatment.
The data described here are combined from 2 prospective observational pharmacokinetic studies in Cape Town, South Africa, which used standard clinical study measures, identical drug formulations, and standard sample collection and laboratory methods. The first observational study (MDRPK1) enrolled HIV-infected and -uninfected children 0 to <15 years of age (n = 173) routinely treated with second-line anti-TB drugs for probable or confirmed drug-resistant TB, who were followed until the end of MDR-TB treatment, from 2011 to 2015. All children from this cohort routinely treated with linezolid were included in the current analysis. The second study (MDRPK2) was a direct follow-up observational study to MDRPK1 and enrolled HIV-infected and -uninfected children 0 to <18 years of age (n = 64) routinely treated for MDR-TB with levofloxacin, moxifloxacin, or linezolid, in the same setting, from 2016. Follow-up for MDRPK2 is ongoing. In MDRPK2, children not prescribed linezolid as a component of their routine MDR-TB care received a single dose of linezolid on the day of pharmacokinetic sampling and therefore contributed data to pharmacokinetic analyses but not to long-term safety data. The total sample size for MDRPK2 was estimated from clinical trial simulations with existing models to ensure sufficient precision primarily of the pharmacokinetic parameter estimates (relative standard error [RSE] of parameter estimates of <10%) (see S1 Text). The minimum overall sample size identified was 80 evaluable participants, with 100 participants being the maximum targeted sample size to allow for attrition. The present report is based on a planned interim analysis after enrolment of 50 participants; 9 patients from MDRPK1 were added to supplement participants from MDRPK2. The timing of this analysis (which included just under the planned 50 participants) was also influenced by the lack of a strong evidence base for linezolid dosing for TB in children and the urgent need for data, as well as the timing of the revision of WHO MDR-TB treatment guidelines and the need to inform that process in the absence of other paediatric data.
Children with MDR-TB were treated individually according to national and international guidelines—with a minimum of 4 confirmed- or likely-effective drugs, usually with the addition of pyrazinamide and ethambutol—generally for 12 to 18 months’ duration [22–24]. Due to its cost and adverse effects, at the time of the study, linezolid was reserved for children with MDR-TB with (1) probable or confirmed additional fluoroquinolone resistance, including those with XDR-TB; (2) MDR-TB meningitis; (3) or documented intolerance to other second-line anti-TB medications. Management of linezolid-associated adverse effects was individualised depending on the event type and severity, the child’s TB disease severity, and the availability of other treatment options. For common, nonsevere events such as low-grade (grades 1 or 2) anaemia, linezolid would typically be temporarily interrupted and then restarted at half the dose. For more severe events (grade 3 or 4), linezolid may have been permanently discontinued, depending on other treatment options and the child’s clinical status at the time. These clinical decisions were individualised and were made at the discretion of the routine treating clinician.
At the time of the study, there was no dosing recommendation based on high-quality evidence for linezolid for TB treatment in children. The routinely prescribed dose of linezolid in the study setting, based on expert opinion, was 10 mg/kg/dose twice daily (20 mg/kg daily) for children <10 years of age and 10 mg/kg/dose once daily for children >10 years of age, up to a maximum total daily dose of 600 mg. This was based on an approximation of the dose required to achieve similar exposure as in adults receiving 600 mg once daily, based on existing pharmacokinetic data from children with non-TB infections [9]. Linezolid was available as 600 mg unscored tablets and as a 20 mg/mL suspension (Pfizer, Sandton, South Africa).
In the MDRPK1 study, an exact 10 mg/kg dose was prepared and administered along with other anti-TB medications in the regimen on the day of pharmacokinetic sampling. Samples were drawn just prior to—and then at 1, 2, 4, 8, and either 6 or 11 hours after—the anti-TB medications’ dose. In MDRPK2, a weight-banded dose (approximately 10 mg/kg) was administered on pharmacokinetic sampling days along with other routine anti-TB medications in the regimen. Samples were drawn just prior to, and at 1, 4, and 10 hours after, the observed dose. For both studies, medications were administered identically, on an empty stomach after an overnight fast, and were given either as whole tablets if the child was able to swallow, as a suspension if available, or as crushed tablets mixed in a small amount of water if the child was unable to swallow and the suspension was unavailable (due to occasional stockout), using identical formulations in both studies. On occasion, all anti-TB medications were administered via nasogastric tube on pharmacokinetic sampling days, only if the child refused to swallow medications orally (e.g., young children). One hour after the anti-TB medication dose, antiretroviral medications were administered in HIV-infected children, and a standard breakfast was offered.
Pharmacological assays were performed at the University of Cape Town, Division of Clinical Pharmacology using a validated liquid chromatography tandem mass spectrometry (LC MS/MS) method. This LC MS/MS method involves a simple protein precipitation extraction using 20 μl plasma, followed by isocratic separation on a Poroshell 120EC-C18, 4.6 × 50 mm, 2.7 μm column. A deuterated internal standard, Linezolid-d3, is used to monitor the method across a calibration range of 0.100 μg/ml (LLOQ) to 30 μg/ml (ULOQ). The method performed well over the period of analysis, with an accuracy ranging from 96.6% to 98.7% and a precision estimate of less than 7.2% (% CV) over all 3 quality control (QC) sample concentrations (QC low at 0.25 μg/ml, QC medium at 12 μg/ml, and QC high at 24 μg/ml).
All children had regular clinical and laboratory safety assessments, including a full blood count monthly for the first 6 months, then every 2 months or as clinically indicated until completion of treatment. All adverse events were recorded and were assessed for attribution to linezolid based on the judgment of the investigator or subinvestigator, who was not blinded to the presence or duration of linezolid treatment, and graded for severity. In MDRPK1, events were graded according to standard Division of AIDS (DAIDS) criteria (“DAIDS Table for Grading the Severity of Adult and Pediatric Adverse Events” version 1.0, December 2004, updated August 2009) [25]. In MDRPK2, events were graded according to the updated DAIDS Table for Grading the Severity of Adult and Pediatric Adverse Events (version 2.0, November 2014 [26]; corrected version 2.1, July 2017 [27]). There were no differences between these versions for most of the adverse events that are relevant to linezolid, including low platelets, low white blood cell count, and peripheral neuropathy; the grading for these events is shown in S1 Table. There were slight differences in the way low haemoglobins were graded, which are summarised separately in S2 Table. The minor differences in low haemoglobin grading are unlikely to have meaningfully changed the results or the study’s overall conclusions.
The full MDRPK2 study protocol Statistical Considerations section is included as S1 Text. The analyses described here and in the following section are consistent with the general statistical approach in the protocol. Baseline characteristics were presented with descriptive statistics. Weight-for-age (WAZ) and height-for-age (HAZ) z-scores were calculated using British reference values, as WHO references only include children <10 years of age [28]. Only children receiving linezolid as a component of their routine drug-resistant TB treatment regimen were included in safety analyses (i.e., not children only receiving single-dose linezolid). Adverse events at least possibly related to linezolid were presented by grade, and the rate of events per person-time of observation were calculated. The development of a pharmacokinetic-pharmacodynamic model to assess the relationship between linezolid exposures and adverse events was planned after the study was fully enrolled and completed. At the time of the current analysis, this was explored using the analyses described below. The median (interquartile range [IQR]) linezolid area under the concentration time curve from 0 to 24 hours at steady state (AUC0–24ss) was reported and compared by the presence of linezolid-related adverse effects using the Wilcoxon rank sum test. Linezolid exposure was further defined using AUC0–24ss and split into 2 groups using the median AUC0–24ss, 107 mg/L × h. Survival curves for time to any linezolid-related adverse effect and time to grade 3 or 4 linezolid-related adverse effects were displayed using Kaplan-Meier plots. Patients were censored at the time of the relevant linezolid-related event or when linezolid was discontinued. The log-rank test was used to assess whether linezolid exposure was associated with the time to linezolid-related grade 3/4 or any-grade adverse events. If the proportional hazards assumption was violated, log-rank test was reported along with a comparison of the restricted mean survival times (RMSTs) [29, 30]. During the peer review process, these analyses were repeated to assess associations between the minimum linezolid concentrations (Cmins) and linezolid-related adverse effects. A linezolid Cmin of 2 mg/L, suggested as a threshold value for risk of adverse events [31], was used to separate groups in survival curves.
Pharmacokinetic data were characterised based on a population nonlinear mixed-effects modelling approach using the software NONMEM 7.41 (ICON Development Solutions, Ellicott City, Maryland). The method of estimation used was the first order conditional estimation (FOCE) method with the option ‘interaction’. Between-subject variability (BSV) was modelled exponentially, and the residual error was described using a combination of an additive and proportional error model. Inter-occasion and inter-cohort variability were also evaluated to deal with the differences between both studies and pharmacokinetic occasions. Model building was performed in 2 stages, with the structural model developed first and the covariate analysis done subsequently. The main covariates evaluated in the analysis were weight, height, age, sex, race, HIV status, linezolid administration method (oral versus nasogastric tube), formulation (whole versus crushed versus suspension), linezolid given as single dose versus at steady state, and interactions with concomitant drugs. Covariate identification was done using the stepwise covariate modelling (SCM) through the PsN software (version 4.6.0). This method consists of the stepwise testing of linear and nonlinear covariate-parameter relationships with forward inclusion and backward exclusion approaches with significance levels of 0.05 and 0.01, respectively. Given the modest sample size and the large number of covariates, the analysis was also done with a forward inclusion level of 0.2 and backward exclusion of 0.05. The final inclusion of the identified significant covariates was done taking into account scientific plausibility, statistical significance, and clinical relevance. Both stages of model building were evaluated by the likelihood ratio test, goodness of fit plots, and visual predictive checks (VPCs). R software (version 3.5.1) was used to make the graphical representation, using the xpose package (version 4.6.1) (reference: https://CRAN.R-project.org/package=xpose) for data set checkout and graphical evaluation of the results.
The final model was used to simulate different linezolid doses depending on children’s weights. Model-identified optimal doses were derived based on the following assumptions with respect to formulation availability: (a) independent of formulation, e.g., exact optimal doses, (b) optimal doses with 20 mg/mL suspension, and (c) optimal doses with 600 mg tablet in one-fourth tablet increments. The target linezolid exposure used for dosing simulations was the AUC0–24ss in adults with TB after a 600 mg once-daily dose (110 mg/L.h). This target is based on data on linezolid pharmacokinetics from a completed phase II dose-ranging trial in adults with drug-susceptible TB (LINCL-001, NCT02279875, N = 99) and an ongoing phase III trial in adults with MDR-TB, (NiX-TB, NCT02333799, N = 88) (S4 Table). Two hundred simulations of the pharmacokinetic profile of a paediatric population with weights ranging from 5 to 56 kg, receiving the calculated weight-banded dose regimen, were performed in NONMEM (Icon Development Solutions, Ellicot City, MD), and their AUCs were reported.
This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 STROBE Checklist).
Written informed consent was provided by the parent(s) or legal guardians, and written informed assent was given by participants ≥7 years of age. Ethical approval for the study was provided by the Health Research Ethics Committee of Stellenbosch University (N11/03/059 for MDRPK1 and N15/02/012 for MDRPK2).
Forty-eight children were included in the present analysis, 9 from MDRPK1 and 39 from MDRPK2. Baseline clinical and demographic characteristics by study (MDRPK1, MDRPK2) are shown in Table 1.
Of the 48 children included, 5 participants contributed data from more than 1 pharmacokinetic sampling occasion. Four patients from MDRPK2 contributed full profiles from 2 sampling occasions, and 1 patient from MDRPK1 contributed 1 full profile and 2 partial profiles. The pharmacokinetic profiles of these 5 children are displayed in S1 Fig, including the raw pharmacokinetic data, their individual prediction, and the dose and formulation received on each occasion. No concentrations were below the limit of quantitation. The final pharmacokinetic model consisted of a one-compartment disposition model characterised by clearance (CL) and volume (V) parameters. CL and V included allometric scaling, using the exponents 0.75 and 1, respectively, to account for changes in weight. The rate of absorption (Ka) was constrained to be faster than the rate of elimination (CL/V) in order to avoid flip-flop kinetics during model estimation [32]. BSV was estimated on both CL and V. Inter-occasion and inter-cohort variability were assessed and were not included in the final model due to the lack of significance. This lack of significance may be explained by the small number of pharmacokinetic measurements available at second and third occasions (only 5 out of 48 children). Similarly, the lack of significance when using inter-cohort variability suggests that the populations between cohorts were similar and comparable. No other covariates tested—including age, HIV status, formulation (whole tablets versus crushed tablets versus suspension), WAZ, HAZ, body mass index, administration by nasogastric tube—significantly improved the model fit for any of the pharmacokinetic parameters after accounting for the effect of weight. S3 Table reports the coefficients and the 95% confidence intervals of several covariate models tested during the covariate analysis. Table 2 describes the final model parameters. The final model fit the observed data well, as shown in the VPC (Fig 1). Calculated AUC and maximum plasma concentrations from our study population are described in Table 3.
The proposed weight-banded doses across weights that approximated the emerging AUC targets reported in adults with MDR-TB receiving a 600 mg once-daily dose are shown in Table 4. The expected linezolid exposures (AUC0–24ss) across weights from simulations using the final model and this weight-banded dosing strategy are shown in Fig 2.
Seventeen children were included in the safety analysis (median age 4.7 years; range 0.6 to 15.3) and followed for a median duration of 17.7 months on linezolid (IQR 7.5 to 19.7). Ten patients (59%) experienced at least 1 adverse event possibly attributed to linezolid, the most frequent being low haemoglobin, with 12 events in 10 patients (Table 5). There were three grade 3 and two grade 4 events in 5 (23%) participants, all anaemia. The linezolid dose was adjusted after 4 (23.5%) events; linezolid was temporarily interrupted and restarted after 5 (29.4%) events, and 4 (23.5%) participants permanently discontinued linezolid due to adverse events. For the single episode of peripheral neuropathy, which occurred in a 13-year-old female, linezolid was temporarily interrupted, and the patient was treated with gabapentin, resulting in symptomatic improvement. The linezolid was restarted at a lower dose after symptoms resolved with no sequelae, and symptoms did not recur at the lower dose. We did not identify any events of optic neuropathy.
The median linezolid exposure was associated with occurrence of a grade 3 or 4 linezolid-related adverse event (p = 0.020) but not with occurrence of any grade linezolid-related adverse event (p = 1.000) (Table 6). The Cmin and AUC were highly correlated (pairwise correlation coefficient = 0.7737, p = 0.0003). The median linezolid Cmin was not significantly associated with either any grade (p = 0.696) or with a grade 3 or 4 (p = 0.206) linezolid-related adverse event.
For children with an adverse event, the median (IQR) time to any linezolid-related adverse event was 3.2 months (1.8 to 13.9) (n = 10), and time to grade 3 or 4 linezolid-related adverse event was 2.4 months (1.8 to 4.7) (n = 5). The Kaplan-Meier survival curves for time to linezolid-related adverse events are displayed in Fig 3. The time to grade 3 or 4 linezolid-related adverse event differed by linezolid exposure (log-rank p = 0.0121), but not the time to any linezolid-related adverse event (log-rank p = 0.5895, RMST p = 0.344) (Fig 4). Because there was crossing of the survival curves for time to any grade linezolid-related adverse event by linezolid exposure (Fig 4A), we censored the curve at 360 days, but there was still no difference (log-rank p = 0.360). Kaplan-Meier survival curves for time to linezolid-related adverse events by Cmins are shown in S2 Fig. There were no differences by linezolid Cmin in time to any grade (log-rank p = 0.9267) or grade 3 or 4 (log-rank p = 0.5694) linezolid-related adverse event.
This study provides data on linezolid pharmacokinetics in children with TB to inform evidence-based dosing. Linezolid-related adverse events were frequent and occasionally of high grade (grade 3 or 4) in children treated long term for MDR-TB.
The linezolid plasma AUCs seen in our cohort were higher than expected compared to previously published values in children (58 mg/L.h for children 3 months to 11 years of age, and 95 mg/L.h for 12 to 17 years of age following a 10 mg/kg dose) [19]. The previously published paediatric data reported on single intravenous doses only, so direct comparison to our data is challenging. In certain age groups, exposures in our paediatric cohort were even higher than expected target exposure in adults. These differences could be related to previously unidentified drug-drug interactions with other anti-TB drugs. No association was found between any pharmacokinetic parameters and concomitant drugs in our study; however, the study may have been underpowered to detect these. Linezolid has complex metabolism, and as previously noted, there is substantial variability in linezolid pharmacokinetics between individuals and by disease state [17, 18, 33, 34]. Differences in bioavailability, including formulation effects, should also be considered. The formulations used in our study were all nongeneric, as used in most adult studies to date, and no differences were observed in pharmacokinetic parameters by formulation administration (whole versus crushed versus suspension) in our cohort.
Our study combined linezolid pharmacokinetic data after both single and multiple doses. Usually, pharmacokinetic parameters (CL, V of distribution, and absorption rate constant) derived based on single-dose pharmacokinetic data can predict well multiple-dose pharmacokinetics, if the pharmacokinetics are linear and there are no time-dependent trends, such as autoinduction or nonlinear elimination. This was the case for linezolid. The multiple-dose data confirmed the absence of these trends in this paediatric population, and the single-dose data were valuable in informing the relationship between children’s CL and weight and/or age, which forms the basis for establishing the dosing algorithm.
Additional data on linezolid pharmacokinetics in children treated for MDR-TB are important to confirm all of these findings.
At the linezolid doses used in this study, 10 of 17 children treated long term experienced a linezolid-related adverse event, all of whom had anaemia in addition to some other, less frequent events. This is similar to adult data and to a previous summary of published case series and case reports in which 9 of 18 children had a linezolid-related event [9]. In our study, regular monitoring was able to identify anaemia at mild grades in most children. The approach of temporarily interrupting the linezolid until the haemoglobin had improved, followed by re-introduction at a lower dose (usually half the previous dose), was generally well tolerated. Five children experienced a grade 3 or 4 anaemia, 4 of whom permanently discontinued linezolid. Although all patients recovered from their anaemia without sequelae, the frequency and severity of these events highlights the importance of careful monitoring of haematological parameters during long-term linezolid treatment. In our study, once haemoglobin values began to drop, they often fell rapidly, so the first signs of a falling haemoglobin should prompt more frequent testing, and there should be a low threshold for temporarily interrupting doses, depending on the importance of linezolid to the efficacy of the regimen. The cumulative incidence of adverse events increased with longer duration of treatment; however, there were events detected even in the first 60 days, including grade 3 and 4 events. Reducing the duration of linezolid treatment would likely reduce, but not eliminate, the risk of adverse events, and careful safety monitoring will be necessary in future clinical trials and in routine care, regardless of the duration of linezolid therapy. Reducing the linezolid dose would be expected to reduce its activity and may increase the risk of acquired resistance. The risk to children, who tend to have paucibacillary disease, would likely be low, especially when the child has already been on treatment for weeks or more with effective companion drugs in the regimen, and the bacillary burden would have been substantially reduced. However, there is still uncertainty about the risk of acquired resistance and the efficacy of these lower doses of linezolid, and an improved understanding of how to weigh up this risk versus benefit is an area for future research.
Grade 3 or 4 events were strongly associated with higher linezolid exposures. The lower doses we have proposed may reduce this risk. Therapeutic drug monitoring may also be a consideration to explore further to reduce the risk of these more severe events.
Peripheral neuropathy was detected in only one child, a 13-year-old female, and improved with temporary discontinuation of linezolid and gabapentin treatment, and it ultimately resolved without sequelae after recommencing a reduced linezolid dose. To our knowledge, there is no reason to believe that children do not develop linezolid-related peripheral neuropathy, although the risk may be lower than in adults. Peripheral neuropathy presents with symptoms of pain, numbness, and tingling in a stocking-glove distribution, and signs of weakness, reduced deep-tendon reflexes, and loss of sensation of vibration, proprioception, and temperature. All of these signs and symptoms can be challenging to elicit in young children. To our knowledge, there are no validated clinical approaches for screening for peripheral neuropathy that can be used in young children. The gold standard for diagnosis is nerve conduction studies, which are invasive and specialised tests not widely available in high–TB-burden settings. The implication is that peripheral neuropathy may be difficult to identify in young children, and thus there is a risk of underdiagnosis. This is an important consideration when weighing the risk to benefit for the use of linezolid in children with MDR-TB. We did not identify any other serious, more rare events such as optic neuropathy or lactic acidosis. This is reassuring but should be interpreted cautiously, as the number of children included here is modest, and formal ophthalmologic examinations were not done. Lactate levels and pancreatic enzymes were not monitored, which is also a limitation. However, in no children were clinical symptoms or signs suggestive of optic neuropathy, lactic acidosis, or pancreatitis reported or suspected.
Using the final model, our simulations identified weight-banded doses that would achieve the proposed target exposure, based on emerging data in adults with MDR-TB receiving a 600 mg daily dose (Table 4, Fig 2). Doses used for children in routine care will be constrained by available formulations. A 20 mg/mL suspension exists but is not widely available due to cost, variable supply, or lack of in-country registrations, and most settings only have access to the unscored 600 mg tablet. Pragmatic doses using the available formulations result in less optimal exposures in some weight bands, which may increase the risk of adverse effects (Table 4, Fig 2). This should not limit the use of linezolid in children when indicated; however, more flexible, child-friendly formulations are urgently needed.
These new proposed doses are lower than those used in this study and are currently used in routine care for MDR-TB in most settings. Lower doses in children would likely reduce the risk of adverse events, which were frequent in this cohort, although this would need to be confirmed in further studies. Because exposures with these doses in children would be expected to approximate those in adults after a 600 mg daily dose, the efficacy would be expected to be as good or better in most children compared to adults receiving this dose, as children often have paucibacillary TB. Careful evaluation of the pharmacokinetics, safety, and treatment outcomes in children treated with these proposed new, weight-banded doses of linezolid is needed.
The linezolid exposure targeted clearly has an impact on paediatric dosing studies and recommendations. The target exposure chosen here (110 mg/L.h) is based on robust data from 2 adult TB trials. This target was chosen because it was taken from recent studies that also included efficacy and toxicity data. The optimal linezolid target exposure and the optimal dosing strategy for linezolid has not yet been established in adults. An ongoing phase III trial (ZeNix, NCT03086486) is evaluating the efficacy and safety of multiple doses and durations of linezolid in combination with bedaquiline and pretomanid in adults with MDR-TB. ZeNix will provide important data to inform future linezolid dosing recommendations in adults. A free AUC/minimum inhibitory concentration (MIC) ratio of >100 has been proposed for efficacy based on hollow fibre model data [35, 36]. A linezolid free-AUC/MIC ratio of <96 or a trough <2 mg/L have been proposed as targets to reduce the risk of adverse effects [31, 35, 36]. However, more data are needed to understand how clinically relevant these targets are in adults and in children of different ages. We are planning additional work to better characterise the relationship between linezolid pharmacokinetics and adverse events using current and future data and pharmacokinetic-pharmacodynamic modelling. Due to its excellent penetration into CSF, linezolid may be an important component of treatment regimens for TB meningitis. Higher doses of linezolid than used for other forms of TB may be needed in order to achieve target exposures in CSF. This is an important area for future research.
There is a growing awareness of the importance of drug exposure at the site of disease, which is influenced by penetration of drugs into different areas of TB lesions [37]. Targeting plasma AUC/MIC does not take this key aspect into consideration. Targeting drug exposures that are linked with efficacy in TB patients with the spectrum of TB disease and lung pathology does incorporate this aspect and, in our estimation, is a more appropriate target given the current state of knowledge. Regardless, paediatric linezolid dosing recommendations may need to evolve as additional data on optimal dosing in adults become available.
The study has a number of important limitations. Although this is a relatively large study of intensively sampled pharmacokinetics in children, the numbers of included children remain small, especially regarding potentially relevant subgroups. There were few very young and small children included in this analysis, and none under 6 months of age. There are data in neonates and young infants indicating that the CL of linezolid is associated with postnatal age, with children <8 days of age having lower CL than those 8 days to 12 weeks of age (i.e., CL rapidly increased in the first week of life) [38]. Our proposed doses for smaller children should be interpreted with caution because infants younger than 6 months were not included. For clinicians in the field responsible for caring for the occasional young, small child with MDR-TB, some practical guidance on dosing—even with inherent limitations—is, however, still valuable. Additionally, there were few children with HIV infection or undernutrition. We are unaware of clinically significant drug-drug interactions between antiretrovirals and linezolid or of other previously described effects of HIV or undernutrition on linezolid pharmacokinetics. However, these are important covariates to consider, and our small data set is underpowered to detect effects of HIV infection or undernutrition that could be clinically significant. A larger study, or pooling of data, would help better characterise the effect of any of these important covariates. Additionally, we did not detect an effect of formulation (whole tablets versus crushed tablets versus suspension). The small sample size and confounding by age (young children received suspension and older children whole tablets) limited our ability to detect such formulation effects if they existed.
There were also limitations to our safety analysis. The modest number of children receiving long-term linezolid and contributing to safety data limits the precision of our estimates of the proportion of children with adverse events detected. The fact that more rare events, such as optic neuropathy, were not detected does not mean they might not be observed with larger numbers of patients. HIV-infected patients receiving linezolid may be at a higher risk of adverse effects because of overlapping mitochondrial toxicity with some antiretroviral drugs, especially the non-nucleoside reverse transcriptase inhibitors. This remains an important area for future research.
This study provides urgently needed data on linezolid pharmacokinetics and prospective safety data in children with TB. Linezolid exposures were higher than expected, and adverse events were common and occasionally severe. Based on these data, we have proposed, for the first time, paediatric weight-banded doses approximating a 600-mg–daily adult linezolid dose. Clear guidance on the optimal dose of linezolid for children with TB is urgently needed given its growing importance in MDR-TB treatment regimens. These data have informed revised WHO-recommended dosing in children.
|
10.1371/journal.pgen.1000604 | Nuclear Calcium Signaling Controls Expression of a Large Gene Pool: Identification of a Gene Program for Acquired Neuroprotection Induced by Synaptic Activity | Synaptic activity can boost neuroprotection through a mechanism that requires synapse-to-nucleus communication and calcium signals in the cell nucleus. Here we show that in hippocampal neurons nuclear calcium is one of the most potent signals in neuronal gene expression. The induction or repression of 185 neuronal activity-regulated genes is dependent upon nuclear calcium signaling. The nuclear calcium-regulated gene pool contains a genomic program that mediates synaptic activity-induced, acquired neuroprotection. The core set of neuroprotective genes consists of 9 principal components, termed Activity-regulated Inhibitor of Death (AID) genes, and includes Atf3, Btg2, GADD45β, GADD45γ, Inhibin β-A, Interferon activated gene 202B, Npas4, Nr4a1, and Serpinb2, which strongly promote survival of cultured hippocampal neurons. Several AID genes provide neuroprotection through a common process that renders mitochondria more resistant to cellular stress and toxic insults. Stereotaxic delivery of AID gene-expressing recombinant adeno-associated viruses to the hippocampus confers protection in vivo against seizure-induced brain damage. Thus, treatments that enhance nuclear calcium signaling or supplement AID genes represent novel therapies to combat neurodegenerative conditions and neuronal cell loss caused by synaptic dysfunction, which may be accompanied by a deregulation of calcium signal initiation and/or propagation to the cell nucleus.
| The dialogue between the synapse and the nucleus plays an important role in the physiology of neurons because it links brief changes in the membrane potential to the transcriptional regulation of genes critical for neuronal survival and long-term memory. The propagation of activity-induced calcium signals to the cell nucleus represents a major route for synapse-to-nucleus communication. Here we identified nuclear calcium-regulated genes that are responsible for a neuroprotective shield that neurons build up upon synaptic activity. We found that among the 185 genes controlled by nuclear calcium signaling, a set of 9 genes had strong survival promoting activity both in cell culture and in an animal model of neurodegeneration. The mechanism through which several genes prevent cell death involves the strengthening of mitochondria against cellular stress and toxic insults. The discovery of an activity-induced neuroprotective gene program suggest that impairments of synaptic activity and synapse-to-nucleus signaling, for example due to expression of Alzheimer's disease protein or in aging, may comprise the cells' own neuroprotective system eventually leading to cell death. Thus, malfunctioning of nuclear calcium signaling could be a key etiological factor common to many neuropathological conditions, providing a simple and unifying concept to explain disease- and aging-related cell loss.
| Physiological levels of synaptic activity are required for neurons to survive [1]. Activity-dependent neuroprotection is induced by calcium entry through synaptic NMDA receptors and requires that calcium transients invade the cell nucleus [2]–[6]. Procedures that interfere with electrical activity and compromise NMDA receptor function or nuclear calcium signaling can have deleterious effects on the health of neurons both in vitro and in vivo. For example, blockade of NMDA receptors in vivo following intraperitoneal injections of the NMDA receptor antagonist MK-801 into seven day-old rats triggers, within 24 hours, a wave of apoptotic neurodegeneration in many brain regions, including the parietal and frontal cortex, the thalamus and the hippocampus [7]. Likewise, the selective blockade of nuclear calcium signaling prevents cultured hippocampal neurons from building up anti-apoptotic activity upon synaptic NMDA receptor stimulation [2],[3],[6]. Conversely, enhancing neuronal firing and synaptic NMDA receptor activity is neuroprotective: networks of cultured hippocampal neurons that have experienced periods of action potential bursting causing calcium entry through synaptic NMDA receptors are more resistant to cell death-inducing conditions [2]–[4]. Moreover, stimulating synaptic activity in vivo by exposing rats to enriched environments reduces spontaneous apoptotic cell death in the hippocampus and protects against neurotoxic injuries [8].
Neuronal activity and NMDA receptor-induced calcium signaling pathways can suppress apoptosis and promote survival through two mechanistically distinct processes. One process is independent of on-going gene transcription and involves the phosphatidylinositide 3′-OH kinase (PI3K)-AKT signaling pathway which promotes survival while neurons are being electrically stimulated [3]. However, the principal pathway conferring long-lasting neuroprotection requires the generation of calcium transients in the cell nucleus [2]–[6],[9]. The aim of this study was to investigate how nuclear calcium promotes neuroprotection. Using tools to selectively block nuclear calcium signaling in hippocampal neurons in conjunction with microarray technologies and bioinformatics, we uncovered a genomic survival program that is induced by calcium transients in the cell nucleus. The core components of this program, referred to as Activity-regulated Inhibitor of Death (AID) genes, can provide neurons with a broad-spectrum neuroprotective shield against cell death.
To identify genes regulated by nuclear calcium signaling in hippocampal neurons, we carried out comparative whole-genome transcriptional profiling. Hippocampal neurons were infected with a recombinant adeno-associated virus (rAAV) expressing either the calmodulin (CaM) binding-peptide, CaMBP4 (rAAV-CaMBP4 [6]) or β-galactosidase (rAAV-LacZ) as a control. CaMBP4 is a nuclear protein that contains 4 repeats of the M13 calmodulin binding peptide derived from the rabbit skeletal muscle myosin light chain kinase; it binds to and inactivates the nuclear calcium/CaM complex [10]. Inhibition of nuclear calcium signaling with CaMBP4 in hippocampal neurons blocks synaptic activity-evoked CREB-mediated transcription and prevents the induction of a genomic neuroprotective program by neuronal activity [3],[6]. Hippocampal neurons were stimulated by exposing the network to the GABAA receptor antagonist, bicuculline. GABAergic interneurons, which represent about 11% of the neuron population, impose a tonic inhibition onto the network [11]. Removal of GABAAergic inhibition with bicuculline leads to action potential (AP) bursting, which stimulates calcium entry though synaptic NMDA receptors, generates robust cytoplasmic and nuclear calcium transients, induces CREB-dependent transcription, and strongly promotes neuronal survival [2]–[4],[6],[11],[12]. RNA isolated from these hippocampal neurons was used for microarray analyses on Affymetrix GeneChips. The Affymetrix microarray data were analyzed by a two-step process; details of the data analysis are described in Text S1. First, we determined all genes induced or repressed by AP bursting (which gives rise to robust nuclear calcium signals) in control-infected (rAAV-LacZ) hippocampal neurons. A threshold of 2.0 fold was chosen, which, given that microarray data are compressed and generally underestimates the fold differences in gene expression [6],[13], filters out genes that are likely to undergo signal-induced changes in their expression that are in the range of at least 2.5 to 3 fold. This analysis revealed 302 genes that were induced and 129 genes that were repressed in rAAV-LacZ infected hippocampal neurons 4 hours after the induction of AP bursting. A color-coded map provides an overview of these 431 AP bursting-regulated genes (Figure 1A). A comparison of the genes identified in this study using rAAV-LacZ infected hippocampal neurons with the pool of activity-regulated genes described in a previous study [6] revealed a high degree of overlap. However, due to the higher threshold applied in this analysis (2 fold vs. 1.5 fold change used in our previous study [6]), the current analysis filtered out fewer genes.
In a second analysis step, we compared the expression of genes regulated by AP bursting in hippocampal neurons infected with rAAV-LacZ and in hippocampal neurons infected with rAAV-CaMBP4 to block the nuclear calcium/CaM complex. The regulation of a gene was considered dependent on nuclear calcium signaling if based on the microarray data its induction or repression by AP bursting is reduced by at least 40% in rAAV-CaMBP4 infected neurons compared to rAAV-LacZ infected neurons. We found that 183 genes (plus Btg2 and Bcl6; see below) fulfill these criteria (Figure 1B); a list of those genes including their fold changes following AP bursting and percent inhibition by CaMBP4 is given in Table 1. We are likely to underestimate the total number of genes regulated by nuclear calcium signaling for several reasons. First, our screen did not identify genes controlled by the downstream regulatory element antagonist modulator (DREAM) [14]. DREAM is a transcriptional repressor that can directly bind calcium through its EF hands. In its calcium-bound form, DREAM is released from the DNA allowing transcription to be activated in a nuclear calcium-dependent but calmodulin-independent manner [14]. Second, our analysis was restricted to one time point (i.e. 4 hours after induction of AP bursting) and therefore it is possible that we have missed genes that have peak expression levels at time points significantly earlier or later than 4 hours. Although those genes may also show changes in expression levels at 4 hours after AP bursting, if this induction was less than two fold they were not scored as induced in our study. Third, a possible regulation by nuclear calcium could also have been missed because for genes weakly induced by AP bursting, the accuracy of the microarray data–based assessment of nuclear calcium regulation decreases and the necessary statistical criteria may not be met. Indeed, these conditions apply to two neuronal survival-promoting genes, Btg2 and Bcl6 [6]. Btg2 is induced with fast kinetics after AP bursting; the induction peaks at 2 hours [about 8 fold induction based on microarray data and 17 fold induction based on quantitative reverse transcriptase (QRT)-PCR analysis [6]], whereas at the 4 hours time point the change in expression based on microarray data relative to unstimulated control is only about 2 fold [6]. In the case of Bcl6, the fold changes observed at 4 hours after the on-set of AP bursting are about 1.5 fold based on microarray data analysis and about 2.5 fold based on QRT-PCR analysis [6]. Both, Btg2 and Bcl6, are regulated by nuclear calcium signaling [6] and have therefore been included in the list of nuclear calcium-regulated genes (Table 1). The identified nuclear calcium-regulated gene pool comprises 43% of all activity-regulated genes. It contains a large variety of gene products with different catalytic and binding activities (Figure 1C).
Because nuclear calcium signals evoked by AP bursting strongly promote neuronal survival [2]–[4],[6], we next aimed at identifying putative neuroprotective genes present in the nuclear calcium-regulated gene pool. Using a Gene Ontology (GO) analysis with the GO term ‘Apoptosis’ and a literature search we were able to identify 20 nuclear calcium-regulated genes that have been implicated in cell death/survival processes in non-neuronal or neuronal cells (Table 1). Our attention was drawn in particular to 8 nuclear calcium-regulated genes that, based on the microarray data, showed a very robust induction (more than 10 fold changes) of expression following neuronal activity. This includes 6 genes with known or putative functions in the cell nucleus (Atf3, GADD45β, GADD45γ, Interferon activated gene 202b (Ifi202b), Npas4, and Nr4a1) and 2 genes encoding secreted proteins (Inhibin β-A and Serpinb2). Atf3 (activating transcription factor 3) is a member of the ATF/cAMP-responsive element-binding protein (CREB) family of transcription factors [15] that has been implicated in survival processes in neuronal and non-neuronal cells [16]–[18]. The family of growth arrest and DNA damage-inducible 45 (GADD45) genes comprises three members (GADD45α, GADD45β and GADD45γ) that are expressed in response to stress stimuli and DNA damage. GADD45 genes have been implicated in DNA excision-repair processes [19],[20] but may also contribute to gene transcription via a process that involves DNA demethylation [21]. Ifi202b belongs to the interferon-activated p202 gene family that plays a role in cell survival and the regulation of caspase activation [22],[23]. Npas4 (also known as NxF) is a member of the basic helix-loop-helix/Per-Arnt-Sim (bHLH-PAS) homology protein family [24]; it functions as a transcriptional regulator with possible roles in cell survival and differentiation [25],[26]. Nr4a1 (also known as nur77 or NGFIB) is an orphan nuclear hormone receptor with possible pro-apoptotic and anti-apoptotic functions [27]–[29]. Serpinb2 [also known as plasminogen activator inhibitor type-2 (PAI-2)] is a serine proteinase inhibitor that can influence cell proliferation, differentiation and cell death [30],[31]. Inhibin β-A is a member of the transforming growth factor (TGF)-β superfamily [32] that can protect human SH-SY5Y neuroblastoma cells from chemical-induced death and may mediate neuroprotective actions of basic FGF [33],[34]. We considered these 8 genes plus the previously identified pro-survival gene Btg2 (which is robustly induced by neuronal activity in a nuclear calcium-dependent manner [6]) as the core components of the putative neuroprotective gene program and refer to this set of genes hereafter as Activity-regulated Inhibitor of Death (AID) genes (Table 1; AID genes are boxed).
Using QRT-PCR, we confirmed the regulation of each AID gene by AP bursting and nuclear calcium signaling; the regulation of Btg2 by nuclear calcium signaling has been established previously [6]. The expression of GADD45β, GADD4γ, and Nr4a1 increased about 20 to 35 fold after AP bursting, which were the weakest inductions among this group of genes (Figure 2). Significantly higher fold changes relative to unstimulated control in either uninfected or rAAV-LacZ infected hippocampal neurons were observed for Atf3 (63±4 fold, uninfected; 49±4 fold, rAAV-LacZ), Npas4 (203±17 fold, uninfected; 186±9 fold, rAAV-LacZ), Ifi202b (288±19 fold, uninfected; 246±30 fold, rAAV-LacZ), and Inhibin β-A (432±41 fold, uninfected; 388±30 fold, rAAV-LacZ) (Figure 2). For Serpinb2, QRT-PCR revealed a very dramatic, 1839±45 fold increase in expression following AP bursting in uninfected hippocampal neurons and 1781±48 fold increase in rAAV-LacZ infected hippocampal neurons (Figure 2). This is - to the best of our knowledge - the highest fold change ever observed for a signal–regulated gene. For all AID genes, we confirmed the requirement for nuclear calcium signaling for their induction by AP bursting (Figure 2). Most dramatic inhibitions of well over 80 percent were observed for Inhibin β-A (93±3% inhibition), Npas4 (92±3% inhibition), Ifi202b (87±4% inhibition), and Serpinb2 (85±3% inhibition), and Atf3 (80±5%) (Figure 2). We also included prostaglandin-endoperoxide synthase 2 (Ptgs2; also known as Cox2) in the QRT-PCR analysis. Expression of Ptgs2 is robustly induced by neuronal activity (80±24 fold, uninfected; 78±18 fold, rAAV-LacZ), and this induction was inhibited by CaMBP4 by 68±20% (Figure 2). Because there is no available evidence for a role of Ptgs2 in promoting survival of neuronal or non-neuronal cells, this gene served as one of the negative controls in the in vivo survival experiments (see below).
Given the importance of CREB and its co-activator CREB binding protein (CBP) in mediating transcriptional activation by synaptic activity and nuclear calcium signaling [12],[35],[36] and the critical role of CREB in neuronal survival [3],[37],[38], we carried out data base searches to determine whether the nuclear calcium-regulated genes, in particular AID genes, contain putative CREB binding sites. Information retrieved from two databases (the CREB ‘regulon’ (http://saco.ohsu.edu/) [39] and the CREB target gene database (http://natural.salk.edu/creb/) [40] indicated that a large fraction of the nuclear calcium regulated gene pool (56%) and all AID genes except Npas4 contain one or several CREs or CRE-like sequences, suggesting that they could be CREB target genes (Table S1). In addition to CREB, other CBP-recruiting transcription factors may contribute to the regulation of AID genes by nuclear calcium signaling.
We next investigated the role of the nuclear calcium/calmodulin-dependent protein kinase IV (CaMKIV) in the regulation of AID genes. CaMKIV is one important mediator of nuclear calcium/CREB-regulated transcription [36], [41]–[46]. To inhibit CaMKIV activity, we infected hippocampal neurons with a rAAV containing an expression cassette for a kinase-inactive form of CaMKIV (CaMKIVK75E) that functions as a negative interfering mutant of CaMKIV [36],[42],[43]; immunoblot analysis of expression of CaMKIVK75E in hippocampal neurons infected with rAAV-CaMKIVK75E is shown in Figure 3A. We found that in hippocampal neurons infected with rAAV-CaMKIVK75E the induction by AP bursting of all AID genes and induction of Ptgs2 was inhibited (Figure 2). For 5 AID genes (i.e. GADD45β, GADD45γ, Serpinb2, Inhibin β-A, and Ifi202b) and for Ptgs2, the percent inhibition by rAAV-CaMKIVK75E was very similar to the inhibition obtained by the blockade of nuclear calcium signaling using CaMBP4 (Figure 2). CaMKIVK75E was slightly less potent than CaMBP4 in inhibiting induction of Atf3, Npas4, and Nr4a1 (Figure 2), suggesting that targets of nuclear calcium other than CaMKIV may contribute to regulation of these genes by neuronal activity. These results indicate that nuclear calcium-CaMKIV is an important regulatory module of AID genes.
We next investigated the role of AID genes in neuronal survival. We first carried out gain-of-function experiments in which we used rAAV-mediated gene delivery to over-express Flag-tagged AID proteins in cultured hippocampal neurons. Expression of the proteins was assessed with immunoblots using antibodies to the Flag-tag (Figure 3A). Infection rates were determined immunocytochemically and ranged from 80 to 95 percent of the viable neurons (data not shown). We used two types of assays to assess apoptotic cell death: growth factor withdrawal and treatment of cultured hippocampal neurons with a low concentration of staurosporine [2],[6], a classical inducer of apoptotic cell death. We found that compared to control (i.e. non-infected neurons or neurons infected with rAAV-LacZ), cell death induced by either growth factor withdrawal or staurosporine treatment was inhibited in neurons infected with rAAV carrying Flag-tagged AID proteins (Figure 3B). Inhibition of apoptosis ranged from about 30 to 95% for growth factor withdrawal-induced apoptosis and from about 40 to 80% for apoptosis induced by staurosporine (Figure 3B). Over-expression of Atf3, Nr4a1, GADD45β, and GADD45γ yielded the most potent inhibition for growth factor withdrawal-induced apoptosis, whereas expression of Npas4 was most efficient in protecting against staurosporine-induced apoptosis (Figure 3B). These results indicate that AID proteins can confer robust neuroprotection to cultured hippocampal neurons.
We next investigated whether AID genes contribute to activity-dependent survival induced by AP bursting and activation of synaptic NMDA receptors [2]–[4],[6]. For this analysis, we selected three genes (Atf3, GADD45β, and GADD45γ) that protected efficiently against growth factor withdrawal-induced apoptosis and Npas4 that protected efficiently against staurosporine induced apoptosis (see Figure 3B). RNA interference (RNAi) was used to inhibit expression of these genes following synaptic activity. DNA sequences encoding short hairpin RNAs (shRNAs) designed to appropriate target regions were inserted downstream of the U6 promoter of a rAAV vector that also harbors an expression cassette for humanized Renilla reniformis green fluorescent protein (hrGFP) [6] (for details see Text S1). To control for non-specific effects of infections with rAAVs carrying an expression cassette for shRNAs, a rAAV was used that contains a universal control shRNA (rAAV-Control-RNAi), which has no significant sequence similarity to the mouse, rat, or human genome. For all rAAVs carrying an expression cassette for shRNAs, infection rates of 80 to 95 percent of the neuron population were obtained (data not shown). QRT-PCR analysis revealed that RNAi was effective in eliminating induction of Atf3, GADD45β, and GADD45γ, and Npas4 by AP bursting in hippocampal neurons. The inhibition of expression was in the range of 85% for all 4 genes; rAAV-Control-RNAi had no significant effect (Figure 4A). Given the neurotropism of rAAVs used in the study [47], the results also indicate that the induction of these genes occurs in hippocampal neurons and not in glial cells.
To assess activity-dependent survival, apoptotic cells were counted after treatment with staurosporine or withdrawal of growth factors with and without previous periods of neuronal activity (Figure 4B and 4C) [2],[3],[6]. In the staurosporine assays, we detected about 10 to 15% of apoptotic cells in the control condition, which increased to about 60% after treatment (Figure 4B). In the growth factor withdrawal assays, basal cell death was slightly higher (about 20 to 25%) and also increased to about 60% (Figure 4C). Upon subjecting the neurons to a period of 12–16 hours of synaptic activity (induced by bicuculline treatment in the presence of 4-amino pyridine, which increases the burst frequency [2]) prior to growth factor withdrawal or staurosporine exposure, far fewer cells underwent apoptosis (Figure 4B and 4C). As shown previously, this activity-dependent survival is triggered by calcium entry into the neurons through synaptic NMDA receptors and involves nuclear calcium signaling [2]–[4],[6]. In uninfected neurons and neurons infected with rAAV-Control-RNAi, we observed the typical stimulus-induced increase in apoptotic cells; following the period of synaptic activity prior to staurosporine exposure or growth factor withdrawal, the stimulus-induced cell death was reduced (Figure 4B and 4C). In contrast, in neurons infected with rAAV-Atf3-RNAi, rAAV-GADD45β-RNAi, rAAV-GADD45γ-RNAi, and rAAV-Npas4-RNAi the basal level of cell death was slightly elevated and activity-dependent survival was severely compromised (Figure 4B and 4C). These results indicate that the AID genes Atf3, GADD45β, GADD45γ and Npas4 are important for neuronal survival and represent key components of the synaptic NMDA receptor-induced genomic neuroprotective program.
Virtually all cell death processes involve the deregulation of mitochondrial functions. One important early event in excitotoxic cell death is the collapse of the mitochondria membrane potential and the shift in the mitochondrial membrane permeability, known as mitochondrial permeability transition (MPT) [48]–[50]. Using Rhodamine 123 (Rh123) imaging techniques to monitor the mitochondrial membrane potential, we have recently shown that one mechanism through which down-regulation of the tumor suppressor gene, p53, or increasing expression of Btg2 can enhance the survival of hippocampal neurons involves inhibition of the NMDA-induced break-down of the mitochondrial membrane potentials [51]. We therefore investigated whether the identified AID genes can also act through a process that guards mitochondria against toxic insults. Rh123 imaging of uninfected hippocampal neurons and hippocampal neurons infected with rAAV-LacZ revealed the typical increase in Rh123 fluorescence after application of NMDA (30 µM), which is indicative of the break—down of mitochondrial membrane potential [2],[51],[52]. In contrast, in hippocampal neurons that had been infected with rAAV to express the AID genes Npas4, Inhibin β-A, Ifi202b, and Nr4a1, or to express the previously identified neuroprotective gene, Bcl6 [6], the NMDA-induced loss of mitochondrial membrane potential occurred with slower kinetics and reached significantly lower magnitudes (Figure 5A–5C). A quantitative analysis of the imaging data revealed that expression of Npas4 and Bcl6 had the largest inhibitory effects on the NMDA-induced break—down of mitochondrial membrane potential (Figure 5B and 5C). The inhibitions by Inhibin β-A, Ifi202b, and Nr4a1 were smaller but also significant, whereas no significant reductions were observed for Atf3, GADD45β, GADD45γ, and Serpinb2 under the conditions used (Figure 5B and 5C). These results suggest that one converging point common to several AID genes is the mitochondria, which are rendered more resistant against death signal-induced dysfunction.
We next analyzed the neuroprotective activity of AID genes in vivo. Stereotaxic injection was used to deliver rAAVs carrying expression cassettes for AID genes or appropriate negative controls (i.e. rAAV-LacZ and rAAV-Empty) to the dorsal hippocampus of male Sprague-Dawley rats weighing 230 to 250 g. We also included Ptgs2 in our in vivo analysis as an additional negative control. Expression of Ptgs2 is robustly induced by neuronal activity in a nuclear calcium/CaMKIV dependent manner (Figure 2). However, there is no available evidence for a role of Ptgs2 in promoting survival of neuronal or non-neuronal cells and we therefore expected that expression of Ptgs2 in vivo would not provide neuroprotection. Two weeks after viral delivery, the rats were injected intra-peritoneally with kainic acid (KA), which induces seizures leading to cell death in the hippocampus [53]. The animals were sacrificed three days after KA injection. The brains were removed, cut into slices, and stained with the histofluorescent label, Fluoro-Jade C, which serves as a very reliable marker for degenerating neurons [54]. The slices were immunostained with antibodies to the neuronal marker, NeuN, and antibodies to the Flag tag to detect the over-expressed proteins. We found widespread KA-induced cell death in the CA1 region of the hippocampus of animals injected with rAAV-Empty, rAAV-LacZ, and rAAV-Ptgs2, as well as in the CA1 region of the non-injected side of the hippocampus (Figure 6). In contrast, expression of AID genes in the CA1 area of the injected side of the hippocampus protected against KA-induced cell death (Figure 7). Quantification of the Fluoro-Jade C signals revealed inhibitions of KA-induced cell death of 85±7% (Atf3), 87±5% (Btg2), 92±6% (GADD45β), 93±2 (GADD45γ), 96±1% (Ifi202b), 70±8% (Inhibin β-A), 92±3% (Npas4), 90±7% (Nr4a1), and 96±1% (Serpinb2); over-expression of Ptgs2 or LacZ did not inhibit KA-induced cell death (Table 2). Expression of Btg2, Ifi202b, and Serpinb2 consistently led to neuroprotection also on the contralateral (i.e. non-injected) side (Figure 7, Table 2). This could be due to secretion of the neuroprotective protein, which may be the case for the serine proteinase inhibitor, Serpinb2. It is also conceivable that expression of neuroprotective proteins in the processes of infected neurons, which project to the contralateral hippocampus, can promote survival of target neuron; this may be the case for flag-tagged Ifi202b which is readily detectable in the hippocampus of the non-injected, contralateral side (see Figure 7).
Neuronal activity boosts neuroprotection. Our study has identified the key players in this process. Using a strategy that combined the identification of all genes controlled by neuronal activity and nuclear calcium signaling with subsequent bioinformatics filtering procedures, a set of genes was unearthed that provides neurons with a robust neuroprotective shield. The core neuroprotective program contains 9 genes, which are referred to as AID genes.
Activity-dependent, long-lasting neuroprotection as well as other adaptive responses in the nervous system require the dialogue between the synapse and the nucleus. There is growing evidence to suggest that calcium signals propagating from their site of activation at the plasma membrane towards the cell soma and the nucleus are key mediators of synapse-to-nucleus communication. Nuclear calcium, most likely acting via nuclear calcium/calmodulin dependent protein kinases, controls CREB/CBP-dependent transcription [12], [35], [36], [43]–[46] and is thought to regulate genomic programs critical for neuronal survival, synaptic plasticity, memory formation, and emotional behavior [3], [6], [55]–[59]. In this study, we identified the nuclear calcium-regulated gene pool in hippocampal neurons. The large number of 185 genes induced or repressed by nuclear calcium signaling is not unexpected. A genome-wide analysis revealed that CREB – the principal target of nuclear calcium signaling - can occupy between 4000 and 6000 promoter sites in the rat or human genome, although the transcription of only a subset of those genes is signal-induced in a given cell type perhaps due to preferential recruitment of CBP [39],[40],[60]. Indeed, 56% of nuclear calcium-regulated genes and all AID genes except Npas4 are known or putative CREB targets (Table S1), underscoring the importance of the nuclear calcium-CREB axis in neuronal survival.
The AID genes characterized in this study fall into two functional categories: regulators of gene transcription and secreted proteins. Although they may act in concert to collectively provide full neuroprotection, over-expression of individual AID genes is sufficient to promote survival. Moreover, RNAi-based loss-of-function experiments indicate that the selective reduction of individual AID genes can compromise the activity-induced build-up of a neuroprotective shield. These results could be explained by a possible convergence of AID genes on one or a small number of targets that execute protection. Given that 7 out of 9 AID genes are putative regulators of gene expression, the existence of common target genes that are part of a ‘second wave’ transcriptional response is conceivable. Under physiological condition, a signal-regulated, coordinate induction of transiently expressed genes may be required for the regulation of common targets; thus interference with one or a small number of AID genes would disturb the system. However, high-level, constitutive expression of individual AID genes may be sufficient to activate or inactivate down-stream regulators of survival. The regulation of a putative common target could involve direct trans-activation by AID gene products through binding to the target genes' promoter elements, although other modes of regulation (such as control of mRNA or protein stability or activity-regulating post-translational modifications) are conceivable and may involve secondary responses triggered by AID genes. Additional components of such a regulatory network may include other survival-promoting transcriptional regulators such as C/EBPβ [61]–[64], or the secreted AID genes Serpinb2 and Inhibin β-A, or Bdnf [65] (see Table 1); these genes may contribute through transcription-dependent and transcription-independent processes to the funneling of information flow and the reduction of complexity to few molecules implementing neuroprotection. Mitochondria, which are vital for supplying the energy required to maintain life, may be the end-point of neuroprotective processes. In this study we show that expression of AID genes renders the mitochondria of hippocampal neurons more resistant to harmful conditions. Thus, neuroprotection may ultimately guard mitochondria against stress and toxic insults to prevent mitochondrial dysfunction.
The finding that activation of synaptic NMDA receptors and calcium signaling to the cell nucleus builds up a strong and lasting neuroprotective shield may change our view of neurodegenerative disorders and cell death associated with aging. Proper functioning of the endogenous neuroprotective machinery requires a sequence of events that can be disturbed at the level of synaptic transmission, synaptic NMDA receptor activation, the generation of calcium signal and their propagation to the cell nucleus, and the regulation of gene transcription. Malfunctioning of calcium signaling towards and within the cell nucleus may lead to neurodegeneration and neuronal cell death. In Alzheimer's disease cell death may be caused by compromised endogenous neuroprotection due to impaired synaptic transmission and synapse loss caused by A-β or changes in calcium homeostasis or calcium signaling in neurons expressing mutant presenilin-1 [66]–[71]. Consistent with this concept is the observation that compared to an age-matched healthy control group, individuals with Alzheimer's disease have reduced levels of the activated (i.e. phosphorylated) form of CREB [72]; calcium signaling to the cell nucleus is the key inducer of CREB phosphorylation on its activator site serine 133 [12]. Similarly, in aged neurons, calcium signaling may be altered at the level of calcium signal generation and/or calcium signal propagation [73]–[75]. This could explain the reduced levels of serine 133-phosphorylated CREB in the hippocampus of aged, learning-impaired rats [76]–[78], which could lead to compromised endogenous neuroprotection, progressive cell loss and cognitive decline. The development of strategies to boost the endogenous neuroprotective machinery may lead to effective therapies of neurodegenerative condition. Both in disease and aging, health and functionality of neurons may be preserved by expressing AID genes or by restoring or enhancing key signals, in particular nuclear calcium.
Hippocampal neurons from newborn C57/Black mice were cultured in Neurobasal media (Invitrogen, Carlsbad, CA, USA) containing 1% rat serum, B27 (Invitrogen, Carlsbad, CA, USA), and penicillin and streptomycin (Sigma). The procedure used to isolate and culture hippocampal neurons has been described [79],[80]. The hippocampal cultures used for this study typically contained about 10 to 15% glial cells and therefore a fraction of the RNA isolated from the cultures was derived from glial cells. Stimulations were done after a culturing period of 9 to 12 days during which hippocampal neurons develop a rich network of processes, express functional NMDA-type and AMPA/Kainate-type glutamate receptors, and form synaptic contacts [12],[81]. Action potential bursting was induced by treatment with the GABAA receptor antagonist bicuculline (Sigma) (50 µM) as described previously [2],[11],[12]. In the survival experiments, neurons were treated for 16 hours with bicuculline in the presence of 250 µM 4-amino pyridine (4-AP; Calbiochem) [2]. 4-AP increases the frequency of the bicuculline-induced action potentials bursts, thereby enhancing nuclear calcium, CREB-mediated transcription, and activity-induced neuroprotection [2],[11],[12].
DNA microarray analysis was done using Affymetrix GeneChip Mouse Genome 430 2.0 Arrays. See Text S1 for details.
The vectors used to construct and package rAAVs have been described previously [6]. The rAAV cassette for mRNA expression contains a CMV/chicken β actin hybrid promoter. The following rAAVs were generated and confirmed by DNA sequencing: rAAV-Atf3, rAAV-CaMKIVK75E, rAAV-GADD45β, rAAV-GADD45γ, rAAV-Ifi202b, rAAV-Inhba, rAAV-LacZ, rAAV-Npas4, rAAV-Nr4a1, rAAV-Ptgs2, and rAAV-Serpinb2. rAAV-Btg2 and rAAV-CaMBP4 have been described [6]. All rAAV-expressed proteins except hrGFP carry a Flag tag. For shRNA expression, a rAAV vector was used that contains the U6 promoter for shRNA expression and a CMV/chicken β actin hybrid promoter driving hrGFP expression [6]. For details on the construction of rAAVs expressing shRNAs see Text S1. Hippocampal neurons were infected with rAAVs at 4 days in vitro (DIV). Infection efficiencies were routinely determined immunocytochemically at 9 DIV or 10 DIV using antibodies to the Flag tag or to hrGFP or by analyzing the fluorescence of hrGFP; they ranged from 80 to 95 percent of the viable neurons [6].
To determine the mRNA expression levels of Atf3, GADD45β, GADD45γ, Ifi202b, Inhba, Npas4, Nr4a1, Ptgs2, Serpinb2, Gusb, and glyceraldehyde-3-phosphate dehydrogenase (Gapdh), QRT-PCR was performed using real-time TaqMan technology with a sequence detection system model 7300 Real Time PCR System (Applied Biosystems, Foster City, California, USA). For further details, see Text S1.
As described previously [2],[3], two types of assays were used to investigate apoptotic cell death and the protection from cell death afforded by a period of action potential bursting. At 10 DIV, activity-dependent survival was induced by treatment of the neurons for 16 hours with bicuculline (50 µM) and 4-AP (250 µM). All electrical activity of the network was subsequently stopped using tetrodotoxin (TTX; TOCRIS Bioscience) (1 µM) followed by keeping the cells either in regular medium (containing growth and trophic factors) with or without staurosporine (Calbiochem) (10 nM), or in medium lacking growth and trophic factors, all in the presence of TTX (1 µM). The principal growth and trophic factors in the regular, serum-free hippocampal medium [termed transfection medium (TM) [76]] are insulin, transferrin, and selenium. Staurosporine-induced and growth factor withdrawal-induced apoptosis was assessed after 36 hours and 72 hours, respectively, by determining the percentage of hippocampal neurons with shrunken cell body and large round chromatin clumps characteristic of apoptotic death [2],[3]. In the growth factor withdrawal assays, basal cell death is slightly higher due to the differences in the time that the neurons are kept in serum free, TTX-containing media. At least 20 visual fields from each coverslip (corresponding to 1500–2000 cells per coverslip) were counted with Hoechst 33258 (Serva) and the percentage of dead cells was determined. TUNEL assays (Roche, Mannheim, Germany) were done according the instructions provided by the manufacturer and were used to validate the analysis of cell death using the Hoechst 33258 stain. Photomicrographs of examples of healthy and apoptotic hippocampal neurons stained with TUNEL and with Hoechst 33258 are shown in Figure S1. All cell death analyses were done without knowledge of the treatment history of the cultures. All results are given as means±SEM; statistical significance was determined by ANOVA.
Imaging of mitochondrial membrane potential was done using Rhodamine 123 (Rh123; Molecular Probes, Eugene, OR) as described [2],[51],[52]. Imaging and data analysis were performed without knowing the experimental conditions. Quantitative measurements are given as means±SEM from n≥4 experiments, with at least 100 cells analyzed each. Statistical significance was determined by ANOVA.
rAAVs were delivered by stereotaxic injection into the dorsal hippocampus of male Sprague-Dawley rats weighing 230–250 g. Rats were randomly grouped and anaesthetized with ketamine. A total volume of 3 µl containing 3×108 genomic virus particles were injected unilaterally over a period of 30 min at the following coordinates relative to Bregma: anteroposterior, −3.8 mm; mediolateral, 2.8 mm; dorsoventral, −2.8 to −3.8 mm from the skull surface. Procedures were done in accordance with German guidelines for the care and use of laboratory animals and to the respective European Community Council Directive 86/609/EEC.
Two weeks after rAAV delivery, rats were injected with kainate (Sigma; 10 mg/kg i.p.) or vehicle (phosphate-buffered saline, PBS), and monitored for at least 4 hours to categorize the severity of epileptic seizures according to following criteria: level 1, immobility; level 2, forelimb and/or tail extension, rigid posture; level 3, repetitive movements, head bobbing; level 4, rearing and falling; level 5, continuous rearing and falling; level 6, severe tonic-clonic seizure [82]. Only animals that exhibited at least level 4 or 5 of epileptic seizure behavior were analyzed further. Three days after seizure induction, animals were deeply anesthetized with an overdose of Nembutal (300 mg/kg), pre-perfused transcardially with PBS, and perfused with 200 ml of neutral phosphate buffered 10% formalin (Sigma). Brains were removed and post-fixed overnight in the same fixative solution. For cryoprotection, brains were incubated in 30% sucrose in PBS for 2 days. Brains were rapidly frozen on dry ice. Frozen sections (40 µm thick) were collected in PBS. Three consecutive sections separated by a 240 µm distance were used for immunostaining and Fluoro-Jade C staining (Histo-Chem, Inc., Jefferson, Arkansas, USA), which selectively stains degenerating neuronal cell bodies and processes, regardless of the mechanism of cell death. Transgene expression was detected with anti-Flag antibodies (1∶2500, M2 mouse monoclonal; Sigma). Neuronal cell loss was assessed with NeuN immunostaining (1∶500, mouse monoclonal; Chemicon). Immunostaining was done using standard procedures. Fluoro-Jade C staining was done as described previously [54]. Images of Fluoro-Jade C staining were taken and the signals in the CA1 area of the hippocampus were quantified. The quantification was performed without knowing the experimental conditions. Images (10× objective, 1600*1200 pixels) of Fluoro-Jade C stained sections were taken at central CA1 region infected with rAAVs. Sections were collected every 240 µm by cryostat sectioning at the level of 3.0∼5.0 mm posterior to Bregma. Five sections from each brain hemisphere were chosen for image analysis. Fluoro-Jade C signals from the images were quantified with NIH ImageJ software (National Institute of Health, Bethesda, MD, USA). Background intensity was measured from the CA1 area lacking positively-stained neuronal cell bodies. Threshold level was set as means of the background +3 SD. The CA1 pyramidal cell layer was encircled manually and particle analysis was performed. Particles were defined as 30 pixels to infinity, roundness 0∼1.0. Total pixel of Fluoro-Jade C positive particles from each section was obtained to calculate the mean cell death area for each hemisphere. All pixel values were normalized to the average kainate-induced Fluoro-Jade C signal from the control (i.e. non-injected) hemispheres of kainic acid treated animals; the signal obtained from the non-injected hemispheres of the animals injected with rAAV-Btg2, rAAV-Ifi202b, and rAAV-Serpinb2 was not included into the calculation of the average kainate-induced Fluoro-Jade C signal because for rAAV-Btg2, rAAV-Ifi202b, and rAAV-Serpinb2 we observed neuroprotection on the contralateral, non-injected hemispheres.
All animal experiments were done in accordance with the international ethical guidelines for the care and use of laboratory animals and were approved by the local animal care committee of the Regierungspräsidium Karlsruhe. We minimized the number of animals used and their suffering.
|
10.1371/journal.ppat.1005799 | Compartmentalization of Total and Virus-Specific Tissue-Resident Memory CD8+ T Cells in Human Lymphoid Organs | Disruption of T cell memory during severe immune suppression results in reactivation of chronic viral infections, such as Epstein Barr virus (EBV) and Cytomegalovirus (CMV). How different subsets of memory T cells contribute to the protective immunity against these viruses remains poorly defined. In this study we examined the compartmentalization of virus-specific, tissue resident memory CD8+ T cells in human lymphoid organs. This revealed two distinct populations of memory CD8+ T cells, that were CD69+CD103+ and CD69+CD103—, and were retained within the spleen and tonsils in the absence of recent T cell stimulation. These two types of memory cells were distinct not only in their phenotype and transcriptional profile, but also in their anatomical localization within tonsils and spleen. The EBV-specific, but not CMV-specific, CD8+ memory T cells preferentially accumulated in the tonsils and acquired a phenotype that ensured their retention at the epithelial sites where EBV replicates. In vitro studies revealed that the cytokine IL-15 can potentiate the retention of circulating effector memory CD8+ T cells by down-regulating the expression of sphingosine-1-phosphate receptor, required for T cell exit from tissues, and its transcriptional activator, Kruppel-like factor 2 (KLF2). Within the tonsils the expression of IL-15 was detected in regions where CD8+ T cells localized, further supporting a role for this cytokine in T cell retention. Together this study provides evidence for the compartmentalization of distinct types of resident memory T cells that could contribute to the long-term protection against persisting viral infections.
| Some viruses have the capacity to establish chronic infections in humans. How different T cell populations effectively control these infections has not been clear. Continuous circulation of memory T cells was thought to be crucial for effective immune surveillance against such infections. Recent studies in mice however, have shown that non-circulating tissue resident memory populations can also contribute to protective immunity. In this study we have examined the distribution, localization and characteristics of Epstein-Barr virus and Cytomegalovirus-specific T cells in different human tissues. This showed that virus-specific T cells were differentially distributed in different tissues and there was preferential accumulation of EBV-specific resident memory T cells at sites where EBV reactivates. In vitro studies showed that IL-15 and TGF-β could cooperate to extinguish tissue exit signals in T cells and therefore potentiate their retention within tissues. IL-15 expression was also detected in areas where T cells aggregated within the tissue. Together our study provides insight into how distinct memory T cells are compartmentalized in tissues to maintain long-term protection against persisting viral infections.
| It has recently become evident that protective T cell immunity relies not only on circulating memory T cells but also on non-circulating resident memory populations [1–5]. These resident memory T (Trm) cells have been identified in a variety of different non-lymphoid and lymphoid tissues in mice [6–10]. Importantly, when compared to their circulating counterparts Trm cells provide superior protection against reinfection at their site of localisation [6,11–16]. The characteristics of these cells in humans however, are poorly understood. A greater understanding of the mechanisms that regulate their development and maintenance is paramount for future vaccine strategies.
Residence within tissue environments depends upon the ability of T cells to overcome the egress signals. This is achieved by acquiring expression of receptors that enhance cellular interaction within the tissue and facilitate survival for prolonged periods within that tissue. The exit signal for T cells is largely mediated by the concentration gradient of sphingosine-1-phosphate and expression of its receptor, S1P1, on T cells [17]. Accordingly, studies in mice have shown that Trm cells completely lack the expression of S1P1 as well as its transcriptional regulator KLF2 [18,19]. T cell exit through the efferent lymphatic system is facilitated by the expression of CCR7. KLF2 is also known to positively regulate the transcription of CCR7 [20], and therefore the loss of KLF2 may also abrogate the CCR7 mediated T cell exit. Trm cells are further distinguished from their circulating counterparts by constitutively expressing CD69. This C-type lectin has traditionally been considered as a marker of T cell activation, but its role in promoting tissue residence, through binding to and down-modulating pre-existing S1P1 on the T cell surface, has only recently been recognized [21–23]. A second surface marker associated with tissue residence is CD103, the alpha chain of the integrin αEβ7 which mediates T cell binding to E-cadherin expressed on epithelial tissues [24]. In the mouse, at least two distinct subsets of Trm cells have been identified based on the presence or absence of CD103, with the CD103+ subset largely found at barrier surfaces [6,13,25–27].
To date, studies in man suggest that Trm cells are likely to be present in both lymphoid and non-lymphoid organs [28–32]. A recent study has shown that human skin is populated with at least two distinct memory T cell subsets that are non-circulating resident populations. This was demonstrated in patients who underwent alemtuzumab treatment, which selectively depleted the circulating T cell populations [29]. Most other studies however, have relied solely on the expression of CD69 as a Trm marker and, while large numbers of CD69+CD8+ T cells have been reported in human lymphoid organs [28,30], the significance of such findings is difficult to judge. Firstly, it is not clear whether these CD69+CD8+ T cells possess other aspects of the Trm phenotype, such as loss of S1P1 and KLF2, or whether they merely express CD69 as a result of recent activation. Furthermore a crucial feature attributed to Trm cells in mouse models, namely the ability of antigen-specific cells to persist at sites of potential antigen encounter [10], has only been examined in human skin [33] or lungs [34]. Whether antigen-specific Trm cells accumulate in human lymphoid tissues is unknown.
In that context, it is relevant to compare the distribution of CD8+ T cells to two common human herpesviruses, Epstein-Barr virus (EBV) and cytomegalovirus (CMV) in different human tissues. Both elicit numerically strong CD8+ T cell responses, but are harboured at different sites in vivo [35,36]. EBV persists as a latent infection of a memory B cell population that preferentially recirculates between the blood and oropharyngeal lymphoid tissues such as the tonsil [37]. Occasional reactivation from the latent reservoir is thought to seed foci of lytic infection in oropharyngeal epithelium, leading to periods of viral shedding into throat washings [36]. By contrast, CMV persists as a latent infection of the myeloid lineage with the capacity to reactivate to lytic infection at various tissue sites [35]. In this study we have focused on two human lymphoid tissues, namely spleen and tonsils, and have used CD69 and CD103 to identify two distinct subsets of memory T cells that are retained within these two organs in the absence of recent T cell activation. These two populations are transcriptionally distinct by S1P1 and KLF2 expression and have different anatomic distributions, with a selective retention of EBV-specific T cells in tonsillar tissues suggesting that they are strategically positioned at sites of possible antigen encounter. Our data also indicated the important roles for two locally-produced cytokines, IL-15 and TGF-β, in determining tissue residence of CD8+ T cells.
We chose to compare circulating T cells with T cell populations isolated from human spleens and tonsils, two lymphoid tissues that are important for infections acquired through blood and the oropharynx respectively. As previously reported [28], CD69 expression was minimal on circulating T cells; however, was consistently detected on 25–75% of splenic and tonsillar CD8+ T cells (Fig 1A). CD69 expression was largely confined to memory CD8+ T cell populations, with the CCR7—CD45RA—effector memory and CCR7—CD45RA+ TEMRA subsets together accounting for over 75% of CD69+CD8+ T cells in the spleen and over 90% in the tonsils (Fig 1B). Since CD69 is also expressed on recently activated T cells, we examined CD8+ T cell subsets for early (CD137, CD25) or late (HLA-DR) T cell activation markers. When compared to CD69—CD8+ T cells however, there were no differences in the expression levels of these markers on CD69+ CD8+ T cells (Fig 1C and 1D). In addition, recently activated effector T cells can be identified by their high expression of KLRG1. Both CD69+ and CD69—CD8+ T cells in spleen (Fig 1C) and tonsils (Fig 1D) however, were largely KLRG1low. This demonstrated that CD69 was expressed on memory CD8+ T cells in the absence of recent T cell activation. Enhanced T cell survival capacity is also crucial for the persistence of memory T cells [38]. In this regard we determined the expression of the pro-survival gene BCL-2 in purified CD69+ and CD69—effector memory CD8+ T cells. Compared to CD69—memory CD8+ T cells, CD69+ memory CD8+ T cells expressed higher levels of BCL-2 (Fig 1E), suggesting that CD69+ memory CD8+ T cells are better equipped for survival than their CD69—counterparts. Together these data clearly demonstrate that CD69 is expressed on memory CD8+ T cells in these lymphoid tissues in the absence of recent T cell activation.
Since at least two distinct subsets of Trm cells can be identified in mouse tissues and human skin by differential expression of CD103 [6,13,26,27,39], we also examined CD8+ T cells from human spleens and tonsils for co-expression of CD103 and CD69. As shown in Fig 2A and 2B, in both tissues CD103 expression was largely restricted to the CD69+ subset; however, while in the spleen CD103+ cells made up only a small fraction of CD69+ population, in the tonsil these cells comprised around 50% of the CD69+ subset. As in the mouse therefore, human lymphoid tissues contain both CD69+CD103—and CD69+CD103+ subsets of CD8+ T cells. To gain further insight into possible differences between these two subsets, we characterized both populations phenotypically using the T cell differentiation markers CD45RA and CCR7 (the latter known to be involved in T cell exit from peripheral tissues and T cell retention in lymph nodes [22,40]), and CD11a, the alpha chain of the integrin LFA-1 known to be associated with memory T cell retention in murine tissues [41]. As shown in Fig 2C and 2D, while CD69+CD103—CD8+ T cells were divergent in terms of CD45RA and CCR7 expression, the CD69+CD103+ CD8+ T cell subset was uniformly CD45RA—CCR7—in both the spleen and tonsils. The expression levels of CD11a varied between the organs. In the spleen there were no major differences between the three subsets, however in the tonsils both CD69+ subsets expressed higher levels when compared to the CD69—subset. Studies in mice suggested that Trm cells have high levels of PD-1 and therefore we asked whether any of the subsets in humans expressed PD-1. We also examined TIM-3 and BTLA, two markers often associated with cell exhaustion. These showed that CD69+CD103+CD8+ T cells had the highest level of PD-1, followed by CD69+CD103-CD8+ T cells (Fig 2E). Neither of these subsets however, appeared to express both TIM-3 and BTLA (Fig 2E), suggesting that they are unlikely to be exhausted cells.
To extend the analysis of phenotypic differences between CD69+CD103+ and CD69+CD103—memory cell subsets, we next assessed expression of S1P1 and KLF2, both of which are down-regulated in, and are thus a feature of Trm cells [18]. We purified CD103+CD69+, CD103—CD69+ and CD103—CD69—CCR7—memory CD8+ T cells from spleen and tonsils by sorting and then quantified mRNA levels by Q-PCR. This showed that the CD69+CD103+ subset of memory T cells had significantly down-regulated transcription of both markers, whereas the CD69+CD103—subset had down-regulated S1PR1 substantially but KLF2 only partially (Fig 2F). These data suggested that although both CD69+CD103+ and CD69+CD103—populations are likely to be retained within the tissues, the mechanisms that regulate their retention could be different.
To characterize further the differences between CD103—CD69+ and CD103+CD69+ CD8+ T cell populations, we examined the anatomical location of these subsets. Immunofluorescence microscopy showed clear differences in the localization of these two subsets in tonsils and spleen. Fig 3A shows the overview of the stains in tonsils and spleen. Higher magnification of tonsillor sub-epithelial region (area 1) and extra-follicular region (area 2) revealed that CD103+CD69+CD8+ T cells preferentially localized near the epithelial barrier surface while CD103—CD69+CD8+ T cells were largely localized in the extra-follicular regions (Fig 3B). Quantitative analysis of the proportion of CD103+ T cells near the epithelium confirmed that the majority of this subset was indeed localized near the barrier surface (Fig 3C). In the spleen there were only a few CD69+CD103+CD8+ T cells, and these were located within the red pulp (area 1) (Fig 3D). The CD103—CD69+CD8+ T cells however, were localized within the periarteriolar lymphoid sheaths (PALS) (area 2) (Fig 3D). Together this reveals that CD103—CD69+ and CD103+CD69+ CD8+ T cells localize to different anatomical locations within human lymphoid tissues.
Having identified two distinct subsets of CD8+ T cells that were retained within human lymphoid tissues, we wanted to determine the factors that influenced their retention. Studies in mice have shown that the maturation of Trm T cells is largely determined by cytokine exposure [1,19,27]. We therefore asked whether cytokines could potentiate the retention of CD8+ T cells in humans by inducing a Trm-phenotype in circulating CD8+ T cells. To this end, CD8+ T cells from peripheral blood were stimulated for 7 days either with candidate cytokines or with T cell activation and expansion beads (TAE) as a polyclonal TCR stimulus and the expression of CD69 and CD103 examined by flow cytometry. Stimulation with IL-15 alone consistently up-regulated CD69 on resting human CD8+ T cells (Fig 4A and 4B) including a small population of CD103+CD69+ CD8+ T cells (Fig 4A). By contrast, type I interferons IFN-α and IFN-β, which in mice have been shown to induce CD69 expression on T cells, and IL-2 had little effect in these experiments (Fig 4B). Although TGF-β failed to induce CD69 expression by itself, together with IL-15, TGF-β induced a small proportion of CD103+CD69+ CD8+ T cells (Fig 4A and 4B). Stimulation through the TCR and co-stimulatory receptors using TAE beads resulted in a large proportion of CD69+ and small proportion of CD103+ populations (Fig 4A). Importantly, in contrast to this polyclonal TCR stimulation which up-regulated activation markers such as CD137, IL-15 induced CD69 expression in the absence of CD137 expression (Fig 4C). In addition, CD8+ T cells isolated from the spleen not only proliferated in response to IL-15, but also maintained the expression of CD69 (S1A Fig). In order to determine which subsets of circulating CD8+ T cells up-regulated CD69 in response to IL-15 simulation, we purified peripheral blood naïve (CCR7+CD45RA+), central memory (TCM, CCR7+CD45RA—), effector memory (TEM, CCR7—CD45RA—), and TEMRA (CCR7—CD45RA+) CD8+ cells from blood by cell sorting, labeled them with cell trace violet (CTV) to track cell proliferation and stimulated them for 7 days with IL-15. This revealed that although all 4 subsets responded to IL-15 and underwent robust proliferation, expression of CD69 was mainly induced on TEM or TEMRA CD8+ T cells (Fig 4D). These data suggest that effector memory CD8+ T cells may be more responsive to IL-15-induced CD69 expression, which is consistent with the greatest proportion of CD69+ cells within the TEM subset in human tonsils and spleens (Fig 1B).
We next determined whether IL-15, TGF-β and IL-2 could induce the transcriptional down-regulation of S1PR1 and KLF2 in human CD8+ T cells. We stimulated peripheral blood CD8+ T cells with IL-15 or IL-2 in the presence or absence of TGF-β for 7 days, purified CD69+ and CD69—CD8+ T cells by cell sorting, isolated their RNA and determined expression levels of KLF2 and S1PR1 by Q-PCR. As shown in Fig 4E and 4F, IL-15 markedly down-regulated S1PR1 transcription in the cells that had converted to CD69+ status, whereas its effects on KLF2 expression were partial and amplified considerably when TGF-β was also present. Similarly, IL-2 also down-regulated both KLF2 and S1PR1, however, the presence of TGF-β did not result in further down-regulation (S1B Fig). In order to determine whether IL-15 induced CD69+CD8+ T cells were unresponsive to S1P gradient we performed trans-well migration assay. This showed that the migration of CD69+CD8+ T cells towards S1P was significantly lower when compared to their CD69—CD8+ T cell counterparts (Fig 4G), despite being able to migrate efficiently to CCL5 (Fig 4H).
In view of the above findings, we then examined whether IL-15 expressing cells were present within human lymphoid tissues. Frozen sections of tonsils were stained for IL-15, CD8+ T cells and B cells. As recently observed in IL-15 reporter mice [42], IL-15 producing cells were abundant in T cell areas of human tonsils, but largely absent from B cell follicles (Fig 5). In addition, IL-15 was detected in the squamous epithelial cells lining the tonsils. These data provide evidence for the constitutive expression of IL-15 within human tonsils.
In order to understand the significance of CD69+CD103+ and CD69+CD103—CD8+ T cell subsets that persist in human lymphoid tissues, we examined the tissue distribution and phenotype of CD8+ memory T cells specific for two common viral pathogens, the B-cell tropic EBV and largely myeloid cell-tropic CMV. The typical values, including that of some paired blood-tissue samples, of the proportion of EBV and CMV-specific CD8+ T cells in the blood, spleen and tonsils suggested that EBV-specific CD8+ T cells preferentially accumulated in the tonsils, while CMV-specific CD8+ T cells remained in the blood (Fig 6A). This prompted us to examine the phenotype of the virus-specific cells in these tissues to determine the proportion of Trm cells. While both EBV-specific and CMV-specific CD8+ T cells in circulation were CD69—CD103—(Fig 6B), a significant proportion of EBV-specific and a small proportion of CMV-specific CD8+ T cells in the spleen expressed CD69 (Fig 6B and 6C). Strikingly, neither EBV nor CMV-specific CD8+ T cells in the spleen expressed CD103 (Fig 6B and 6C). By contrast, large proportions of EBV-specific CD8+ T cells in the tonsils were CD69+CD103+ and the small fraction of CMV-specific cells found in the tonsils remained CD69—CD103—(Fig 6C). The expression levels of late activation marker, HLA-DR was similar between CD69+ and CD69—EBV-specific cells in the tonsils (Fig 6D), suggesting that the CD69 expression was not due to recent T cell activation. These observations revealed that there was selective accumulation of CD69+CD103+ EBV-specific CD8+ T cells in the tonsils.
Maintenance of memory CD8+ T cells at appropriate anatomical sites appears to be crucial for optimal protection against recurrent virus infections. In this study we demonstrate that two distinct subsets of memory CD8+ T cells, expressing markers of tissue residence but not of recent activation, are retained within human lymphoid tissues such as tonsils and spleen. These two subsets are not only phenotypically distinct, but also anatomically separate within the tissue environment. CD8+ T cell memory to different viruses are differently distributed between the two subsets and show different patterns of tissue retention, with memory to an oropharyngeally-replicating virus acquiring a phenotype that is specifically retained at sites of possible virus reactivation within the tonsil. We also show that IL-15 and TGF-β can potentiate the retention of memory T cells in tissues.
Our initial experiments showed that CD69 is expressed on large proportions of CD8+ T cells in human tonsils and spleen in the absence of recent T cell stimulation. Previous studies have also reported similar high proportions of CD69+CD8+ T cells in human lymphoid tissues but left open the question whether CD69 was simply a marker of recent activation rather than tissue residence. The present study, showing that such CD69+ cells lack a range of T cell activation markers, makes it clear that their expression of CD69 reflects tissue retention or tissue-residency. Our data also show that this population is itself heterogeneous, with CD103 as a marker that differentiates at least two distinct subsets. CD103 is an adhesion molecule that has already been associated with Trm cells in the skin, brain, gut and reproductive tract in mice [6,7,9,13,19,25,43]. Although CD103 expression is not a requirement for Trm cell formation [27,43], its presence on the membrane enables T cells to bind to epithelial surfaces where its ligand E-cadherin is expressed [24]. We have identified crucial differences between the CD103+CD69+ and CD103-CD69+ CD8+ T cell populations in human lymphoid tissues. Both subsets exhibited dramatic down-regulation in expression of S1P1. However, while there was a corresponding strong down-regulation of KLF2 in the CD103+ subset, this was only partial in the CD103- subset. In line with this, the CD103+ subset was uniformly CCR7—, consistent with KLF2 also regulating expression of CCR7 [20]. In addition, the CD69+CD103+ subset preferentially localized to epithelial barrier. Therefore our data from human tissues not only demonstrates the differences between CD69+CD103+ and CD69+CD103—subsets, but also indicates that the CD69+CD103+ subset is more typical of the Trm cell populations described at barrier surfaces in mouse models.
Emerging evidence from different mouse models suggest that Trm cell formation is a two-step process [1]. The first step is the infiltration of a memory precursor population from blood into the tissue; a process that may or may not depend on the presence of antigen [7,9,12,13,27,44]. In the second step, the infiltrated precursor cells mature to become Trm cells through a process that largely depends on cytokines [19,27]. During this maturation process the CD8+ T cells not only express CD69 and CD103, but also substantially down regulate KLF2 and S1P1 [12,19]. To this end, IL-15 and TGF-β have been implicated in the development of Trm cells in mouse models, although the mechanism by which IL-15 enables Trm cell formation is unclear [19]. Here we provide evidence for a key role for these two cytokines in humans. Firstly, we demonstrate that IL-15 alone can up-regulate CD69, probably by down-modulating S1P1 and KLF2. Our data however, also reveals that down-regulation of KLF2 was only partial in the presence of IL-15 alone and TGF-β was required for the complete down-regulation of KLF2. IL-15 is a known growth factor for memory T cells and it enables T cells to persist in the absence of continuous antigen stimulation [45,46]. Therefore IL-15 is a likely candidate to retain and maintain a pool of memory T cells. Although it has been known that epithelial cells produce TGF-β, here we also demonstrate the presence of IL-15 producing cells at the sites where T cells are retained. We therefore propose that IL-15 and TGF-β could be the key regulators of CD69+CD103+ and CD69+CD103- subsets in human lymphoid tissues.
Further insight into the differences between the two subsets came when the specificity of CD8+ T cells in human spleen and tonsils was examined. Although EBV persists within the memory B cell compartment, it largely remains as a true latent infection in these cells, and the evidence for its reactivation outside of oropharynx is limited. For example, the viral load in the spleen of healthy virus carriers was 20-fold lower than that observed in their peripheral blood [47], suggesting that it is unlikely that regular viral replication takes place at sites such as the spleen. By contrast, the lymphoid tissues associated with the Waldeyer’s ring in the oropharynx are likely sites where virus reactivation takes place, probably initiated when virus-infected B cells infiltrate these tissues and the infection switches from latency into lytic cycle to seed foci of replication in permissive oropharyngeal epithelium [37,48,49]. This explains the occasional bouts of clinically silent virus shedding detectable in the throat washings of healthy virus carriers [36]. Such shedding is clearly under some form of T cell control, since levels of shedding are significantly raised in T cell-compromised individuals [36,50]. In this context the phenotype of EBV-specific CD8+ T cells in the tonsils suggests that more than a third of them were CD69+CD103+ and therefore are likely to be positioned near the epithelial barrier where EBV reactivates. Furthermore, the CD103+CD69+ EBV-specific CD8+ T cells were absent in the spleen and the few CMV-specific CD8+ T cells that were retained in the tonsils and spleen were largely CD103—.
Our data reveal that CD103+CD69+ EBV-specific CD8+ T cells are selectively retained at sites of possible antigen encounter, which is consistent with the Trm T cell function implicated in mouse models [10]. Accumulation of virus-specific CD8+ T cells at sites of viral replication has been observed in other infections as well [11,34]. Here we show the distinct compartmentalization of different virus-specific resident memory CD8+ T cells that could be a crucial strategy for sustained protective immunity to pathogens. Our study also provides evidence for locally produced cytokines to potentiate the formation and positioning of Trm T cells. Taking together, the evidence we provide here suggest that the development of EBV-specific resident memory CD8+ T cells within the tonsils is likely to be influenced by the virus and the local environment. During active EBV replication, virus-specific effector CD8+ T cells are recruited to the tonsils. These infiltrating EBV-specific CD8+ T cells do not express CD103 [51]. Overtime, under the influence of IL-15 and TGF-β these effector cells are likely to mature to become CD69+CD103+ resident memory T cells, positioned along the epithelial barrier. This is supported by the finding that CD103+ EBV-specific CD8+ T cells within the tonsils only appear during convalescence [51]. The positioning of CD69+CD103+CD8+ T cells is likely to be facilitated by presence of E-cadherin at these sites [52]. More importantly, when compared to their circulating counterparts these CD103+ EBV-specific CD8+ T cells were highly reactive against EBV antigens [52], suggesting that the primary role of these Trm cells could be to prevent or limit the viral replication at these sites. This strategic positioning of EBV-specific Trm cells is reminiscent of herpes simplex virus (HSV)-specific Trm cells in human skin. A recent study has demonstrated that similar persistence of HSV-specific CD8+ Trm cells at sites of viral reactivation is crucial for the containment of the virus [33]. Therefore the fact that EBV reactivation is often asymptomatic or sub-clinical could be largely due to this effective control at the sites where the virus reactivates.
Buffy coats from healthy blood donors along with spleens and paired blood samples from cadaveric organ donors were obtained from the Australian Red Cross Blood Services. Tonsils were obtained from patients undergoing routine tonsillectomy. Spleen and tonsil specimens were disaggregated to single-cell suspensions and the mononuclear cells were isolated using the standard Ficoll separation method. Cells were either used fresh or cryopreserved in liquid nitrogen for later work. All human experiments were approved by jurisdictional ethics committees in Sydney as well as the institutional review boards.
Approval for this study was obtained from human ethics committees of the Royal Price Alfred Hospital, St Vincent’s hospital and Sydney South West Area Health Services (Australia). Informed written consent was obtained from next of kin.
Mononuclear cells from blood, spleen and tonsils were stained with fluorochrome-conjugated antibodies (mAbs) specific for cell surface proteins. The following mAbs were used for identification of CD8+ T cells and the determination of their phenotype; anti-CD3, anti-CD8 (Biolegend), anti-CCR7 (R&D Systems), anti-CD45RA (BD Biosciences), anti-CD69, anti-CD25, anti-CD137, anti-HLA-DR, anti-CD103, anti-KLRG-1 and anti-CD11a (all obtained from Biolegend). Fluorochrome-conjugated HLA class I dextramers (Immudex) were used to identify virus-specific CD8+ T cells. EBV-specific CD8+ T cells were identified with dextramers specific for the following viral epitopes; GLCTLVAML (derived from EBV-lytic protein BMLF1), CLGGLLTMV (derived from EBV-latent protein LMP2), RAKFKQLL (derived from EBV-lytic protein BZLF1) and FLRGRAYGL (derived from EBV-latent protein EBNA3A). CMV-specific CD8+ T cells were identified with dextramers specific for the following viral epitopes; NLVPMVATV, TPRVTGGGAM, RPHERNGFTV (all derived from pp65 protein), VLEETSVML, ELRRKMMYM, ELKRKMMYM (all derived from IE-1 protein) and VTEHDTLLY (derived from pp50 protein). Stained cells were analyzed on either FACSCanto II or LSRFortessa flow cytometer (BD Biosciences) and the data processed using FlowJo software (Treestar, Ashland, USA).
RNA was isolated immediately after ex vivo purification of T cells or from cells after 7 days of culture using RNeasy kit (Qiagen). Total RNA was then reverse transcribed with oligo-dT. For BCL2, the following Real-time PCR primer set was used; forward, 5’–ttgacagaggatcatgctgtactt– 3’ and reverse, 5’–atctttatttcatgaggcacgtt- 3’. Q-PCR was performed with assorted commercially available Taqman assays (Hs00824723_m1, Hs00984230_m1, Hs02800695_m1, Hs00173499_m1, Hs00360439_g1) and Taqman Fast Advanced Mastermix on a StepOnePlus Real-Time PCR cycler (Life Technologies). The threshold cycle of S1PR1 and KLF2 for each cell population was normalized to the arithmetic mean of HPRT, B2M and UBC housekeeping genes (ΔCt). Normalized gene expression of each cell type was compared to the gene expression of a reference population with expression set to 1 according to the 2(-ΔΔCT) method.
Frozen sections of the tonsils and spleens were acetone-fixed and stained for different markers with mAbs using standard protocols [25]. The following primary mAbs were used to identify lymphocytes, anti-CD8 (Abcam), anti-CD3 (Biolegend/ AbD Serotec) (to reveal CD8+ T cells) and anti-IgM (Life technologies) (to reveal B cells). The expression of CD69 and CD103 was identified using purified, flurophore-conjugated or biotin-conjugated anti-CD69 mAb (BioLegend) and anti-CD103 mAbs (BD Biosciences). The presence of IL-15 producing cells was determined using anti-IL-15 mAb (Abcam). The following secondary antibodies were used to reveal specific staining; affinity purified F(ab’)2 fragments of AF647 or AF488 or CyTM3 conjugated donkey anti-mouse IgG and CyTM3 conjugated donkey anti-rabbit IgG (all obtained from Jackson ImmunoResearch). Control antibodies or secondary mAbs in the absence of primary mAbs were used to determine background fluorescent levels. Images were acquired using Delta Vision Personal (Olympus) or Zeiss LSM700 microscope and analysed Imaris software (Bitplane).
CD8+ T cells were isolated from PBMCs using magnetic separation kit (Dynal). Isolated cells or total PBMCs were cultured for 7 days in the presence of cytokines (50 ng/ml of IL-15, IFN-α, IFN-β, TGF-β and 50 U/ml of IL-2) or with T cell activation and expansion beads (TAE; anti-CD3/CD28/CD2 mAb micro beards, Miltenyi Biotech; Polyclonal stimulation). Different doses of IL-15 (1, 10 or 50 ng/ml) were used for stimulation of virus-specific CD8+ T cells. For experiments where T cell proliferation was measured, 1–2 x 106 purified CD8+ T cells were labeled with CellTrace Violet (CTV; Invitrogen) prior to cell culture. The proliferative history was determined based on the dilution of CTV of the T cells after stimulation.
Sorted CD69+ and CD69—CD8+ T cells were washed in RPMI with 0.05% fatty-acid-free BSA (Sigma) and tested for transmigration across gelatin coated 5 μm transwell filters (Corning) for 4 hours to Shingosine-1-phosphate (S1P) (Sigma) or CCL5 (R&D Systems). Migrated cell numbers were enumerated by flow cytometry.
|
10.1371/journal.pntd.0002176 | Mode of Death on Chagas Heart Disease: Comparison with Other Etiologies. A Subanalysis of the REMADHE Prospective Trial | Sudden death has been considered the main cause of death in patients with Chagas heart disease. Nevertheless, this information comes from a period before the introduction of drugs that changed the natural history of heart failure. We sought to study the mode of death of patients with heart failure caused by Chagas heart disease, comparing with non-Chagas cardiomyopathy.
We examined the REMADHE trial and grouped patients according to etiology (Chagas vs non-Chagas) and mode of death. The primary end-point was all-cause, heart failure and sudden death mortality; 342 patients were analyzed and 185 (54.1%) died. Death occurred in 56.4% Chagas patients and 53.7% non-Chagas patients. The cumulative incidence of all-cause mortality and heart failure mortality was significantly higher in Chagas patients compared to non-Chagas. There was no difference in the cumulative incidence of sudden death mortality between the two groups. In the Cox regression model, Chagas etiology (HR 2.76; CI 1.34–5.69; p = 0.006), LVEDD (left ventricular end diastolic diameter) (HR 1.07; CI 1.04–1.10; p<0.001), creatinine clearance (HR 0.98; CI 0.97–0.99; p = 0.006) and use of amiodarone (HR 3.05; CI 1.47–6.34; p = 0.003) were independently associated with heart failure mortality. LVEDD (HR 1.04; CI 1.01–1.07; p = 0.005) and use of beta-blocker (HR 0.52; CI 0.34–0.94; p = 0.014) were independently associated with sudden death mortality.
In severe Chagas heart disease, progressive heart failure is the most important mode of death. These data challenge the current understanding of Chagas heart disease and may have implications in the selection of treatment choices, considering the mode of death.
ClinicalTrails.gov NCT00505050 (REMADHE)
| Chagas disease remains a burden for public health systems in Latin American countries. Several authors believe that sudden death is the main cause of death in this population. So many efforts have been made to prevent sudden death in Chagas disease. In order to verify if sudden death is the leading cause of death in Chagasic heart failure, we performed a subanalysis of the REMADHE prospective trial, which included a population of outpatients in a tertiary referral center for heart failure. We grouped patients according to etiology (Chagas vs non-Chagas) and modes of death that were classified as progressive heart failure death, sudden death, other cardiovascular death, noncardiovascular death or unknown death. Our study showed that in this end of the spectrum of presentation of Chagas disease, with systolic heart dysfunction, progressive heart failure is the main mode of death. These data have implications for the development of new strategies for prevention of chagasic heart failure.
| Chagas disease (American trypanosomiasis) remains a burden for public health systems in Latin American countries [1]. In fact, the global prevalence of Chagas disease has reached 9 million people and 25 million are estimated to be at risk worldwide [2]. As a result of globalization and migration, non-endemic countries have reported an increase in the prevalence of Chagas disease. It has been estimated that 47,000 T. cruzi-infected persons are now living in Spain and another 300,000 in the United States [3].
Sudden death has been considered the main cause of death in patients with Chagas heart disease [4]. Nevertheless, this information comes mainly from a period before the introduction of drugs that changed the natural history of heart failure such as beta-blockers, angiotensin-converting enzyme inhibitors (ACEI) and angiotensin II receptor blockers (ARB). Furthermore, many studies that support this finding include heterogeneous cohorts of patients, in particular patients without ventricular dysfunction [4].
To the best of our knowledge there is not a recent publication studying the mode of death in Chagas heart disease. The management of Chagas heart disease is complex and includes the treatment of complex arrhythmias and heart failure. The precise understanding of current natural history of patients presenting systolic dysfunction is indispensable to plan adequate health politics and improve survival. Therefore, we sought to study the mode of death of patients with heart failure caused by Chagas heart disease, comparing with non-Chagas cardiomyopathy.
We examined the patients included in the REMADHE (Repetitive Education at Six-Month Intervals and Monitoring for Adherence in Heart Failure Outpatients) trial; For the purpose of current study, patients were grouped according to etiology (Chagas vs non-Chagas groups) and mode of death.
REMADHE is a prospective, randomized, single-center open parallel trial controlled by nonintervention simple randomization and designed to compare a disease management program versus control in patients with chronic heart failure. Patients enrolled in the study were under ambulatory care in a tertiary referral center and were followed by a cardiologist with experience in heart failure. Inclusion elapsed from October 1999 until January 2005. Patients were aged 18 years or older with irreversible chronic heart failure of at least 6-months. Exclusion criteria included patients' inability to attend educational sessions and researchers' inability to monitor patients because of the patients' lack of transportation, social or communication barriers; myocardial infarction or unstable angina within 6 months before randomization; cardiac surgery or angioplasty within 6 months before randomization; hospitalized patients or recently discharged patients; any severe systemic disease that could impair expected survival; procedures that could influence follow-up; pregnancy or childbearing potential (figure 1). The REMADHE trial demonstrated that a disease management program was associated with reduction in unplanned hospitalization, total hospital days and need for emergency care, as well as improved quality of life [5], [6].
Patients were divided into two groups: Chagas and non-Chagas. The diagnosis of Chagas disease was based on epidemiological information along with serological tests (indirect immunofluorescence, passive haemaglutination, immunoenzymatic assay) positive for Trypanosoma cruzi [7]. Patients with an alternative diagnosis, or a mixed etiology for the cardiomyopathy were included in non-Chagas group.
All deaths were classified according to specified definitions. The agreement of two members of the study on the cause of death was mandatory. Sudden death was defined as an unexpected death occurred within 1 hour of the onset of new symptoms or occurred unwitnessed in a previously stable patient [8].
Heart failure death was defined as a death that occurred because of worsening or intractable heart failure, which included cardiogenic shock, pulmonary edema and terminal arrhythmias during hospital stay for aggravated heart failure.
Other cardiovascular death was defined as a death that occurred after a cerebrovascular accident, vascular disease, myocardial infarction or cardiovascular procedure. Cerebrovascular accident was defined as a persistent disturbance of neurological function. Diagnosis required characteristic history, physical examination, imaging techniques and/or autopsy data.
Myocardial infarction (MI) death was defined as a death that occurred after a verified acute MI. Noncardasiovcular death was defined as a death resulting from a noncardiac reason. Unknown death was defined as a death that did not have a definitive cause.
Patients were categorized according to the New York Heart Association functional class [9] during evaluation of the medical staff of our heart failure clinic.
The primary end-point of the study was all-cause, heart failure and sudden death mortality that were obtained during follow-up, either from the trial database, from review of medical records or by telephone contact with family members. Additionally, Cox proportional-hazards regression models were used to explore the relationship of mode of death, etiology and other variables with survival time. For Cox proportional hazards model, each mode of death is looked at in turn and all other modes of death are censored.
Results are presented as frequencies, mean (SD), or median (interquartile range), as appropriate. For effects of group comparison, the t test was used for normal distribution and Mann-Whitney test was used to compare variables without normal distribution. For categorical variables, chi-square test or the Fisher exact test was applied. Survival was estimated by the Kaplan–Meier method, and differences in survival between groups were assessed by the log-rank test. Cox proportional-hazards models were used to compare the rates of deaths for each mode of mortality. In the analysis, data on patients was censored at the time of implantation of a defibrillator or the time of heart transplantation. All analyses and graphs were performed with SPSS statistical software version 13.0 and Graphpad Prism software version 5.0.,
The study protocol was approved by the institutional ethics committee of Hospital das Clinicas da Faculdade de Medicina da Universidade de Sao Paulo and all patients provided written informed consent. All data analyzed were anonymized.
A total of 412 patients were enrolled in the REMADHE trial. For the purpose of the current study, we excluded patients with left ventricular ejection fraction ≥50% (60 patients) and those who had ICD (10 patients) and we censored nine patients at the time of implantation of a defibrillator. Thus, 342 patients were analyzed. The first inclusion occurred in October 1999, and patients were followed until February 2010 with a mean follow-up of 1,284 days±895 days. Despite our cohort included 342 patients, we have information about New York Heart Association functional class (NYHA FC) only in 296 patients. Baseline characteristics of the patients are described in table 1.
As compared to non-Chagas, Chagas patients had lower body mass index, smaller end-diastolic left ventricle diameter, and smaller proportion of patients under beta-blocker therapy. Chagas patients also had higher proportion of females and larger left ventricular ejection fraction (Table 1).
Chagas patients had a higher incidence of death related to cerebrovascular accident and non-cardiovascular deaths. They also had a higher number of hospitalizations and long-term hospital stay (Table 2).
Altogether 185 (54.1%) patients died. Death occurred in 31 (56.4%) Chagas patients and 154 (53.7%) non-Chagas patients. Kaplan-Meier cumulative event curves for all-cause, heart failure and sudden death mortality are shown in Figures 2, 3 and 4.
The cumulative incidence of all-cause mortality was higher in patients with Chagas heart disease compared to non-Chagas patients (figure 2). Additionally, the cumulative incidence of heart failure mortality was higher in Chagas patients compared to non-Chagas (figure 3). Nevertheless, when we analyzed the cumulative incidence of sudden death mortality, there was no difference between the two groups (figure 4).
The influence of etiology on heart failure and sudden death mortality was further evaluated by Cox proportional-hazards regression. As we have information about New York Heart Association functional class (NYHA FC) only in 296 patients, we did two different models of Cox proportional-hazards regression; Model 1 without information about NYHA FC and Model 2 including NYHA FC (I/II vs III/IV). Other variables included were age, gender, etiology (Chagas vs non-Chagas), use of beta-blocker, use of angiotensin-converting enzyme inhibitor or angiotensin II receptor blocker, use of amiodarone, use of spironolactone, body mass index, left ventricular ejection fraction, left ventricular end diastolic diameter (LVEDD), group allocated (disease management program versus control), hemoglobin, serum sodium, creatinine clearance, number of hospitalizations, days of hospitalization and number of admissions to emergency care.
Chagas etiology, LVEDD, creatinine clearance and use of amiodarone were independently associated with heart failure mortality on Model 1 and Model 2 (table 3). On the other hand, LVEDD and use of beta-blocker were independently associated with sudden death mortality on both Models 1 and 2 (table 4).
Concerning the toxic effects of amiodarone, it is important to highlight that none patient died from lung toxicity.
The relationship between mode of death and NYHA functional class at randomization was analyzed for each etiology. As mentioned before, information about NYHA functional class was available in 296 (87%) patients.
In patients with Chagas heart disease, the proportion of patients who died of worsening heart failure increased substantially as increased functional class (38% vs 57%). On the other hand, the proportion of patients who died of sudden death decreased with the worsening of functional class (31% vs 15%). Furthermore, 23% of deaths in NYHA class I/II were secondary to other cardiovascular causes (Figure 5).
In non-Chagas group, sudden death was the main cause of death throughout the spectrum of NYHA functional class; 47% in both class I/II and class III/IV (figure 5).
To the best of our knowledge, this prospective study is the first one to investigate the causes of death in outpatients with Chagas heart disease associated with severe systolic dysfunction in comparison with non-Chagas patients. A recent study investigated risk estimation for cardiac events in Chagas disease but not analyzed the mode of death [10]. The present study found that patients with heart failure caused by Chagas heart disease had a worse prognosis when compared to others as well that the predominant mode of death in these chagasic patients were progressive heart failure throughout the spectrum of NYHA functional class presentation. In addition, the incidence of death related to cerebrovascular accident and non-cardiovascular deaths was higher in Chagas disease. In the same scenario, the main mode of death in non-Chagas group was sudden death. Moreover, in multivariate analysis Chagas heart disease, left ventricular end diastolic diameter, amiodarone use, serum sodium and creatinine clearance were independent risk factors for death from progressive heart failure.
Our finding is divergent from previous anecdotal reports concerning mode of death in chagasic heart failure patients that indicates sudden death as the main cause of death [4]. Prior studies had included few patients with severe heart failure or patients with higher values of left ventricular ejection fraction. Furthermore data were limited to abstracts or the treatments were outdated [4], [11]. Otherwise, other publications in Chagas disease had been selected patients with complex ventricular arrhythmias or patients under secondary prevention for sudden death [12], [13]. In the largest study in Chagas heart disease, 62.3% of the patient died of sudden death [14]. However the patients received outdated treatment characterized by low rate of beta-blockers (0.2%–2.4%) and of angiotensin-converting enzyme inhibitor (1.9%–21.9%), and unexplained high rates of amiodarone (35.1%–71.9%), that is not in accordance with guidelines for treatment of chagasic heart failure [15], [16]. As additional limitations, it was a retrospective study, the left ventricular function was not reported, only 10.4% of the patients were in NYHA functional class III/IV and the mode of death analysis was partially based on precarious Brazilian death certificate. Our results concerning progressive heart failure death are in concordance with other previous publications that included chagasic patients also under outdated treatment [17], [18]. Our finding of high incidence of stroke causing death in Chagas disease in comparison with other etiologies confirms previous studies [18], [19] despite the unexpected recently reported lack of evidence of pro-thrombotic status among patients with Chagas disease [20]. However, cardiac embolus originating from apical aneurysm and left ventricular thrombosis are a recurrent cause of Chagas stroke [21]. The explanation for high incidence of non-cardiovascular death on Chagas heart disease is unknown. However, in general chagasic patients have lower social and nutritional status and higher incidence of comorbidities, which could predispose them to non-cardiovascular death [22].
Some distinguishing clinical and anatomopathological aspects should be noted in Chagas disease that might influence the mode of death. Biventricular impairment with persistent myocarditis, inflammatory infiltrate edema, contraction-band necrosis and myocytolysis, focal and diffuse areas of myocellular hypertrophy, and fibrosis are found in histology of cardiac tissue [23]–[26]. Also, chronic low-grade parasite persistence drives tissue damage and the autoimmune component of Chagas cardiomyopathy [27]. Even though this autoimmunity may play a role in development of Chagas heart disease, this was never demonstrated in clinical setting. The role of anti-muscarinic antibodies in the determination of mode of death should be investigated [28]. Concerning the effects of chagasic myocarditis in the mode of death, our findings are in agreement with a recent study, which included 181 patients suspected of viral myocarditis followed for 58 months and did not show prevalence of sudden death as mode of death [29].
One would expect the influence of fibrosis in Chagas heart disease determining sudden death as the main mode of death as in ischemic cardiomyopathy. Although pathological findings in Chagas disease reported a pattern of diffuse interstitial fibrosis [30], magnetic resonance imaging study reported a pattern of fibrosis that can vary a lot from diffuse, heterogeneous, small focal to big transmural scars indistinguishable from myocardial infarction [31]. Ischemic cardiomyopathy displays regional fibrosis and scars which seem to have a mechanistic link with malignant ventricular arrhythmias [32]. The remarkable chagasic biventricular impairment, especially right-sided failure has been appointed as an independent predictor of survival in Chagas heart disease [18]. In fact, the complex Chagas heart disease is characterized by multiple clinical and anatomopathological factors that determine the mode of death, the clinical manifestation and the worst prognosis observed in our study as well as in other publications [33], [34].
The medical treatment might have influenced our results. The 38% rate of beta-blockers prescription to our chagasic patients cannot be considered substantial but it was in accordance of guidelines at the period of inclusion of REMADHE , when beta-blockers use was controversial in chagasic patients [35]. Only recently the use of beta-blockers was included in guidelines for the treatment of chagasic patients [15]. However, this 38% rate is considerably higher in comparison with previous studies where the rate of beta-blockers were under 15% [13], [14]. In addition, it could partially explain our results given that beta-blockers reduce sudden death in heart failure [36].
Our group has previously published that beta-blockers attenuate the worst prognosis of Chagas heart disease, approaching similar survival to other etiologies [33]. In this study, use of beta-blockers was not independently associated with the risk of death from worsening heart failure but it was independently associated with sudden death mortality. Based on our results, it is reasonable to suppose that beta-blockers had a more prominent effect in preventing sudden death than worsening heart failure death in patients with Chagas cardiomyopathy.
Specifically regarding the influence of amiodarone, which has been widely prescribed in Chagas heart disease to prevent sudden death and has reached up to 70% rate of use [13], [37], [38], our cohort had a lower rate of its use (9.1% in total cohort and 12.7% in Chagas group). Despite this, the percentage of sudden death was strikingly lower in this etiology. According to I Latin American Guidelines for the diagnosis and treatment of Chagas' heart disease, amiodarone is indicated only to (Class I) sustained ventricular tachycardia (symptomatic or not) or symptomatic non-sustained ventricular tachycardia and (Class IIb) for asymptomatic ventricular premature beats or non-sustained ventricular tachycardia [15].
On the other hand, amiodarone was an independent predictor of worsening heart failure death in our cohort, which agrees with previous study in non-chagasic population [39]. This finding could be related to a more strict indication of amiodarone in severe cases. However, this is an issue to be clarified in future studies because amiodarone prescription in our heart failure clinic is indicated according Brazilian Heart Failure Guidelines [16], in clinical situations not necessarily related to worst prognosis in heart failure.
In the face of our findings, an awkward question arises related to the higher incidence of sudden death in populations of chagasic patients despite the high rate of unexplained prescription of amiodarone [13], [40]. One issue to be clarified is the potential influence on mortality of amiodarone side effects such as pro-arrhythmic effects, bradycardia and AV block in a disease known to have high incidence of AV block, sick sinus syndrome, besides the increase deaths from circulatory failure and non-cardiovascular death [39], [41].
In non-Chagas group, our proportion of sudden death and progressive heart failure death is consistent with literature data. It is worth noting that our prescription of beta-blockers, angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker and spironolactone reflects the state-of-the-art treatment of heart failure. Our results corroborate the EMPHASIS trial, which studied mild non-chagasic heart failure and demonstrated sudden death as main mode of death among cardiovascular death [42]. Also, the CIBIS trial reported sudden death as the predominant mode of death in the bisoprolol arm in non chagasic severe heart failure (NYHA class III/IV) [43].
This study contains some limitations that should be acknowledged. First, this study was a single center trial that included patients initially enrolled in another intervention trail. In addition, this was a retrospective analysis though the data were obtained prospectively in REMADHE trial.
Concerning our cohort, Chagas group is considerably smaller than non-Chagas group and Kaplan-Meier curves result may have limited value after 1500 days since the number of Chagas patients at risk is very low. However, it is important to note that the difference in proportion between groups reflects the reality of our population.
In few patients, the mode of death of patients was not clarified. However, it was less than 10%. Unfortunately, the gold standard autopsy studies were not available widespread to confirm certainly the cause of death. Therefore, in most randomized trials as in the REMADHE trial the autopsy is not routinely performed.
Clinicians should keep in mind that in severe Chagas heart disease progressive heart failure plays an important role as the most important mode of death in a scenario of worse prognosis in comparison with other etiologies. The data from this analysis challenges the current understanding of Chagas heart disease and may have implications in the selection of treatment choices considering the mode of death. In addition, our results may influence the development of new strategies for prevention of chagasic heart failure. The amiodarone role in chagasic heart disease should be reevaluated. Moreover, prevention of thromboembolism must be emphatically pursued. Finally, priority should be given to enhance the research related to prevent and treat the progression of heart failure in Chagas disease.
|
10.1371/journal.pgen.0040007 | ER Stress-Mediated Apoptosis in a New Mouse Model of Osteogenesis imperfecta | Osteogenesis imperfecta is an inherited disorder characterized by increased bone fragility, fractures, and osteoporosis, and most cases are caused by mutations affecting the type I collagen genes. Here, we describe a new mouse model for Osteogenesis imperfecta termed Aga2 (abnormal gait 2) that was isolated from the Munich N-ethyl-N-nitrosourea mutagenesis program and exhibited phenotypic variability, including reduced bone mass, multiple fractures, and early lethality. The causal gene was mapped to Chromosome 11 by linkage analysis, and a C-terminal frameshift mutation was identified in the Col1a1 (procollagen type I, alpha 1) gene as the cause of the disorder. Aga2 heterozygous animals had markedly increased bone turnover and a disrupted native collagen network. Further studies showed that abnormal proα1(I) chains accumulated intracellularly in Aga2/+ dermal fibroblasts and were poorly secreted extracellularly. This was associated with the induction of an endoplasmic reticulum stress-specific unfolded protein response involving upregulation of BiP, Hsp47, and Gadd153 with caspases-12 and −3 activation and apoptosis of osteoblasts both in vitro and in vivo. These studies resulted in the identification of a new model for Osteogenesis imperfecta, and identified a role for intracellular modulation of the endoplasmic reticulum stress-associated unfolded protein response machinery toward osteoblast apoptosis during the pathogenesis of disease.
| Osteogenesis imperfecta (OI) is a heterogeneous collection of connective tissue disorders typically caused by mutations in the COL1A1/2 genes that encode the chains of type I collagen, the principle structural protein of bone. Phenotypic expression in OI depends on the nature of the mutation, causing a clinical heterogeneity ranging from a mild risk of fractures to perinatal lethality. Here, we describe a new OI mouse model with a dominant mutation in the terminal C-propeptide domain of Col1a1 generated using the N-ethyl-N-nitrosourea (ENU) mutagenesis strategy. Heterozygous animals developed severe-to-lethal phenotypes that were associated with endoplasmic reticulum stress, and caspases-12 and −3 activation within calvarial osteoblasts. We provide evidence for endoplasmic reticulum stress–associated apoptosis as a key component in the pathogenesis of disease.
| Mutations in type I collagen genes (COL1A1/2) typically lead to Osteogenesis imperfecta (OI), the most common heritable cause of skeletal fractures and bone deformation in humans [1]. OI is classified into eight human subtypes, and to date greater than 500 human COL1A1 mutations have been reported representing a clinical heterogeneity dictated by the complex array of mutations. Recently, novel molecules and loci apart from classic type I collagens have been implicated in both murine [2] and human [3–5] alternative recessive forms of OI, thus expanding the genetic heterogeneity.
Type I collagen is the most common ubiquitously expressed fibrous protein in the extracellular matrix (ECM) of connective tissues with both biomechanical and physiological functions [6]. Type I collagen initially exists as a procollagen precursor with NH2- and COOH-terminal propeptide domains with distinct roles. Type I procollagen molecules consist of three polypeptide coiled subunit chains (two proα1(I) and one proα2(I) chain) that self-associate in the endoplasmic reticulum (ER), and require a highly coordinated post-translational regulation. The helical procollagens are deposited into the extracellular space, proteolytically cleaved, and then organized into highly ordered collagen fibrils covalently cross-linked to increase tensile strength and rigidity. Apart from its biomechanical properties, type I collagen stores key factors for remodeling maintenance, and acts as an adhesive substrate with cellular receptors and other matricellular components along its major ligand binding regions [7]. These properties regulate complex intracellular signal transduction pathways for tissue remodeling and repair, immune response, polarization, migration, proliferation, differentiation, and cell survival within various cellular contexts [8].
Based on detailed radiographic, molecular genetic and morphological analyses, structural collagen mutations are likely associated with lethal (type II) and moderate (types III and IV) forms of OI [1]. Type II OI represents the most severe form of the disease accounting for ∼20 % of cases, and the heterogeneous clinical and biochemical aspects have been described [9]. Most OI-II probands acquire new dominant mutations in COL1A1/2, and a low frequency recurrence risk can be the result of gonadal mosaicism in one of the parents [10]. OI-II-related mutations behave in a dominant negative manner as the mutant proα(I) proteins affect wild-type chain registration and formation due to the polymeric nature of assembly, impairing overall ECM stoichiometry and integrity. Typically in OI-II cells, abnormal trimers are poorly passaged through the secretory pathway, and are more vulnerable to degradation and overmodification, thus the cytoprotective unfolded protein response (UPR) is concordantly instigated [11–13].
Despite the observation of machinery involving BiP chaperones that might target most defective procollagens for subsequent proteasomal degradation in OI-II cells [13], the direct activation of apoptosis has not been shown to date. Here, we provide the first report on the isolation, cloning and characterization of the skeletal phenotype in a novel ENU-induced mouse line for human OI affecting the terminal C-propeptide region of Col1a1. Furthermore, we provide evidence of caspases-12 and −3-mediated apoptosis in Aga2 osteoblasts as a critical step toward increased cell death, thus broadening our molecular insight into OI.
The original Aga2/+ mouse was identified in the Munich ENU dominant mutagenesis screen [14], and displayed an abnormal gait due to deformity of the hind limbs at five weeks of age among the F1 mutagenized progeny. Aga2/+ males have general reproductive success and the mutation segregated in a dominant manner with complete penetrance (Table 1). Aga2/+ females produced smaller litter sizes due to a reduction in body size. A large subset of Aga2/+ animals succumbed to postnatal lethality (Table 1), and featured severe bone deformities and fractures. Aga2/+ inter se mating yielded non-Mendelian ratios for dominant inheritance suggesting embryonic lethality. Various embryonic stages were investigated to determine gestational arrest in homozygotes, which was estimated to occur around embryonic day 9.5 post coitum (data not shown).
Genetic mapping of the Aga2 locus was performed following the standard outcross-backcross breeding strategy. The mutation was mapped to a 700 kb domain on Chromosome 11 (Figure S1). Within the Aga2 candidate region, two skeletal genes were identified, Col1a1 (procollagen type I, alpha1) and Chad (chondroadherin). Both candidate genes were sequenced and a novel T to A transversion mutation within intron 50 of Col1a1 was identified (Figure 1A). The Aga2 substitution generated a novel 3′ splice acceptor site and predicted a terminal frameshift beyond the endogenous stop (Figure 1C). On the transcript level, cDNA from homozygous embryos revealed a 16 bp expanded transcript, whereas in heterozygous embryos equal levels of two transcripts are present (Figure 1B). We further confirmed the mutation at the genomic level. The T to A exchange disrupted the endogenous MspA1I restriction site, generating a cleavage-resistant allele in Aga2/+ samples (Figure 1B).
The C-terminal portion of human, and murine as well as Aga2 mutant Col1a1 was aligned highlighting specific residues of importance (Figure 1C). Of interest, the most terminal conserved cysteine C244 (aa 1451) was ablated, and 48 endogenous amino acids including the stop were frameshifted, concomitantly predicting 90 new amino acids beyond the termination position. The Aga2 frameshift begins 56 amino acids away from the chain selectivity sequence [15]. Additionally, five new cysteines and a potential N-linked oligosaccharide (CHO) attachment site were introduced. The secondary structure of the Aga2 product was predicted to remove the hydrophobicity of the short loop formed by the last intra-chain disulphide bond (data not shown).
Among all Aga2/+ animals, the gross skeletal phenotype was discernable starting between days 6 - 11 after birth. Heterozygous animals that survived to adulthood were classified as moderately-to-severely affected and displayed the hallmark dystrophic limb(s), long bone and pelvis fractures, reduced body size, and generalized decreases in DXA-based bone parameters (Figure 2 and Table S1; data were collected from the German Mouse Clinic dysmorphology primary screen [16]). Lethal animals developed thin calvaria, hemorrhaging at joint cavities and intracranial sites, scoliosis, provisional rib and long bone calluses and deformities, body size deficit, pectus excavatum, gasping, cyanosis, platyspondyly, edema of the eyes, greasy skin, and eczema (Figure 2 and data not shown). Comminuted fractures incurred signifying the severe brittleness of Aga2/+ bones accompanied by many repair blastemas characteristic of a type II OI phenotype.
The effect of the mutation on volumetric bone mineral density (vBMD) and content (vBMC) was evaluated in adult mice using in vivo peripheral quantitative computed tomography (pQCT; Table S2). In Aga2/+ distal femora, trabecular and cortical vBMD as well as cortical vBMC were substantially decreased compared to controls. In contrast, the trabecular vBMC was unchanged due to enlarged medullar areas (suggesting elevated resorption in Aga2/+). Collectively, these results suggest potential defects in mineralization and/or bone formation in Aga2/+ mice.
Several serum biochemical and hormonal markers were evaluated to detect possible bone metabolic disturbances (Table S3). Total alkaline phosphatase (ALP) and osteocalcin levels were significantly increased in both sexes in Aga2/+ compared to littermate controls. In addition, circulating TRACP 5b (an osteoclast marker) was significantly elevated in Aga2/+. No changes in inorganic phosphate levels were observed (data not shown). Aga2/+ animals depicted a significant increase in PTH levels. Furthermore, Aga2/+ mice yielded a distinct increase in calcitonin, which was more pronounced in females. Taken together, the significantly increased bone formation and resorption markers indicate that in Aga2/+ mice metabolic bone turnover is elevated. This was further supported by histomorphometric analysis of bone formation and resorption showing increased numbers of endogenous osteoblasts and osteoclasts, an increased bone formation rate (BFR) and a reduced mineral apposition rate (MAR) in Aga2/+ bone samples (Figures S2, S3 and Table S4).
Fibroblast and type I collagen features in Aga2/+ connective tissues were evaluated by transmission and scanning electron microscopy (TEM and SEM). Aga2/+ dermis contained more heterogeneous populations of fibroblasts with smaller nuclei, aberrant dilated electron-dense ERs, lysosomes and empty autophagic-like vacuoles interspersed throughout the cells (Figure 3A and 3B). Aga2/+ ERs appeared contiguous with secretory vesicles signifying the formation of ER associated compartments. SEM studies on cortical nanostructure of Aga2/+ samples depicted a less-parallel, less-densely packed network of collagen bundles when compared to controls (Figure 3C–3F).
Procollagen trafficking was assessed in immunofluorescence studies using dermal fibroblasts (Figure 4). To uncover mature type I collagen integrity we used an antibody that was directed against the triple helical domain and was also capable of detecting procollagens (Figure 4A). Wild-type cells depicted both intracellular and intact extracellular surface staining (Figure 4B). In Aga2/+ cells, the intact type I collagen surface stain was largely reduced while intracellularly retaining procollagen molecules were stained (Figure 4C). Double immunofluorescence staining of proα1(I) chains was performed using antiserum LF-67 (Figure 4A), protein disulphide isomerase (PDI), and Golgi phosphoprotein 4 (GOLPH4) for detection of the ER and cis-Golgi apparatus, respectively. Merged stainings showed proα1(I) chain retention within the ER (Figure 4E) but not in the cis-Golgi (Figure 4G), when compared to controls (Figure 4D and 4F). It is speculated that aberrant procollagen molecules aggregate into vesicular structures (Figure 4E, merged green dots), which may be part of the proteasome. To further clarify the intracellular abnormalities double immunofluorescence staining of proα1(I) chains was performed using antiserum LF-41 and β-actin antibody for the cytoskeleton. The LF-41 epitope resides within the wild-type terminal C-propeptide overlapping segment of the Aga2 frameshifted region and captures only molecules that contain chains with the normal C-propeptide (Figure 4A). Using antiserum LF-41, control fibroblasts depict dynamic intracellular procollagen and isoform trafficking to the infoldings of the plasma membrane (Figure 4H). Contrarily, Aga2/+ fibroblasts retained intracellular (wild-type) procollagen molecules at near peri-nuclear regions (Figure 4I, arrow). Surface staining in both wild type and Aga2/+ cells was limited given that the LF-41 antiserum was unable to discriminate mature, proteolysed type I collagen in the ECM. Based on our results, the mutation affected proα1(I) chain processing by blocking its ER-to-Golgi anterograde transport, thus inhibiting vesicular exocytosis to the matrix.
OB metabolism and function were characterized utilizing the primary calvarial culture system. We determined the growth curve of Aga2/+ primary OBs and observed a limited saturation density within the stimulated samples (Figure 5A) that was accompanied by a relative increase in ALP levels until day 26 (Figure S4A). Also, Aga2/+ total protein and mitochondrial reductase activity levels were significantly increased (Figure S4B). Consistent with our immunofluorescence studies there was a clear reduction in the total amount of secreted acid-soluble collagens tested in Aga2/+ media (Figure S4B). Of note, unstimulated Aga2/+ OBs depicted no significant change in cellular protein and metabolic activity levels (data not shown).
To address the functional defects in Aga2/+ OBs, nodular formation, growth and binding capacities were studied over time (Figure S4C). The formation of bone-like nodules was persistently reduced between days 9 and 16 in Aga2/+. In addition, the average area of individual Aga2/+ nodules was reduced compared to controls. Lastly, a 3.2 fold comparative increase in nodular dye-binding capacity was observed in Aga2/+ (Figure S4C). Taken together, these results demonstrate clear functional and metabolic disturbances in Aga2/+ OBs.
To show involvement of key regulators of ER stress response pathways, we performed qPCR studies using primary calvarial OBs (Figure 5B). At days 16 and 26 in culture (11 and 21 days after induction, respectively), expression of the molecular chaperones genes Hspa5 (also known as BiP/GRP78) and Serpinh1 (also known as Hsp47) was upregulated ∼1.5 – 2.2 folds above control. The expression of Ddit3 (also known as Gadd153/chop), a transcription factor involved in the induction of apoptosis, was ∼3.7 folds higher in Aga2/+ cells at day 26 compared to control levels. To confirm the induction of cytoprotective unfolded protein response (UPR) in Aga2/+ cells, Western analysis of BiP protein expression was performed using primary calvarial OBs. As shown in Figure 5C BiP protein levels were increased compared to controls.
Within the context of ER-induced stress we investigated cell apoptosis. The Aga2 mutation led to a combined relative increase in the number of early-late stage TUNEL-positive picnotic OBs compared to control (Figure 6E). At 16 days in culture, Aga2/+ preparations depicted a 10 % relative increase in the number of OBs, which contained aberrant caspase-12-immunoreactivity (Figure 6A and 6B; 3.3 ± 0.9 % +/+ and 12.9 ± 1.9 % Aga2/+; p = 0.0007 Student t-test). For specificity, proteolytic processing of procaspase-12 was evident only in Aga2/+ samples via Western analysis (Figure 6G). Caspase-3/7 activation was independently confirmed depicting a significant 9 % comparative increase in activity in Aga2/+ OBs (Figure 6F). Lastly, activated caspase-3 immunoreactivity was observed within femoral periosteum (Figure 6C and 6D). Aga2/+ tissue contained elevated relative numbers of activated caspase-3-positive OBs (28.9 ± 2.7 % +/+ and 42.6 ± 4.1 % Aga2/+; p = 0.02 Student t-test), suggesting a 14 % comparative basal increase in apoptosis. Also, TUNEL experiments revealed increased numbers of DNA-fragmented OBs within Aga2/+ periosteum (data not shown).
ENU mutagenesis is a powerful approach to identify mouse models for human skeletal disorders by generating novel phenotypes. Human mutations affecting the α1/2(I) C-propeptide coding and non-coding regions have been identified less often than other types of mutations and reflect strong variability in clinical outcomes. Here, we provide the first report on the cloning and phenotypic description of the novel mouse line Aga2, which bear a chemically induced genetically stable mutation affecting the terminal C-propeptide domain of Col1a1. We provide evidence that abnormal proα1(I) chains accumulated intracellularly and were poorly secreted extracellularly. Furthermore, we show for the first time that this was associated with the induction of an ER stress-specific unfolded protein response (UPR) with caspases 12 and 3 activation and apoptosis of osteoblasts both in vitro and in vivo. The Aga2 line provides a unique tool for understanding the underlying mechanisms of the incurable debilitating human disease Osteogenesis imperfecta (OI) and the efficacy of novel therapeutic strategies.
Aga2/+ mice feature phenotypic variability ranging from postnatal lethality to moderate-to-severe changes including increased bone fractures, fragility, deformity, osteoporosis, and disorganized trabecular and collagen structures. Compared to the lethal transgenic lines [17,18], the Aga2/+ phenotype is stronger and associated with increased lethality, but comparable to the BrtlIV knock-in (Col1a1G349C) line [19]. The biochemical serum and histomorphometric results of Aga2/+ mice indicated an elevated bone turnover that is described similarly for OI patients [20–23]. Our ex vivo studies showed that Aga2/+ OBs deposited less collagen matrix, and were overly active, affecting nodule growth and function.
The procollagen triple helix proceeds from the carboxyl to amino end, and this association is modulated by correct folding via intra-chain disulphide bonds in the C-propeptide region before chain association and inter-chain disulphide linkage [24,25]. In certain C-propeptide OI-II lethal conditions, the mutated proα1/2(I) chains can associate with normal chains to form secreted triple helical procollagen molecules despite alterations in endogenous disulfide bonds [12,26], while in other lethal cases chain formation with the mutated propeptide is entirely or largely precluded [27,28]. Based on these results, the lethal OI phenotypes are heterogeneously derived from structural and/or cellular metabolic defects.
At the biochemical level, we report significant increased retention of aberrant procollagen molecules within Aga2/+ cells, compromising cellular metabolism and the ECM. Moreover, we show that the majority of accumulated chains were wild type in nature. Gel analysis of cDNA showed significant splicing of the new Aga2 splice site. Although cryptic mRNA transcripts were readily available for translation, unstable mRNA decay and/or anomalies in transport from the nucleus to the cytoplasm due to secondary structure may have mediated low-level translation of mutant chains. The presumptive elongated chain with newly introduced cysteins and disruption of a crucial beta sheet due to the Aga2 mutation likely destabilized protein conformation, thus affecting preferential chain assembly and proteolysis. Also, a N-linked oligosaccharide unit was introduced capable of altering intracellular processing and secretion of procollagens as well [29].
Our findings were consistent with observations made by Willing et al. [30] who investigated a human OI family that harbored a mutation in the 3' end of COL1A1, which ablated the intra-chain disulfide cysteine as in the Aga2 condition. Fitzgerald et al. [13] reported that the fate of mutated unassembled C-propeptide chains from the proband described by Willing et al. were intracellularly degraded in the proteasome. Similar to Aga2, it was unclear if the mutated chains ever incorporated into procollagen molecules causing protein suicide and the expected downstream effects [31], or involved an alternative mechanism to influence the clinical outcome.
Many disorders result from the cell's inability to export mutated proteins and enzymes from the ER, including OI [32,33]. Depending on the nature of the mutation, incorrectly folded proα1/2(I) molecules are managed via multipartite highly regulated sorting pathways in order to retain ER homeostasis and to prevent the secretion of abnormal proteins. The cytoprotective unfolded protein response (UPR) involves the activation of complex regulatory ER stress signaling mechanisms that can either repress protein synthesis or upregulate ER-resident chaperons and other translation regulators [34]. BiP/Grp78, a central regulator of ER function, was shown to bind to proα (I) chains from cell strains, which harbored unique C-propeptide mutations that inhibited chain association [11] and mediated intracellular ER-degradation [12]. Contrarily, mutations that inhibit folding in the triple helical domain of type I procollagen chains do not result in abnormal BiP binding or induction even though the aberrant molecules are retained within the ER [35].
In Aga2/+, the presence of dilated ERs and accumulation of aberrant procollagens in the ER strongly support intracellular trafficking defects and degradation within the proximal region of the secretory pathway. We speculate that aberrant procollagen molecules aggregated into vesicular structures (Figure 4E), which are part of the proteasome. Degradation did occur within more distal secretory compartments by the presence of lysosome-like structures [36]. In Aga2/+ OB cultures treated with ascorbate, BiP levels were slightly upregulated implicating its constitutive involvement during ER stress. Previous studies from Chessler et al. suggested that in skin fibroblasts from OI patients with mutations in the C-terminal propeptide BiP peaked about 48 hours after exposure to ascorbate and that longer exposure resulted in decreasing levels [11]. As cells studied here were exposed to ascorbate continuously during growth as part of their medium enrichment, we performed Western analysis and confirmed increased BiP protein levels. Hsp47, another molecular chaperone, which is part of the ‘quality control' system for procollagens, is enhanced in certain OI cases [37]. In our studies, the degree of Hsp47 induction was also elevated. Thus, in Aga2/+ both BiP and Hsp47 seemed to choreograph the intracellular regulation of mutant type I procollagens with concomitant induction of apoptosis.
Many diseases are associated with either inhibition or increase of apoptosis [34,38], but to date the active involvement of apoptosis during OI pathogenesis is inconclusive. Caspase-12 is a specific mediator of the ER stress-induced UPR within skeletal tissue [39]. In primary Aga2/+ OBs, removal of the procaspase-12 adaptor protein-binding domain was evident, and the processing of procaspase-12 demonstrated the ER stress origin of mutated proα1(I) initiator signals. In addition, the presence of increased activated caspase-3 and TUNEL-positive OBs in Aga2/+ primary culture and periosteum also implicated a tendency toward apoptosis commitment and cell death, respectively.
Although several models have been proposed for the direct activation of caspase-12 the entire process has yet to be described. The Bcl2 family-proteins are well-established components of the apoptotic machinery, some of which are associated with ER stress pathways. For example, Bim (Bcl2-interacting mediator of cell death) activates caspase-12 via translocating from the dynein-rich ER compartment upon stress [40]. BiP is known to inhibit caspase-mediated cell death by forming complexes with procaspases 7 and 12, and BiP disassociation facilitates procaspase activation [41]. Gadd153/CHOP is a key transcription factor in the regulation of cell growth, differentiation and ER stress-induced apoptosis. Gadd153 induces death by promoting ER client protein load and oxidation via transcription of target genes, which mediate apoptosis, presumably leading to caspase cascades [34]. In our studies, Gadd153 was highly induced over time, and was presumably triggered by mutated procollagens and the modulation of ER stress signaling. Of interest, the anti-apoptotic component Bcl2 is down regulated during ER stress conditions attributed to Gadd153 upregulation, leading to enhanced oxidation and apoptosis [42]. Thus, given the upregulation of Gadd153 within Aga2/+ OBs, Bcl2 was likely down regulated, exacerbating and contributing to cell death. Further studies are necessary to elucidate the pro- and anti-apoptotic components that modulated the severe clinical outcome in Aga2/+ animals. In addition to severe long bone deformities we observed immunological, blood, heart, lung and energy metabolic defects pointing to a systemic effect of the Col1a1 mutation. Further experiments will be necessary to elucidate the influence of the systemic defect on bone and early lethality (unpublished data).
The molecular diversity of apoptosis in OI is unknown, and it is unclear if the apoptotic program is a phenomenon of type I procollagen mutations that severely affect chain assembly and retention caused by triple helical versus C-propeptide structural defects, or by both. Recently, Forlino et al. [43] evaluated phenotype heterogeneity within the BrtlIV knock-in OI mouse line and identified increased Gadd153 transcript and protein levels within calvaria of lethal animals. In this study we isolated and characterized a unique mouse model for OI, and provided evidence that the Aga2 C-propeptide α1(I) mutation induced ER-mediated osteoblast apoptosis that affected cellular function and metabolism, which we now suggest to be a key component, among others, that influenced disease severity in Osteogenesis imperfecta.
Activated caspase-3 (R&D Systems, Germany), caspase-12 (BD Pharmingen, Germany), anti-α-tubulin (Sigma, Germany), β-actin, type I collagen, and BiP/GRP78 (all Abcam, UK), and fluorescently labeled secondary antibodies (Molecular Probes, Germany) were commercially purchased. Antisera LF-41 and -67 [44] were generous gifts provided by Dr. L. Fisher (NIH, USA). The serum biochemical analysis has been previously described [45]. The mouse TRACP 5b (IDS Ltd), osteocalcin ELISA (BTI), and mouse/rat intact PTH as well as calcitonin immunoradiometric assay (Immundiagnostik, Germany) kits were commercially purchased. All measurements were performed based on the manufacturers' recommendations.
Mouse husbandry was conducted under a continuously controlled specific-pathogen-free (SPF) hygiene standard according to the Federation of European Laboratory Animal Science Associations (FELASA) protocols. Standard rodent diet and water were provided ad libitum. All animal experiments were conducted under the approval of the responsible animal welfare authority.
The F1 dominant ENU mutagenesis screen was conducted on the inbred strain C3HeB/FeJ and has been previously described [14]. Aga2 was backcrossed to wild-type C3H mice for five generations, and then outcrossed to C57BL/6J (F1) and backcrossed to the outcrossed strain (N2). Phenotype-positive N2 progeny were tail-genotyped with microsatellite markers. Recombination fractions and map distances were calculated using Mapmanager QTX. All primers for sequencing are available upon request. For genotyping, a PCR fragment containing the entire intron 50 of Col1a1 was generated using primers for-5′-ggcaacagtcgcttcaccta-3′ and rev-5′-ggaggtcttggtggttttgt-3′. The product was then cleaved using MspA1I for gel analysis.
Skeletal preparations were generated as previously described [46]. Simple and compound fractures were grossly examined. Compression fractures were evaluated using a dissecting microscope. For μCT scans (Tomoscope 10010m; VAMP, Germany), image reconstruction and visualization were performed using the ImpactCB and ImpactView software (VAMP), respectively. pQCT analysis was performed as previously described [47].
SEM was performed as previously described [48]. For TEM, lower dorsal skin was prepared as previously described [49]. Sample preparation, embedding and double labeling experiments were performed as previously described [50]. For the demonstration of endogenous ALP and TRAP, the procedure of Miao and Scutt [51] was performed with modifications. All bone histomorphometric parameters were derived from the standardized nomenclature [52]. Non-sequential sections were collected every 70 - 100 μm. All images and measurements were captured and determined using an Axioplan2 workstation (AxioVision v3.1; Zeiss, Germany) and the ImageJ program (v1.36; NIH, USA). Automatic color thresholding was applied to stacked image sets. Labels were traced with minimal operator bias. 6 - 20 repeated measurements were made per section using four non-sequential sections, where n (4 - 6) equals the number of animals examined.
Dermal fibroblasts were cultured as previously described [53]. For caspase-12 and BiP detection, cells were lysed in RIPA buffer with protease inhibitors. Supernatants were analyzed using the NuPAGE Novex gel system (Invitrogen, UK).
OBs were prepared from 3-day-old pups with similar appearance as previously described [54]. Cells were plated at 2 × 104 cells/cm2 for all experiments. OB differentiation was initiated via media exchange (i.e. α-MEM w/10 % FCS, 2 mM glutamine, 1 % pen-strep, 50 μg/ml ascorbic acid, 10 mM β-glycerophosphate). A minimum of four calvaria/genotype was pooled representing an independent experiment (n) with four repeated measurements. DNA, total protein, ALP and MTT (kit) measurements were performed in 24-well plates as previously described [55]. Caspase-3/7 activity was monitored using a fluorogenic substrate (Ac-DEVD-AMC; AXXORA, Germany) at excitation 380 nm and emission 460 nm. Mineral content was quantified by establishing an alizarin red standard curve and measuring dye release using 10 % hexadecylpyridinium in 6-well plates at 570 nm. For nodule characterization, samples were analyzed using the ImageJ software.
Cells were cultured on glass coverslips coated with poly-L-lysine. For immunohistochemical analysis, sodium citrate (10 mM, pH 6.0 in microwave) and proteinase K (10 μg/ml, 37 °C) antigen retrieval methods were performed, and the Vectastain ABC kit (Vector Labs, USA) utilized. The TUNEL assay was performed with an ALP-conjugated kit (Roche, Germany). Independent preparations (n = 4) were analyzed with repeated measurements. For cell culture, 300 - 400 cells were counted per image field. For in vivo caspase-3 analysis, a region 0.4 mm from the growth plate was analyzed, and 400 - 480 total cells were evaluated within set regions (i.e. magnification field).
RNA and cDNA were prepared as previously described [56]. qPCR was performed using the TaqMan and QuantiTect ready-to use primers (Qiagen, Germany). Duplicate crossing points per marker were averaged per independent experiment (n = 4). Significance was depicted between time points. Values were normalized to levels of β-actin from the same pool for fold differences.
Level of significance was set at p < 0.5 by ANOVA test on the influence of genotype and sex, and subsequent pairwise mean comparisons performed by Student t- or post hoc (Bonferroni) tests. All statistics were calculated using the Prism 4 (GraphPad, USA) and SigmaStat v. 3.1 (Systat Software, Germany) software.
Col1a1, MGI:88467 (murine) and HGNC:2187 (human); Chad, MGI:1096866; Hspa5 (BiP), MGI:95835; Serpinh1 (Hsp47), MGI:88283; Ddit3 (Gadd153), MGI:109247. Data depicted from Mouse Genome Informatics (MGI; http://www.informatics.jax.org/) and HUGO Gene nomenclature Committee (HGNC; http://www.genenames.org). |
10.1371/journal.ppat.1000582 | A Novel System for the Launch of Alphavirus RNA Synthesis Reveals a Role for the Imd Pathway in Arthropod Antiviral Response | Alphaviruses are RNA viruses transmitted between vertebrate hosts by arthropod vectors, primarily mosquitoes. How arthropods counteract alphaviruses or viruses per se is not very well understood. Drosophila melanogaster is a powerful model system for studying innate immunity against bacterial and fungal infections. In this study we report the use of a novel system to analyze replication of Sindbis virus (type species of the alphavirus genus) RNA following expression of a Sindbis virus replicon RNA from the fly genome. We demonstrate deficits in the immune deficiency (Imd) pathway enhance viral replication while mutations in the Toll pathway fail to affect replication. Similar results were observed with intrathoracic injections of whole virus and confirmed in cultured mosquito cells. These findings show that the Imd pathway mediates an antiviral response to Sindbis virus replication. To our knowledge, this is the first demonstration of an antiviral role for the Imd pathway in insects.
| Alphaviruses are arthropod-borne viruses maintained primarily in an endemic cycle between mosquitoes and rodents or birds. Transmission to humans may result in wide ranging symptoms from subclinical to fatal encephalitis. While infection of vertebrates causes disease, infection of mosquitoes results in a life-long, persistent infection. In order to examine arthropod host pathways involved in controlling alphavirus infections, we have employed a novel system for the controlled launch of Sindbis virus RNA replication from the genome of the fruit fly, Drosophila melanogaster. We present data showing robust replication of a Sindbis virus RNA following its cell-mediated transcription in flies using the UAS-GAL4 misexpression system. Using this system we have genetically demonstrated that the immune deficiency pathway (Imd) suppresses viral RNA replication as a consequence of the activation of the transcription factor Relish. Additionally, we confirmed the activation of the Relish ortholog as a consequence of Sindbis virus infection of mosquito cells. Our work is the first direct demonstration that the Imd pathway plays a role in arthropod antiviral immunity.
| Arboviruses are a large group of RNA viruses that are transmitted between vertebrate hosts by arthropod vectors, primarily mosquitoes. Several arboviruses including members of alphavirus and flavivirus genera are important human pathogens causing severe arthritis, encephalitis, and hemorrhagic fever. Arboviruses are distributed globally, but individual virus species tend to have a focused geographic range. In the recent past, some viruses have expanded globally, and have caused more frequent and larger epidemics. For example, a strain of Chikungunya virus (an alphavirus) endemic to Africa caused an epidemic outbreak in the Indian sub-continent and the Indian Ocean islands leading to more than a million cases of disease and hundreds of death [1],[2]. Similarly West Nile virus (a flavivirus) originally isolated from Uganda has caused about 100,000 cases of neuroinvasive disease and numerous deaths in North and South America [3]. The periodic nature of the infections along with increasing morbidity and mortality in several parts of the world poses a persistent public health risk [4],[5]. Restriction of arbovirus transmission may be accomplished by vector control, vaccination, and/or antiviral treatment. However, currently there are few vaccines and no effective antiviral therapies available, nor are there efficient and safe means of vector control, underscoring the need to understand how arboviruses interact with vertebrate and arthropod hosts.
Alphaviruses form an important group of arboviruses that causes human disease. They are divided into two clinical groups; those that cause serious but primarily non life-threatening illness like rash and arthritis and those that cause fatal encephalitis. The arthritogenic viruses include Sindbis, Chikungunya, and O'nyong-nyong viruses, while the encephalitogenic viruses include Venezuelan, western, and eastern equine encephalitis viruses [2],[4],[6],[7]. Alphaviruses replicate efficiently in both arthropod and vertebrate hosts, however the pattern of infection differs in a host-dependent fashion; in vertebrate cells alphaviruses cause an acute cytolytic infection, whereas in mosquito cells the infection is predominantly persistent and non-cytolytic. This observation strongly suggests that the virus interacts with the host cells in different ways. Most studies of alphavirus pathogenesis and host responses have been performed in mammalian systems and there is a great deal of information available regarding the antiviral response in vertebrates [8],[9]. However, less is known about the antiviral immunity against alphaviruses in arthropods.
Innate immunity plays an important role in limiting microbes in arthropods, through humoral responses (production of effector molecules such as antimicrobial peptides [AMP]), physical barriers, phagocytosis, encapsulation, and melanization [10]. Drosophila melanogaster has been used as an excellent model to study innate immune responses against pathogens that infect insects. Immune responses to various bacterial and fungal pathogens have been well characterized in Drosophila and primarily consist of the Toll and Imd pathways. The Toll pathway is activated by Gram-positive bacteria and fungi. The pathogen associated microbial patterns (PAMPs) such as lysine type-peptidoglycan are recognized by peptidoglycan receptor proteins (PGRPs) and this binding initiates a serine protease cascade. The cleaved form of the cytokine Spätzle activates transmembrane Toll receptor, which directs the phosphorylation and degradation of Cactus, an IκB-like protein that inhibits the NF-κB like transcription factors Dorsal and Dif. Translocation of these transcription factors to the nucleus causes a rapid increase in expression of multiple AMPs including Drosomycin [10]–[15]. The Imd pathway is stimulated by Gram-negative bacteria. When bacterial PAMP's such as monomeric or polymeric diaminopimelic acid peptidoglycan, bind to the transmembrane PGRP-LC receptor [16], a death domain adaptor protein Imd is recruited. Imd binds to dFadd, another death domain protein which in turn interacts with the apical caspase Dredd [17]–[19]. This caspase then cleaves phosphorylated Relish, a NF-κB-type transcription factor [20]. Relish is phosphorylated by the IKK signaling complex, which is itself thought to be activated by TGF-β activated kinase 1 (Tak1) and its adaptor TAK1-associated binding protein2 (Tab2) [21]–[23]. The cleaved N-terminal domain of Relish then translocates to the nucleus and leads to transcriptional activation of several AMPs including Diptericin [11],[20].
In contrast to the abundant information available for fungal and bacterial infections, less is known about how insects respond to viral infections. Recent studies have pointed to the role of RNA interference (RNAi) in generating antiviral immunity in arthropods [24]–[28]. RNAi, is triggered by the recognition of intracellular long double-stranded RNAs (produced during viral genome replication). The endoribonuclease Dicer-2 processes these into small interfering RNA (siRNA). These siRNA duplexes are then separated by R2D2, and incorporated into the RNA-induced silencing complex (RISC) [29]. The guide strand of siRNA targets the RISC complex to complementary single-stranded RNA, which is then cleaved by the RNaseH like enzyme Argonaute 2 (Ago2) [30]. Flies deficient in Dicer-2, R2D2, or Ago2 exhibit increased sensitivity to infection by Flock house virus (FHV) (Nodaviridae), Drosophila C Virus (DCV) (Dicistroviridae), Drosophila X virus (DXV) (Birnaviridae), and Sindbis virus (Togaviridae, alphavirus) [24],[25],[28].
In addition to RNAi, DCV activates the Jak/STAT pathway in Drosophila. Global transcription profiles of flies infected with DCV showed induction of a set of genes distinct from the Toll- and Imd-induced target genes. vir-1 (virus-induced RNA 1) was strongly induced by DCV and its expression was dependent on Hopscotch, the sole Jak kinase of Drosophila. Also, flies deficient in Hopscotch, showed increased viral load and sensitivity to DCV infection [31],[32]. Studies using DXV demonstrated the role of the Toll pathway in antiviral response. Infection with DXV leads to a strong induction of Drosomycin, a marker of the Toll pathway. Also a loss-of-function mutant in Dif (NF-κB component of Toll pathway) and gain-of-function mutant in the Toll receptor were more susceptible to viral challenge and allowed increased viral replication [33]. Even though some of the mechanisms by which Drosophila controls viral infections are known, the molecular mechanism by which the Jak/STAT and Toll pathways are triggered or the effector mechanisms that control viral infections through these pathways are not yet understood.
The innate immune responses characterized in mosquitoes have been largely based on what is known in Drosophila. The mosquito genome has orthologs to the components of the innate immune machinery of Drosophila. Keene et.al. have shown that in Anopheles gambiae, ago2 and ago3 are required for defense against O'nyong-nyong virus [34] while ago2, r2d2 and dcr2 are required for anti-dengue defense in Aedes aegypti [35],[36]. RNAi is also important in defense against SIN; silencing RNAi components in Ae. aegypti resulted in transient increases in SIN replication [37]. In addition to RNAi, the Toll pathway is also implicated in antiviral defense in mosquitoes. SIN infection induced the expression of Toll pathway-related rel1 transcription factor (ortholog of dif) and genes involved in the vesicular transport in mid-guts of Ae. aegypti [38]. A recent study showed that Toll pathway regulates resistance to dengue virus. Microarray analysis of dengue infected Ae. aegypti resulted in up-regulation of Toll pathway associated genes. Activation of the Toll pathway through RNAi-mediated silencing of the negative regulator Cactus reduced dengue virus infection level while repression of the Toll pathway through gene silencing resulted in higher dengue virus infection levels [39].
Although studies have begun to address the antiviral response in insects, much more needs to be known in order limit the spread of alphaviruses and other arboviral infections. In the present study we have taken advantage of the genetic tools available in Drosophila to study what host factors effect SIN replication. We generated a transgenic fly line that expresses SIN replicon RNA capable of autonomous replication. Previously, transgenic animals expressing viral genomes have been generated and used to study antiviral responses [24],[40]. In the system we generated primary transcription of the replicon is under the control of the UAS/GAL4 system and hence can be launched in a temporally and spatially specific manner that is dependent on the enhancer/promoter driving GAL4 transcription [41]. We have demonstrated that SIN RNA replication can be launched using this system, providing a powerful tool for the genetic analysis of host genes affecting virus RNA replication. The SIN replicon fly line was crossed to fly lines carrying mutations in the innate immune pathways (Toll, Imd and Jak/STAT) to determine the role these pathways play a role in curtailing SIN replication. SIN replication remained unchanged in flies that were heterozygous for NF-κB orthologs Dif and Dorsal (activated by the Toll pathway) however SIN replication was higher in flies heterozygous for Relish. SIN replication was also enhanced in flies heterozygous for upstream members of the Imd pathway. Furthermore, intrathoracic injections of SIN virus into relish−/− flies showed higher viral loads and enhanced replication in mutant flies compared to wild type. These findings demonstrate that the Imd pathway is involved in antiviral defense against SIN and provide the first direct evidence for the involvement of the Imd pathway in antiviral defense in insects.
The UAS/GAL4 system allows for targeted gene expression by selective activation of any cloned gene in a wide variety of tissue- and cell-specific patterns [41]. We utilized this system to introduce RNA analogous to the genome of the alphavirus- SIN into Drosophila. Alphavirus genomes can be engineered to express heterologous proteins by substituting the structural protein genes with the heterologous protein gene. This replicon RNA is capable of self-replication but is not able to produce infectious virus particles. The SIN replicon used contains the nonstructural protein genes encoding the viral replicase, the 5′- and 3′-UTRs and a subgenomic promoter that directs expression of green fluorescent protein (GFP). A DNA copy of the SIN replicon genome was cloned behind five UAS enhancer sequences and a minimal heat shock promoter. Transcription from these upstream sequences is activated by the yeast transcriptional activator GAL4 expressed from a specific enhancer/promoter, hence primary transcription of the replicon RNA occurs in temporal and spatial pattern analogous to the gene from which the enhancer/promoter driving GAL4 expression was derived.
We generated two transgenic fly lines; 1) UAS-SINrep:GFP encodes a SIN replicon RNA capable of GFP expression from the subgenomic mRNA, and 2) UAS-SINΔrep:GFP encodes a mutant form of SIN replicon lacking sequence coding for the nonstructural proteins and hence is incapable of replication. When these fly lines are crossed to “driver” lines expressing GAL4 RNA pol II-mediated transcription of the replicon RNA is activated in the progeny. A schematic of the RNAs encoded by these flies is shown in Figure 1A.
While primary, cell-based, transcription of the SIN replicon RNA is under the control of the UAS/GAL4 system (Figure 1B, steps 1 and 2), GFP expression from this RNA is dependent on the replicon encoded viral RNA synthetic complex comprised of the viral nonstructural proteins. This complex copies the plus-strand replicon RNA (analogous to the SIN genome) into a minus-strand copy which in turn serves as a template for plus-strand RNA synthesis, both full-length replicon RNA and the subgenomic mRNA encoding GFP. Figure 1B shows the hypothesized launch of SIN replicon replication under the control of UAS/GAL4 (Figure 1B) compared to a natural virus infection (Figure 1C). Following the introduction of the viral genomic plus-sense RNA into the cytoplasm, which differs for each system (steps 1 and 2), the process of genome replication and subgenomic mRNA expression is the same for each system (steps 3 to 8), meaning host factors that inhibit or support viral genome replication are the same in each case.
We crossed UAS-SINrep:GFP line to an Act5C-GAL4 activator line to determine if alphavirus genome replication can be launched by Drosophila RNA polymerase II-mediated transcription. Primary transcription of the replicon RNA is dependent on the activity of the Act5C enhancer/promoter to drive expression of GAL4. Act5C was chosen as the driver for these experiments as it has been shown to be broadly expressed during Drosophila development and hence provided the greatest opportunity for driving primary transcription of the replicon RNA in a tissue that was permissive for viral RNA replication [42].
Expression patterns of GFP in Act5C-GAL4,UAS-SINrep:GFP flies (hereafter referred to as SIN replicon fly) were compared to those in the control Act5C-GAL4,UAS-GFP flies (hereafter referred to as control GFP fly). F1 progeny at various stages of development were examined for GFP expression. In the third instar of control GFP larvae, GFP expression was observed throughout the body with areas of high expression in the anterior end of the larvae, while SIN replicon-derived GFP expression was characterized by punctate areas of high expression throughout the body (Figure 2A and 2B). The expression pattern however changed in late pupae. Act5C driven GFP expression in the control was predominantly in the abdomen with low levels of expression in thorax and head (Figure 2A), whereas replicon derived GFP expression was predominantly in the thorax with little expression in the abdomen and head (Figure 2B). The pattern of expression in adult flies was similar to that of pupae. This result suggests that once viral RNA replication is initiated, the pattern of GFP expression is no longer defined by the pattern of GAL4 expression, and, as long as the cell carries replicon RNA in the cytoplasm and is permissive for viral RNA replication there is no requirement for continuous primary cell-mediated transcription of replicon RNA.
To determine the viral dependence of the GFP expression observed in the SIN replicon expressing flies we crossed the UAS-SINΔrep:GFP line to Act5C-GAL4 line and checked for GFP expression in F1 Act5C-GAL4,UAS-SINΔrep:GFP flies (hereafter referred to as mutant SIN replicon flies). Since the mutant SIN replicon RNA lacks a significant portion of the nonstructural protein coding region, it is not capable of replication and, therefore should be incapable of subgenomic mRNA synthesis and GFP expression. As hypothesized, we observed no GFP expression in the F1 progeny at any developmental stage (Figure 2C). This result demonstrated that the GFP produced in the SIN replicon flies was dependent on the viral non-structural proteins and was therefore a consequence of viral genome replication.
Results of observed GFP fluorescence were confirmed by qRT-PCR of GFP mRNA. Extremely low levels of GFP mRNA were detected in flies containing the DNA copy of the SIN replicon but lacking the GAL4 driver (UAS-SINrep:GFP). In SIN replicon flies with the driver there was a ∼550-fold increase in the level of GFP transcripts (Figure 2D). Additionally replicon derived minus-strand replication intermediates were detected in Act5C-GAL4,UAS-SINrep:GFP by RT-PCR (supplementary data, Figure S1). These data suggest that binding of GAL4 activated primary transcription of the replicon RNA that then replicated autonomously leading to the production of high levels of GFP encoding RNA. In mutant SIN replicon flies without the driver there was again low levels of GFP mRNA detected, however in mutant SIN replicon flies with the driver there was a 2-fold change in GFP transcripts. This change in RNA levels can be attributed to GAL4 activation of primary transcription of mutant replicon RNA (Figure 2E).
Drosophila lines expressing replicon RNA, mutant replicon RNA, and GFP were stabilized in order to ensure the co-segregation of the GAL4 and UAS elements. These fly lines were used in the experiments that follow.
A previous study showed that flies homozygous for a mutant allele of Dicer-2 (dcr-2L811FXS) were more susceptible to SIN infection. Dicer-2 mutant flies when infected with SIN, had increased viral RNA accumulation and higher viral loads compared to wild type flies [24]. To determine if Dicer-2 played a role in controlling the level of SIN RNA replication following host-derived launch we crossed the SIN replicon fly with Dicer-2 mutant flies. The F1 progeny heterozygous for the SIN replicon and Dicer-2 showed increased levels of GFP indicating increased viral replication (Figure 3A). Viral RNA replication was measured by GFP fluorescence and GFP mRNA levels. The fluorescence and mRNA levels were 1.8 and 2- fold higher respectively in flies possessing only one functional copy of Dicer-2 when compared to SIN replicon flies homozygous for wt Dicer-2 (Figure 3C and 3E). There was no change in GFP expression levels or mRNA levels in the control GFP flies heterozygous for Dicer-2 (Figure 3B, 3D and 3F). Our results verified the previously reported role of Dicer-2 and RNAi in controlling SIN infection and confirmed that the UAS/GAL4 system for replicon launch could be used to genetically examine antiviral responses in Drosophila.
The antimicrobial pathways in Drosophila play a very important role in combating infections. The Toll pathway results in the activation of NF-κB homologues Dif and Dorsal, the Imd pathway activates Relish, while Jak-STAT pathway triggers STAT. These transcription factors are central to the pathways, in the sense that they activate the antimicrobial effector molecules that eventually eradicate microbes. To determine if any of the known antimicrobial transactivators function to inhibit SIN RNA replication we crossed the SIN replicon fly with Dif, Dorsal, Relish and STAT mutants and examined their effects on SIN RNA replication. We measured GFP expression in F1 progeny as a gauge of viral RNA replication. In flies heterozygous for SIN replicon and dif or dorsal mutations there was no change in the levels of GFP. However, there was a 2.3-fold increase in GFP levels in SIN replicon flies heterozygous for a relish mutation (Figure 4A and 4B). SIN replication measured by qRT-PCR of nsP1 mRNA showed similar results. Levels of nsP1 mRNA was 3- fold higher in flies heterozygous for relish mutation compared to wt fly background (Figure 4C). This result suggests that Relish-dependent transcription may be involved in the suppression of SIN replication. Flies heterozygous for SIN replicon and stat also displayed increased SIN replication. GFP expression levels were 1.7- fold higher in STAT mutant flies compared to SIN replicon flies, suggesting that STAT might also play a role in inhibiting SIN RNA replication.
Relish is activated as a result of signaling through the Imd pathway, therefore the data above indicated that this pathway is involved in an antiviral response. To determine the role of the Imd pathway in the suppression of SIN RNA replication we crossed the SIN replicon fly to flies mutant in upstream components of the Imd pathway and examined their effect on virus replication. SIN replication was measured by qRT-PCR of nsP1 mRNA. Replication of SIN RNA increased in flies containing mutations in the Imd pathway (Figure 5A). The levels of nsP1 transcript were 2.8, 2.7, 4.5 and 3.3- fold higher in F1 progeny heterozygous for relish, imd, dfadd and dredd respectively when compared to the replication in a wt fly background. Similarly, nsP1 mRNA levels were also higher by 1.5, 3.6 and 3.1- folds in flies heterozygous for tab2, ird5 and key respectively compared to levels in wt flies. We also measured SIN replication via levels of GFP transcript. The levels of GFP transcripts were 2.3, 2.4, 3.7 and 3.9- fold higher in F1 progeny heterozygous for relish, imd, dfadd and dredd respectively when compared to the replication in a wt fly background. GFP mRNA levels were also higher by 2.4, 3.1 and 2.1- folds in flies heterozygous for tab2, ird5 and key respectively compared to levels in wt flies (Figure 5B). These data demonstrate that the Imd pathway plays a role in the control of SIN genome replication.
The activation of Relish leads to transcription of the AMPs Diptericin and Metchnikowin, while activation of Toll pathway leads to the expression of Drosomycin. Levels of these transcripts were used as markers of antimicrobial pathway activation. A large induction of both Diptericin and Metchnikowin was detected in SIN replicon flies. Diptericin was up by 4.8- fold and Metchnikowin by 9.2- fold as compared to w1118 flies. However, there was no difference in the levels of Drosomycin in SIN replicon flies when compared to w1118 flies. These results confirmed that SIN replicon was stimulating the Imd pathway that activated Relish and expression of AMPs. We also examined the expression of these AMPs in mutant SIN replicon flies expecting no increase in Relish-dependent mRNA expression of AMPs in these flies since viral replicon replication was not occurring. However, a 2.4 and 3.1- fold increase in diptericin and metchnikowin transcripts respectively was detected in the mutant SIN replicon flies (Figure 5C). This suggests that viral replication is being detected through recognition of the viral RNA and replication is not necessary for stimulation of this pathway. It is important to note that significant increases in the levels of AMP encoding mRNAs were not observed as a consequence of UAS/GAL4 based GFP expression, demonstrating that Relish activation is not occurring simply as a consequence of over-expression of a heterologous gene (Figure 5C).
Finally to confirm the role of Relish in antiviral defense against alphavirus in Drosophila, we infected relish−/− flies, dif−/− intrathoracically with 200 pfu of SIN. Viral loads were measured by levels of RNA containing nsP1 sequence. Relish mutant flies had 9.3- fold higher level of viral RNA compared to w1118 flies 5 days post-infection (Figure 6A). The levels of viral RNA however remained the same in dif−/− flies as compared to w1118 flies confirming that Imd but not Toll pathway is involved in controlling SIN infection. Also, SIN viral titers were 3-fold higher in relish−/− flies compared to w1118 and dif−/− flies (Figure 6B).
The anti-viral role of relish was also verified by overexpressing relish. UAS-Relish.his6 flies were crossed to a hemocyte GAL4 driver. The hemocyte driver was chosen to maximize expression of Relish [43]. The F-1 progeny overexpressing Relish were injected with 200 pfu of SIN virus intrathoracically and viral replication was measured five days post- infection. The levels of viral RNA in flies overexpressing Relish was down by 0.51-fold as compared to wt w1118 flies or UAS-Relish.his6 flies without GAL4 driver (Figure 6C).
The Drosophila Relish consists of an N-terminal Rel/NF-κB homology domain (RHD) and a C-terminal IκB-like domain with ankyrin repeats. Relish is activated by endoproteolytic cleavage, the RHD translocates to the nucleus and the IκB domain is retained in the cytoplasm. The RHD binds to DNA and activates the transcription of AMPs [44]. Although the exact mechanism of activation of mosquito Relish is still not known, mosquitoes produce three isoforms of Relish from the rel2 gene by differential mRNA splicing. The first Relish isoform resembles the Drosophila Relish; it contains the RHD and IκB -like domain. The second isoform has a RHD but lacks the IκB-like domain but has a unique 3′-UTR. The third isoform lacks the RHD but has an intact IκB-like domain. To verify the relevance of our findings in Drosophila we examined the cellular localization of the N-terminal RHD in uninfected and SIN infected cultured mosquito cells (c6/36). Infected and uninfected cells were fractionated into cytoplasmic and nuclear fractions and quantities of the RHD were examined by western blot. While levels of the RHD were consistently high in the cytoplasm of both infected and uninfected cells, we observed that SIN infection repeatedly resulted in an increase in the amount of the RHD in the nucleus of infected cells following 48 h of infection (Figure 7). This strongly implies that SIN activates Relish-mediated transcription during persistent infection of cultured mosquito cells.
In the present study we have developed a powerful system to genetically examine the effects of host factors in suppressing SIN RNA replication in Drosophila. SIN RNA replication was launched by cellular transcription of the DNA copy of the viral genome using the UAS-GAL4 system. We crossed the SIN replicon fly to flies carrying mutations in specific components of antimicrobial pathways to determine their role in anti-SIN defense. Replication of SIN RNA was higher in flies heterozygous for a mutation in relish (Imd pathway) but not for dif or dorsal (Toll pathway). Additionally, SIN replication was higher in flies heterozygous for upstream components of the Imd pathway. Furthermore, intrathoracic injections of SIN virus into relish−/− flies showed higher viral loads and enhanced replication whereas SIN replication was unchanged in dif−/− flies. These findings indicate that the Imd pathway is involved in antiviral defense against SIN. This is the first report of the Imd pathway's involvement in antiviral defense in Drosophila.
The data presented in this manuscript demonstrated that SIN replicon mediated RNA synthesis could be launched from the Drosophila genome using the UAS-GAL4 system. Using the Act5C-GAL4 driver we observed robust SIN RNA replication at numerous stages of development. The replicon replication was not pathogenic, although there was 1–2 day delay in the developmental cycle. The pattern of GFP expression resulting from the SIN replicon was different from that observed in the control UAS-GFP fly and was not ultimately defined by the pattern of driver expression. We hypothesize that cells that contained replicating replicon RNA at an early stage in development, when act5C expression is ubiquitous [42], continued to host viral RNA synthesis at later developmental stages even in the absence of primary transcription of the replicon RNA from the fly genome. The pattern of GFP expression in the SIN replicon containing flies also showed that not all tissues are permissive for virus genome replication. For instance while Act5C-GAL4 drives primary UAS-dependent transcription in the abdomen, the lack of GFP signal in the abdomen of replicon-containing flies suggests tissues in this body segment are significantly less permissive for virus replication than thoracic muscle in which replicon derived GFP expression was high.
This system for the launch of SIN RNA replication allows a significant amount of control over where and when replicon RNA is produced in the developing fly. By using different GAL4 drivers we can launch replicon RNA production in a temporal and spatially specific fashion. This provides a greater degree of control than with other transgenic systems of virus launch [24],[40] in which a generalized heat-shock is used to induce cell-mediated transcription of the viral RNA. This flexibility has allowed us to begin to map the permissivity of tissues for viral RNA replication during fly development, including salivary glands, muscle, mid-gut, and CNS (data not shown). A consistent finding when analyzing viral RNA replication using different drivers to launch SIN replicon replication has been that thoracic muscle is highly permissive for virus replication confirming our initial findings with the Act5C driver.
Additionally this system allows us to genetically screen for host factors that are both pro- and antiviral. Our confirmation of the previously reported involvement of the RNAi pathway in the control of SIN replication led us to examine the role other antimicrobial pathways play in the control of SIN RNA synthesis [24]. Examination of the effects of transcription factors associated with antimicrobial signaling pathways revealed SIN replicon replication was 2.3- fold higher in flies heterozygous for Relish mutation. Relish is the terminal transcription factor in the Imd signaling cascade. This pathway is usually activated by Gram-negative bacteria, however activation can also occur by some fungi that do not activate the Toll pathway [45]. Our observation of enhanced RNA replication in flies deficient in components of the Imd pathway, in combination with the observed increase in Relish dependent transcription in flies harboring SIN replicon RNA, has for the first time demonstrated that Imd/Relish pathway is activated by a virus. Earlier studies have indirectly implied a role for viruses in activation of the Imd pathway but have not found specific antiviral effects. Zambon et.al. found that DXV activates sets of AMPs transcribed by both Relish and Dif but only the Toll pathway was involved in anti-DXV defense [33]. Sanders et.al. studied the transcriptional expression profiles of mosquito midguts infected with SIN. Based on the expression pattern they hypothesized that early innate immune responses to SIN infection was through Toll pathway, which is later shut-off and the Imd pathway is activated later in infection [38]. Also, RNAi mediated silencing of native regulators- Cactus (Toll) and Casper (Relish) in mosquitoes followed by infection with Dengue virus activated a considerable number of genes by Relish, however only genes activated by Dif had anti-Dengue effect [39]. Our analyses of Dif and Dorsal imply that the Toll pathway does not have an antiviral effect at the level SIN RNA replication.
While we have strong evidence that the Imd pathway is activated in response to SIN RNA, how the Imd pathway is activated by SIN remains an open question. Gram-negative bacteria activate the Imd pathway when bacterial proteoglycans are recognized by host transmembrane receptors - PGRP-LC and PGRP-LE. This binding leads to the recruitment of Imd by an unknown protein [16]. The SIN replicon does not produce proteins resembling the peptidoglycans of bacteria. We therefore assumed that PGRPs have no role in SIN replication. To address this assumption we measured SIN replication in PGRP-LE and LC mutant flies and we found no difference in the replication of SIN in flies heterozygous for PGRP-LE and LC (Figure S2). We believe that in the SIN replicon flies, the activation of the Imd pathway occurs within the cell since the replication complexes and the replicating RNA are intracellular. We therefore hypothesized that an intracellular receptor recognizes either viral RNA or replication complexes and feeds a signal into the Imd pathway that ultimately activates Relish. Dicer-2 serves as a cytoplasmic sensor of viral RNA similar to mammalian RIG-I and Mda5 and induces the expression of antiviral protein Vago [46]. Since our data indicate that viral RNA is responsible for Relish activation, we examined the role of Dicer-2, and also Dicer-1 in activation of the Imd pathway. However to this point, we have observed no role for Dicer-2 or Dicer-1 in induction of Relish-mediated transcription in SIN replicon flies (Figure S3).
It is currently unclear what the effectors of the Relish-mediated anti-SIN response might be. While increased transcription of the AMP genes diptericin and metchnikowin was used as a measure of Relish activation, a role for these AMPs in antiviral immunity seems unlikely. The induction of AMP expression as a consequence of SIN replication may however be an important prophylactic immune response preventing secondary bacterial infection due to virus-induced tissue damage. We are currently performing comparative transcriptome analyses to identify differences in transcript levels unique to the SIN replicon flies in order to facilitate the identification of antiviral effector molecules.
Another variable that may significantly affect the antiviral response of Drosophila is the presence of Wolbachia. Wolbachia are Gram-negative bacteria that manifest intracellular, inheritable infections. In Drosophila melanogaster the infection is transmitted vertically through the female, and previous studies have reported that Drosophila infected with Wolbachia are less susceptible to infections with RNA viruses [47],[48]. We tested all the lines used in this study for presence of Wolbachia by PCR. Among the lines tested two lines were positive for Wolbachia, the Dif mutant line (w[1118];Df(2L)Exel8036/CyO) and PGRP-LC mutant (w67c23 P{lacW}l(1)G0414G0414/FM7c) line (Figure S4). In our crosses we used females from the SIN replicon expressing line (Act5C-GAL4/UAS-SINrep:GFP) that were negative for Wolbachia and males from the mutant lines. Since Wolbachia is transmitted maternally, the progeny resulting from these crosses are not infected with Wolbachia, and hence virus replication was consistently analyzed in a Wolbachia negative background.
SIN replication was enhanced in STAT mutant flies suggesting that the Jak/STAT pathway may also be involved in controlling SIN replication. Previous data have shown DCV infection induces the expression of vir-1 and expression of vir-1 is dependent on Hopscotch- the Jak kinase. Further genetic experiments suggested that Hopscotch was required but not sufficient for the induction of DCV -regulated genes [31]. It is possible that SIN infection activates both Imd and Jak/STAT pathways and that multiple pathways are required for effective viral clearance. However the potential role of Jak/STAT in SIN infections needs to be completely understood.
While Drosophila represents a genetically accessible model organism, alphaviruses are naturally transmitted between vertebrate hosts by mosquitoes. The mosquito genome has orthologous genes for dif and relish; rel1 and rel2 respectively [49]–[52]. We verified the relevance of our findings in Drosophila by infection of cultured mosquito cells. The results indicated that Rel-2 is activated during SIN infection of c6/36 cells. The RHD containing isoforms of Rel-2 localize to the nucleus later during infection. These results imply that Relish-mediated transcription may be important in controlling virus replication during the persistent phase of infection in mosquitoes. These results also suggests that the results generated by using Drosophila as model organism can be compared and verified in mosquito cells.
In summary, we have developed a system for the controlled launch of SIN RNA replication from the genome of Drosophila. Using this system we have demonstrated that, in addition to RNAi, Jak/STAT, and Toll, the Imd pathway plays an important role in the antiviral response in flies. Further characterization of how the virus is recognized by the host and what downstream effector molecules are required for the control of virus replication will provide additional insights into the role of this pathway in particular and the antiviral response of arthropods in general.
BHK-21 and C6/36 cells (American Type Culture Collection) were grown in Alpha MEM and L15 media respectively (Invitrogen) supplemented with vitamins 10% of fetal bovine serum or heat inactivated FBS (C6/36). SIN:GFP is wild type SIN expressing GFP from a second subgenomic promoter was generated by transfection of BHK-21 cells with in-vitro transcribed infectious SIN:GFP TE RNA [53].
The pUAST- SINrep: GFP plasmid was constructed by replacing the Sbf1 and Not1 fragment of pUAST vector with pSINrep/GFP that encodes the non-structural proteins and GFP from a sub-genomic promoter preceded by 5 UAS sequences. The pUAST- SINΔrep: GFP construct was made by deleting 5.7 kb fragment in the non-structural region of pUAST- SINrep: GFP. RsrII and KpnI were used to remove the 5.7 kb fragment, and the remaining product gel purified and treated with DNA polymerase I large (Klenow) fragment (NEB) to remove the 3′ overhang and fill-in the 5′ overhang. The plasmid was the phenol/chloroform extracted, precipitated and ligated using T4 DNA ligase (NEB). Stable transgenic lines harboring the UAS-SIN constructs were generated via standard methods [54]. We obtained a transformant line for SINrep: GFP that mapped to the third chromosome and one line for SINΔrep: GFP that mapped to the second chromosome.
Fly lines (listed in Table S1) were obtained from the Bloomington stock center. dcr-1Q1147X and dcr-2L811FXS flies were provided by R Carthew (Northwestern University). Dif1 were provided by D Ferrandon (Institut de Biologie Moléculaire et Cellulaire). Fly stocks were raised on standard cornmeal–agar medium at 25°C.
Live flies, pupae and larvae were anesthetized with CO2 and viewed under on a Nikon SMZ1500 dissecting microscope. Photographs were taken using Nikon DXM1200 camera.
Five adult flies were homogenized in 10 mM Tris (pH 8.4), 100 mM NaCl, 1 mM MgCl2, 10 mM dithiothreitol [55]. Homogenates were centrifuged at 15000 g for 5 min to remove debris and fluorescence was detected using a Synergy 4 HT Multi-Detection Microplate Reader (Biotek) with excitation filter set to 485 nm and emission filter at 520 nm.
For viral injections, flies were anesthetized with CO2 and injected with SIN:GFP virus or control alpha MEM media in the thorax using a glass capillary needle. To estimate the number of viral plaque forming units injected into flies, injected flies were immediately flash frozen in liquid nitrogen, homogenized in PBS, centrifuged at 15000 g for 15 minutes to remove the debris and viral titers determined by plaques assays of homogenates. Approximately 200 pfu of Sin:GFP virus was injected into the flies. Five days post-infection flies were collected and viral titers determined as mentioned above. For the survival experiments, the injected flies were put on fresh food, and the number of surviving flies was counted at regular intervals.
RNA was extracted by homogenizing flies in TRIzol reagent (Invitrogen). cDNA was made using AffinityScript QPCR cDNA synthesis kit (Stratagene), and PCR amplification was done using Brilliant II SYBR Green QPCR master mix (Stratagene) following manufacturer's protocol. Gene expression was normalized to the actin mRNA expression. The comparative threshold cycle (CT) method was used to determine fold changes of transcript present in samples. Oligonucleotides used are listed in Protocol S1.
C6/36 cells were not infected or infected with SIN:GFP virus at MOI of 0.1 for 6 h and 48 h. Western blot analysis was performed using standard procedures. Rabbit anti-N Rel antibody (kindly gifted by S. Stoven of Umea University) was used to detect Relish.
The FlyBase (http://flybase.org/) accession numbers for the genes used in the text include actin5C (CG4027), dfadd (CG12297), dicer1 (CG4792), dicer2 (CG6493), dif (CG6794), diptericin (CG12763), dorsal (CG6667), dredd (CG7486), drosomycin (CG10810), imd (CG5576), ird5 (CG4201), kenny (CG16910), metchnikowin (CG8175), pgrp-lc (CG4432), pgrp-le (CG8995), relish (CG11992), stat92E (CG4257), tab2 (CG7417).
|
10.1371/journal.pgen.1007170 | Cdc73 suppresses genome instability by mediating telomere homeostasis | Defects in the genes encoding the Paf1 complex can cause increased genome instability. Loss of Paf1, Cdc73, and Ctr9, but not Rtf1 or Leo1, caused increased accumulation of gross chromosomal rearrangements (GCRs). Combining the cdc73Δ mutation with individual deletions of 43 other genes, including TEL1 and YKU80, which are involved in telomere maintenance, resulted in synergistic increases in GCR rates. Whole genome sequence analysis of GCRs indicated that there were reduced relative rates of GCRs mediated by de novo telomere additions and increased rates of translocations and inverted duplications in cdc73Δ single and double mutants. Analysis of telomere lengths and telomeric gene silencing in strains containing different combinations of cdc73Δ, tel1Δ and yku80Δ mutations suggested that combinations of these mutations caused increased defects in telomere maintenance. A deletion analysis of Cdc73 revealed that a central 105 amino acid region was necessary and sufficient for suppressing the defects observed in cdc73Δ strains; this region was required for the binding of Cdc73 to the Paf1 complex through Ctr9 and for nuclear localization of Cdc73. Taken together, these data suggest that the increased GCR rate of cdc73Δ single and double mutants is due to partial telomere dysfunction and that Ctr9 and Paf1 play a central role in the Paf1 complex potentially by scaffolding the Paf1 complex subunits or by mediating recruitment of the Paf1 complex to the different processes it functions in.
| Maintaining a stable genome is crucial for all organisms, and loss of genome stability has been linked to multiple human diseases, including many cancers. Previously we found that defects in Cdc73, a component of the Paf1 transcriptional elongation complex, give rise to increased genome instability. Here, we explored the mechanism underlying this instability and found that Cdc73 defects give rise to partial defects in maintaining telomeres, which are the specialized ends of chromosomes, and interact with other mutations causing telomere defects. Remarkably, Cdc73 function is mediated through a short central region of the protein that is not a part of previously identified protein domains but targets Cdc73 to the Paf1 complex through interaction with the Ctr9 subunit. Analysis of the other components of the Paf1 complex provides a model in which the Paf1 subunit mediates recruitment of the other subunits to different processes they function in. Together, these data suggest that the mutations in CDC73 and CTR9 found in patients with hyperparathyroidism-jaw tumor syndrome and some patients with Wilms tumors, respectively, may contribute to cancer progression by contributing to genome instability.
| Gross chromosomal rearrangements (GCRs), such as translocations and deletions, are common in many cancers [1]. DNA repair and DNA damage signaling defects that cause increased rates of accumulating GCRs in model systems like Saccharomyces cerevisiae have been identified in sporadic tumors and in inherited cancer predisposition syndromes, suggesting that increased genome instability plays a role in the development of some cancers [2–7]. In addition to defects in DNA metabolism [8,9], defects in transcription are also a source of genome instability. How transcriptional defects cause GCRs is not completely understood, but collisions with the replication machinery, formation of RNA:DNA hybrids, and/or transcription-associated homologous recombination (HR) are potential mechanisms [10,11].
Recently we identified CDC73 in a large-scale screen for genes that suppress the formation of GCRs in S. cerevisiae [6]. CDC73 encodes a subunit of the Paf1 complex, and CDC73 has been previously implicated as playing a role in maintaining the stability of yeast artificial chromosomes, chromosome transmission fidelity, and suppression of direct repeat HR [12–14]. The Paf1 complex, which is comprised of Paf1, Cdc73, Rtf1, Ctr9, and Leo1, binds to and modifies the activity of RNA polymerase during transcription [15–20]. This complex has been implicated in a variety of processes, including transcription elongation, mRNA 3’-end maturation, histone methylation and ubiquitination, expression of normal levels of telomerase RNA TLC1 and maintenance of normal telomere lengths [16,21–24], and is conserved among eukaryotes [25]. Somatic mutations in CDC73 in humans are associated with breast, renal, gastric, and parathyroid cancers [26–28], and germline mutations in CDC73 cause the cancer susceptibility syndrome hyperparathyroidism-jaw tumor syndrome (HPT-JT) [29,30]. In addition, a small fraction of familial Wilms tumor cases have been attributed to germline mutations in CTR9 [31]. However, little is known about how CDC73 and CTR9 function as tumor suppressors, particularly since mutations in the genes encoding the other members of the Paf1 complex have not yet been linked to the development of cancer.
Here we have investigated how the Paf1 complex acts to suppress genome instability with the goal of shedding light on how the human homolog of CDC73 may function as a tumor suppressor. We have found that PAF1, CDC73, and CTR9 play the most important roles in suppressing the accumulation of GCRs among the genes that encode subunits of the Paf1 complex. Strains with CDC73 defects appear to have perturbations in telomere maintenance that result in increased GCR rates and that these defects result in synergistic increases in GCR rates when combined with defects in TEL1 and YKU80, which cause other types of defects in telomere maintenance that also result in increased GCR rates. Deletion analysis identified a 105 amino acid region of Cdc73 that was necessary and sufficient for its incorporation into the Paf1 complex, nuclear localization, and Cdc73 function. These analyses enhance our understanding of how Cdc73, as a subunit of the Paf1 complex, suppresses genome instability, and provide insights into how its human homolog may function as a tumor suppressor.
Because we previously identified CDC73 as a genome instability suppressing (GIS) gene [6], we tested if other genes encoding subunits of the Paf1 complex suppressed the formation of GCRs selected in the duplication-mediated GCR (dGCR) assay (Fig 1A). The cdc73Δ, ctr9Δ, and paf1Δ single mutations caused the largest increases in dGCR rate (9–22 fold), and the leo1Δ and rtf1Δ single mutations caused small increases in the dGCR rate (3–4 fold; Fig 1B; S1 Table). As will be discussed in detail below, we found that the cdc73Δ mutation caused a synergistic increase in the dGCR rate when combined with yku80Δ or tel1Δ mutations (Fig 1B) and tested if the yku80Δ or tel1Δ mutations synergized with deletions of other Paf1 complex genes. Similar to the effects of the single mutations, the cdc73Δ, ctr9Δ, and paf1Δ mutations caused strong synergistic increases in the dGCR rate when tested in combination with either a yku80Δ or tel1Δ mutation whereas the rtf1Δ and leo1Δ mutations did not cause a synergistic increase in the dGCR rate in combination with either a yku80Δ or tel1Δ mutation relative to the respective single mutations (Fig 1B; S1 Table). Interestingly, the mutations that caused the strongest increases in GCR rates in these experiments, cdc73Δ, paf1Δ and ctr9Δ, caused the largest decreases in telomere lengths and TLC1 levels along with causing strong defects in telomere gene silencing (see below), whereas the mutations that caused little if any increases in GCR rates in these experiments, rtf1Δ and leo1Δ, also caused the smallest decreases in telomere lengths and TLC1 levels [32].
Given the differences in the roles of the Paf1 complex subunits in suppressing the accumulation of GCRs, we also tested if transcriptional elongation defects, which are caused by Paf1 complex defects [33–35], might correlate with the increased GCR rates in mutant strains. We measured transcriptional elongation defects that result in sensitivity to 6-azauracil, which depletes cellular rGTP levels [36]. Deletion of PAF1 or CTR9 caused strong sensitivity to 6-azauracil, deletion of CDC73 caused weaker sensitivity, and deletion of RTF1 or LEO1 caused no sensitivity (Fig 1B; S1A Fig). These results are in accord with the results of studies employing other transcriptional elongation assays [33–35]; however, it should be noted that the magnitude of the effect caused by defects affecting the Paf1 subunits, including Cdc73, varies between the transcriptional elongation assays used, and 6-azauracil sensitivity assays can show strain-to-strain variation [37].
Strains with deletions of PAF1 or RTF1 have defects in the silencing of telomere-proximal genes (CDC73, CTR9 and LEO1 were not tested) [23], which has been termed the telomere position effect (TPE) [38] and deletions in PAF1, CTR9, RTF1, and to a lesser extent CDC73, but not LEO1 cause defects in the histone H3 modifications required for gene silencing including TPE [23,39–41]. To determine if TPE defects correlated with increased GCR rates, we measured TPE by monitoring the survival of strains with a telomere-proximal URA3 gene in the presence of 5-fluoroorotic acid (5FOA), which is toxic to strains expressing URA3. Deletion of PAF1 and RTF1 caused the greatest loss of TPE (Fig 1B; S1A Fig), whereas milder TPE defects were observed in cdc73Δ and ctr9Δ strains, and no TPE defect was observed in the leo1Δ strain. The stronger TPE defects caused by the paf1Δ and rtf1Δ mutations are consistent with the known role of Paf1 and Rtf1 in the specific recruitment of histone modifiers [23,39]. To verify that the TPE defects in the cdc73Δ strain were due to loss of telomere silencing and not due to induction of ribonucleotide reductase, which accounts for the apparent TPE defect in pol30-8 and cac1Δ strains [42], we tested the 5FOA sensitivity of the cdc73Δ strain in the presence of sublethal concentrations of hydroxyurea (HU), which rescues the TPE in pol30-8 and cac1Δ strains [42]. Consistent with the results in the absence of HU, growth on 5FOA-containing plates was not restored by addition of HU (S1B Fig). We did not test the deletion of the other Paf1 complex genes because their role or lack of a role in transcriptional silencing is well established [23,39–41] and because paf1Δ and ctr9Δ strains are HU sensitive [37].
The data presented here along with published data [32] suggest that PAF1 plays important roles in genome stability, transcriptional elongation, telomere silencing, maintaining TLC1 levels, and telomere length maintenance. CDC73 has an important role in genome stability, maintaining TLC1 levels, telomere length maintenance and a lesser but detectable role in telomere silencing but little if any role in transcriptional elongation. CTR9 has important roles in genome stability, transcriptional elongation, maintaining TLC1 levels, and telomere length maintenance, and a role in telomere silencing that was similar to that observed for CDC73. RTF1 has the most important role in telomere silencing, but plays little if any role in genome stability and transcriptional elongation, and lesser roles in maintaining TLC1 levels, and telomere length maintenance. And LEO1 plays little if any role at all in the Paf1 complex functions considered here and only a modest role in maintaining TLC1 levels and telomere length maintenance. These observations suggest a model for the complex in which Paf1 facilitates the functions of the other subunits potentially by mediating recruitment of the complex to the different processes it functions in (Fig 1C), consistent with the results of coimmunoprecipitation experiments in S. cerevisiae and binding assays performed with human homologs [22,43,44].
Since PAF1 and CDC73 played the largest roles in suppressing GCRs and the cdc73Δ mutation caused fewer additional defects, we sought to understand how the Paf1 complex suppresses genome instability by focusing on CDC73. We crossed strains containing the dGCR assay and a cdc73Δ mutation or an rtf1Δ mutation as a control to a 638-strain subset of the S. cerevisiae deletion collection that contained deletions of known GIS genes and cooperating GIS (cGIS) genes [6,45]. The resulting haploid double mutant strains were scored by patch tests for the increased accumulation of CanR 5FOAR papillae that are a measure of the formation of GCRs relative to the single mutant strains (Fig 2A). Forty-three mutations caused increased strain patch scores when combined with the cdc73Δ mutation (Fig 2B); potential suppressive interactions were not investigated as slow growth phenotypes can also cause reduced strain patch scores. Selected interactors causing increased patch scores were verified by quantitative fluctuation assays (S2 Table). Almost none of the mutations that caused increased scores when combined with the cdc73Δ mutation interacted with an rtf1Δ mutation (Fig 2B), consistent with the more modest effects of rtf1Δ on GCR rates (Fig 1B).
Among the CDC73 interactors were 7 genes (YKU70, YKU80, TEL1, MRC1, NUP60, RAD6, and VPS20) in which mutations cause shorter telomeres [46–48]. Combined with reports that cdc73Δ strains have reduced levels of the telomerase RNA TLC1 [32], these results suggested that defects in telomere homeostasis could be responsible for some of the strong interactions. To extend these results, we focused on YKU80, YKU70, and TEL1 because the role of these genes in telomere homeostasis is better understood than the other 4 genes. A cdc73Δ mutation showed synergistically increased patch scores when it was combined with either yku70Δ or yku80Δ mutations, which disrupt the Ku complex and cause both shortened telomeres and non-homologous end joining (NHEJ) defects [49,50]. Quantitative rate measurements demonstrated that the cdc73Δ yku80Δ double mutant had a 162-fold increase in dGCR rate as compared to the 4- to 9-fold increase in dGCR rate seen for the respective single mutants (Table 1). In contrast, deletion of DNL4, which encodes the DNA ligase required for NHEJ but not telomere maintenance, did not result in a synergistic increase in GCR rates when combined with the cdc73Δ mutation (S2 Table) suggesting that the increased GCR rates of the cdc73Δ yku70Δ and cdc73Δ yku80Δ double mutants do not involve a defect in NHEJ.
Similarly, the cdc73Δ mutation showed a strong interaction with a tel1Δ mutation in the dGCR assays as measured by patch scores (Fig 2A and 2B), and the cdc73Δ tel1Δ double mutant had a 236-fold increase in the dGCR rate (Table 1). TEL1 encodes a protein kinase involved in the DNA damage checkpoint that also plays a role in maintaining normal telomere lengths such that a tel1Δ mutation causes shortened telomere lengths [53,54]. In contrast, mutant strains that contained a cdc73Δ mutation in combination with defects in other checkpoint genes either did not have increased dGCR patch scores (RAD9, DUN1, and RAD53) or only had small increases in dGCR patch scores (MEC3, RAD17, RAD24, and MEC1), supporting the view that the genetic interaction between the cdc73Δ and tel1Δ mutations reflects the telomere maintenance defect caused by the tel1Δ mutation. The tel1Δ yku80Δ double mutant had a 28-fold increase in the dGCR rate (Table 1 and S2 Table) and the cdc73Δ tel1Δ yku80Δ triple mutant had a 2024-fold increase in the dGCR rate (S2 Table), consistent with the hypotheses that loss of CDC73, YKU80, and TEL1 cause partial defects in different telomere maintenance pathways and that the increased GCR rates that result from combining mutations in these genes may reflect increased telomere maintenance defects.
GCRs selected in the dGCR assay are most commonly generated by non-allelic HR between the DSF1-HXT13 region on the left arm of chromosome V (chrV L) and divergent homologies on chrIV L, chrX L, and chrXIV R [51]. PCR analysis of GCRs formed in the cdc73Δ, cdc73Δ tel1Δ and cdc73Δ yku80Δ dGCR strains showed that the distribution of GCRs were essentially the same as that from the wild-type strain, despite the >200-fold increase in GCR rate in some of the strains analyzed (Fig 2C). As expected, introduction of the HR-defective rad52Δ mutation decreased the dGCR rates of the cdc73Δ tel1Δ and cdc73Δ yku80Δ double mutants by 45-fold and 25-fold, respectively (S2 Table). In addition, the rad52Δ mutation shifted the spectrum of GCRs recovered in the cdc73Δ mutant to GCRs that were not formed by non-allelic HR (Fig 2C).
As observed in the dGCR assay, synergistic increases in GCR rates were also observed when the cdc73Δ mutation was combined with either the yku80Δ or tel1Δ mutation in strains containing either the unique sequence (uGCR) assay or the short homology GCR (sGCR) assay (Table 1). Since the sGCR assay selects for a somewhat broader diversity of types of GCRs including de novo telomere additions than the uGCR assay and is not dominated by a single type of GCR as compared to the dGCR assay (summarized in Fig 1A), we used the sGCR assay to determine if the absence of CDC73 altered the distribution of the GCRs formed.
We analyzed 1 parental strain and 11 independent GCR-containing isolates by paired-end next-generation sequencing for the wild-type strain, the cdc73Δ single mutant strain, and the cdc73Δ tel1Δ and cdc73Δ yku80Δ double mutant strains (Fig 3; S3 and S4 Tables; S2–S8 Figs). In the wild-type sGCR strain, 46% of the GCRs analyzed (5 of 11) were produced by de novo telomere addition, 18% (2 of the 11) were produced by HR between the SUP53 tRNA gene introduced by the can1::PLEU2-NAT marker and another leucine tRNA gene, and 36% (4 of the 11) were produced by HR between the YCLWdelta5 fragment introduced by the can1::PLEU2-NAT marker and another Ty-related sequence (Fig 3A; S5,S9 and S10 Figs; S4 Table). The presence of both de novo telomere addition and HR-mediated GCRs among the GCRs selected in the sGCR assay is useful for characterizing mutations that alter the GCR spectra.
Analysis of GCRs formed in the cdc73Δ, cdc73Δ tel1Δ, and cdc73Δ yku80Δ sGCR strains revealed that no de novo telomere addition GCRs were recovered when CDC73 was deleted (Fig 3A; S6–S8 Figs; S4 Table). Remarkably, the majority of the GCRs selected in strains containing a cdc73Δ mutation were inverted duplications, and most of these contained a second breakpoint that was mediated by HR (Fig 3G; S11 Fig). Inverted duplications mediated by hairpins were frequent in the cdc73Δ tel1Δ sGCR strain (Fig 3H; S12 Fig), which is consistent with the previously observed increase in hairpin-mediated inverted duplications observed in the uGCR assay for the tel1Δ single mutant strain [52]. For inverted duplication GCRs, the initial inversion GCRs would be predicted to be dicentric, but in all cases identified here, these GCRs underwent additional rearrangements to generate stable monocentric chromosomes. These additional rearrangements commonly involved HR between repetitive elements on chrV L and other repetitive elements elsewhere in the genome, including an unannotated delta sequence on chrV R (S13 and S14 Figs). All of the GCRs observed other than de novo telomere addition-mediated GCRs were different types of translocations; the rates of accumulating these translocations in the sGCR assay relative to the wild-type rate were increased 52-fold for the cdc73Δ single mutant, 242-fold for the cdc73Δ tel1Δ double mutant, and 460-fold for the cdc73Δ yku80Δ double mutant sGCR strains. Most GCR-containing strains contained a normal complement of chromosomes, except for one cdc73Δ yku80Δ GCR-containing strain that contained two copies of chrXVI (S15 Fig).
Taken together, the shift in the GCR spectra in sGCR strains lacking CDC73 is consistent with an underlying defect in telomere homeostasis as most mutations that result in high GCR rates result in increased levels of de novo telomere addition GCRs as long as functional telomerase is present [55]. Given the limits on the numbers of GCRs we can presently sequence, our analysis cannot definitively prove that de novo telomere addition GCRs are not formed when CDC73 is deleted, but does demonstrate that other types of GCRs, which are all different types of translocations, are selectively increased (e.g., the increase in the rate of de novo telomere additions in the cdc73Δ mutant relative to the wild-type is <5-fold compared to a 52-fold increase in the rate of translocations). The relative lack of de novo telomere addition GCRs among the GCRs selected in the sGCR assay in strains containing cdc73Δ mutations could indicate a complete failure of de novo telomere additions, as is observed with strains with deletions of YKU80 or genes encoding telomerase subunits [55], or a partial defect that only decreases the efficiency of de novo telomere additions relative to other GCR-forming mechanisms, as is observed with tel1Δ strains [52,55]. We therefore combined the cdc73Δ mutation with a deletion of PIF1, which causes a substantial increase in GCRs formed through an increase in de novo telomere additions due to decreased inhibition of telomerase at DSBs [56,57], even under conditions where pif1 mutations potentially prevent the formation of GCRs mediated by break-induced replication [52,55]. Mutations inhibiting de novo telomere addition suppress the increased GCR rate caused by the pif1Δ mutation, whereas mutations causing only reduced efficiency of de novo telomere addition do not [52,55]. The cdc73Δ mutation partially suppressed the increased GCR rate caused by the pif1Δ mutation (Table 1), suggesting that the cdc73Δ mutation causes a substantial, but incomplete, defect in the formation of GCRs mediated by de novo telomere addition. This could be due to reduced levels of functional telomerase resulting from the partial reduction of TLC1 telomerase RNA levels observed in cdc73Δ mutants [32].
Deletions of CDC73, YKU70, YKU80 and TEL1 all result in shortened telomeres [32,46,47]. To investigate if cdc73Δ double and triple mutant strains have increased telomere defects in addition to increased GCR rates, we generated haploid single, double, and triple mutant strains containing different combinations of CDC73, EXO1, TEL1 and YKU80 deletions by crossing mutant strains to each other to generate fresh mutant haploid spore clones for telomere length analysis. Consistent with previous results [32,46,47], telomere lengths were reduced in cdc73Δ, and to a greater extent in tel1Δ, and yku80Δ single mutant strains (Fig 4A). Exo1 plays a role in resection of deprotected telomeres [58] and deleting EXO1 partially restored the shortened telomeres caused by the cdc73Δ, tel1Δ, and yku80Δ mutations; this is consistent with prior observations that exo1Δ yku80Δ double mutants have slightly longer telomeres than yku80Δ single mutants [59]. The tel1Δ yku80Δ, cdc73Δ yku80Δ, and cdc73Δ tel1Δ double mutant combinations all showed potential signs of additional telomere dysfunction compared to the respective single mutants, which included: (1) a telomere length that was shorter than seen in the respective single mutants (cdc73Δ tel1Δ) or potentially shorter than seen in the respective single mutants (tel1Δ yku80Δ, which was previously reported [60], and cdc73Δ yku80Δ); and (2) a smeared telomere pattern (tel1Δ yku80Δ and cdc73Δ yku80Δ), which was reminiscent of the telomere pattern seen in telomerase-defective post-senescent survivors that maintain their telomeres by alternative mechanisms [61]. Remarkably, the cdc73Δ tel1Δ yku80Δ triple mutant strain did not have a distinct telomere-containing band, but rather had only a smeared pattern, suggestive an even stronger telomere defect. The genetic interactions observed between the cdc73Δ, tel1Δ, and yku80Δ mutations resulting in increased telomere dysfunction mirrors the synergistic increases in GCR rates seen in strains containing combinations of these mutations.
The cdc73Δ single mutant and the tel1Δ yku80Δ, cdc73Δ yku80Δ, and cdc73Δ tel1Δ double mutants all grow slowly and have evidence of telomere defects. We therefore investigated whether or not these strains would show evidence of crisis, escape from senescence and improved growth by serially restreaking the mutant strains on non-selective medium (S16 Fig). To ensure that our serial restreaking procedure could detect senescence and recovery, we tested the tel1Δ yku80Δ double mutant strain and found it initially grew slowly but eventually recovered a wild-type growth rate as previously reported [62] (not illustrated). In contrast, the slow growth of the cdc73Δ single mutant and the even slower growth of the cdc73Δ yku80Δ, and cdc73Δ tel1Δ double mutants showed only partial improvement in growth after 11 rounds of restreaking and never achieved wild-type growth rates. One possible explanation for this difference is that telomere maintenance-independent effects on transcription could also contribute to the slow growth phenotype caused by the cdc73Δ mutation.
The telomere structures of these serially propagated strains were analyzed by Southern blot and the telomere species of the tel1Δ yku80Δ, cdc73Δ tel1Δ, cdc73Δ yku80Δ, and cdc73Δ tel1Δ yku80Δ strains were all observed to contain smeared telomere fragments (Fig 4B); this suggests that the telomeres in these mutants may be partially maintained by one of the RAD52-dependent telomerase-independent telomere maintenance pathways [61,63]. Consistent with this, the cdc73Δ tel1Δ rad52Δ and cdc73Δ yku80Δ rad52Δ triple mutants all had very short telomeres, but lacked the smeared pattern seen in the Southern blots (Fig 4B). In contrast, we were unable to generate a cdc73Δ tel1Δ yku80Δ rad52Δ quadruple mutant by either PCR mediated gene disruption or by crossing different mutant strains to each other; this is consistent with a requirement of RAD52-dependent HR in the cdc73Δ tel1Δ yku80Δ triple mutant either for telomere maintenance or for the repair of some other type of spontaneous DNA damage in this triple mutant.
Pulse field gel electrophoresis (PFGE) was used to analyze chromosomes from cdc73Δ single, double and triple mutant strains for the presence of aberrant sized chromosomes (Fig 4C). The cdc73Δ, tel1Δ, and yku80Δ single mutant strains and the cdc73Δ tel1Δ, cdc73Δ yku80Δ, and tel1Δ yku80Δ double mutant strains had chromosomal banding patterns that were similar to that from the respective wild-type strain, although the double mutants showed more chromosomes with abnormal sizes despite being grown in the absence of any selection for chromosome rearrangements. The cdc73Δ tel1Δ rad52Δ and cdc73Δ yku80Δ rad52Δ triple mutants had increased numbers of chromosomes with abnormal sizes compared to the respective cdc73Δ tel1Δ and cdc73Δ yku80Δ double mutants. In contrast, no chromosome bands were visible when the cdc73Δ tel1Δ yku80Δ triple mutant was analyzed, which is consistent with reports that chromosomes from post-senescent survivors are unable to enter PFGE gels, likely because of the structure of the HR intermediates that act in telomere maintenance [61]. The aberrant chromosomes observed in this experiment were not studied further; however, the structures of GCRs selected in many of these mutant strains have been determined (Fig 3).
We also investigated whether cdc73Δ single and double mutants with telomere defects had TPE defects. Consistent with previous results [64], we found that deletion of YKU80 caused significant TPE defects relative to wild-type cells and hence a decreased ability to grow on plates containing 5FOA (Fig 4D). In contrast, the cdc73Δ and tel1Δ single mutant strains had modest but easily detectible or no sensitivity to 5FOA, respectively (Fig 4D, S1A Fig, S17A Fig). However, the cdc73Δ yku80Δ and cdc73Δ tel1Δ double mutant strains showed increased sensitivity to 5FOA, suggesting increased perturbation of the chromatin structure proximal to the telomeres, and hence loss of silencing in these double mutants. Consistent with a synergistic defect in TPE rather than an indirect effect due to induction of ribonucleotide reductase [42], growth on 5FOA-containing plates was not restored by addition of HU (S1B Fig, S17A Fig, S20B Fig).
To test interactions between cdc73Δ and additional telomere homeostasis mutations, we measured the dGCR rates of strains containing a cdc73Δ mutation in combination with deletions of SIR2, SIR3, or SIR4, which cause defects in TPE, telomere chromatin structure and, at least in the case of SIR3 and SIR4 (SIR2 does not appear to have been tested) also cause shortened telomeres [61,65], but were missing from our screen as these genes are required for mating [61,66]. The single sir2Δ, sir3Δ and sir4Δ mutant dGCR rates were increased 6 to 8-fold relative to the wild-type dGCR rate, and the double mutation combinations with the cdc73Δ mutation resulted in a synergistic increase in dGCR rates that were 41 to 190-fold higher than the wild-type dGCR rates (S2 Table).
In contrast, only 9 of the 36 mutations tested (including sir3Δ and sir4Δ) that were known to cause shortened telomeres [46–48,67] resulted in synergistic increases in dGCR rates when combined with cdc73Δ (S18 Fig). However, of the 27 mutations that did not interact, 3 mutations caused extremely high GCR rates and 1 mutation was in a Paf1 complex genes making it unlikely that interactions could be detected. Of the remaining 23 non-interacting mutations, many caused weak or inconsistent phenotypes (lst7Δ was reported to cause both long and short telomeres), 20 were identified in only one of two genetic screens performed suggestive of causing weak or inconsistent phenotypes and in most cases have not yet been demonstrated as causing a defect in a specific aspect of telomere homeostasis such as defects in TPE. Moreover, the cdc73Δ mutation also caused a strong synergistic increase in the dGCR rate when combined with a deletion of EXO1 (Table 1). EXO1 encodes a 5’ to 3’ exonuclease that acts in different DNA repair pathways and is the primary nuclease that resects deprotected telomeres [68–70]. Unlike the case of the cdc73Δ mutation, combining the exo1Δ mutation with either a yku80Δ or a tel1Δ mutation did not cause synergistic increases in the dGCR rate (S2 Table). Taken together, these data do not argue that the cdc73Δ mutation causes synergistically increased GCR rates in strain backgrounds that have short telomeres per se. Rather, the interaction of cdc73Δ with tel1Δ and yku80Δ may reflect an interaction between mutations that disrupt specific aspects of telomere structure including telomere chromatin structure [61,65], nuclear localization of telomerase [71,72], and/or telomerase recruitment to telomeres [73,74].
The data described above are consistent with a role for CDC73 in suppressing genome instability arising due to telomere dysfunction. This effect could be due to roles of CDC73 in promoting TLC1 transcription [32] or causing defects in transcriptional elongation that give rise to recombinogenic RNA:DNA hybrids (R-loops) [75–78], particularly at the sites of long noncoding telomeric repeat containing RNA (TERRA) [79]. We measured the TLC1 levels in cdc73Δ, tel1Δ, and yku80Δ single and double mutant strains and found that the yku80Δ and cdc73Δ mutations caused a small and large decrease in TLC1 levels, respectively, as previously reported [32] and that the cdc73Δ tel1Δ and cdc73Δ yku80Δ double mutants had the same level of TLC1 as the cdc73Δ single mutant (S17B Fig). Introduction of a plasmid expressing TLC1 into strains in the uGCR assay caused a statistically significant ~4-fold decrease in the GCR rate of the cdc73Δ tel1Δ double mutant and caused a small, but not statistically significant, decrease in the GCR rate of the cdc73Δ yku80Δ double mutant (S5 Table). Consistent with the suppression results, the TLC1 expression plasmid caused 1) increased TLC1 levels in all strains tested, 2) increased the telomere lengths in the cdc73Δ single mutant and the cdc73Δ tel1Δ double mutant, and 3) potentially a small increase in telomere length in the cdc73Δ yku80Δ double mutant as evidenced by a modest increase in more slowly migrating telomere species (S19 Fig). We also measured the TERRA levels in cdc73Δ, tel1Δ, and yku80Δ single and double mutant strains and found that these mutants did not significantly affect TERRA accumulation, except for an increase of the chrXV L TERRA in a yku80Δ single mutant (S17C Fig). To test if the effects of cdc73Δ might be due to the accumulation of R-loops, we introduced a plasmid bearing RNH1, which encodes S. cerevisiae RNase H1, into uGCR assay strains. In contrast to TLC1 overexpression, the RNH1 plasmid did not substantially affect the uGCR rate of either the cdc73Δ tel1Δ double mutant or the cdc73Δ yku80Δ double mutant (S5 Table). Taken together, these data suggest that the increased GCR rate caused by the cdc73Δ mutation may in part reflect an alteration in telomere structure caused by reduced telomerase activity due to reduced TLC1 levels. However, the synergistic increases in GCR rates seen in the cdc73Δ tel1Δ and cdc73Δ yku80Δ double mutants (and potentially the ctr9Δ and paf1Δ double mutants) is unlikely to be explained solely by reduced TLC1 levels as these double mutants have the same TLC1 levels as the cdc73Δ single mutant.
Cdc73, like other members of the Paf1 complex, has no known enzymatic activity [24]. The N-terminal region (S. cerevisiae residues 1–229) lacks identifiable domains; whereas the C-terminal domain (S. cerevisiae residues 230–393) has a conserved GTPase-like fold [41,80] and has been proposed on the basis of chemical crosslinking and cryo-electron microscopy to make direct interactions with the RNA polymerase II subunit Rpb3 [81]. We replaced the wild-type chromosomal copy of CDC73 with various CDC73 deletion mutations to gain insights into Cdc73 function (Fig 5A; S20 Fig; S6 Table). We found that deletion of the C-terminal domain (cdc73Δ230–393) resulted in wild-type dGCR rates, normal sensitivity to 6-azauracil and normal TPE. This result contrasts with a previous report suggesting that a cdc73Δ231-393-TAP construct causes increased sensitivity to 6-azauracil relative to wild-type CDC73 [41]; this difference may be due to the presence of the TAP tag in the previous study. On the other hand, deletion of the N-terminal region (cdc73Δ2–229), caused defects in all three assays that were similar to those caused by the cdc73Δ single mutation.
Additional analysis of CDC73 (Fig 5A) defined a minimal deletion, cdc73Δ125–229, that caused a similar fold-increase in the dGCR rate compared to that caused by the cdc73Δ single mutation (17.3-fold increase vs. 9.3-fold increase) and caused a synergistic increase in the dGCR rate when combined with the yku80Δ mutation that was similar to that observed with the cdc73Δ mutation (98.1-fold increase vs. 162-fold increase). This minimal deletion also caused increased sensitivity to 6-azauracil and reduced TPE (Fig 5A, S20A Fig) as well as reduced TLC1 levels (Fig 5B) and shorter telomeres (Fig 5C) similar to that caused by the cdc73Δ single mutation; as before, addition of sublethal concentrations of HU to distinguish TPE from 5FOA-induced overexpression of ribonucleotide reductase verified that the cdc73Δ125–229 mutation, like the cdc73Δ mutation, caused TPE defects (S20B Fig). As the effect of the cdc73Δ125–229 mutation could either have been due to loss of a functional region of Cdc73 or due to causing defects in folding Cdc73, we generated a gene construct that encoded only residues 125–229 (cdc73:125–229). This gene construct, which encoded 105 residues from the center of Cdc73, was sufficient to substantially restore Cdc73 functions in suppressing GCRs, maintenance of TLC1 levels, TPE, and telomere length homeostasis (Fig 5, S20 Fig). These results define a minimal functional Cdc73 construct, Cdc73:125–229, and a minimal non-functional Cdc73 construct, Cdc73Δ125–229.
Residues 125–229 of Cdc73 precede the C-terminal GTPase domain and lie in a region that is predicted to be less ordered by IUPRED [82] (S21A Fig) and that has reduced conservation (S21B Fig). Previous chemical crosslinking of the Paf1 complex bound to RNA polymerase II identified 22 crosslinks between Cdc73 and other Paf1 subunits of which 19 were between Cdc73 and Ctr9, which is primarily composed of tetratricopetide repeat (TPR) domains [81]. Analysis of these data also revealed that the Cdc73 region containing residues 125–229 had 9 crosslinks to Ctr9 (~50% of Cdc73-Ctr9 crosslinks), 2 crosslinks to Leo1, 2 crosslinks to Rpb11, and 1 crosslink to Rpb2 (S21C Fig). Together these data are consistent with the possibility that the TPR domains of Ctr9 bind to an unstructured Cdc73 peptide or Cdc73 alpha helices, rather than a folded Cdc73 domain, like other known TPR-peptide interactions [83]. To test for a direct Cdc73-Ctr9 interaction in the Paf1 complex, we tested the ability of Paf1 and Cdc73 to co-immunoprecipitate in a wild-type strain or strains with deletions of LEO1, RTF1, or CTR9 (S21D Fig). Consistent with this hypothesis, the Paf1-Cdc73 interaction was lost in the ctr9Δ strain, whereas deletions of LEO1 and RTF1 had only modest effects on the Paf1-Cdc73 interaction.
As the paf1Δ mutation causes increased dGCR rates similar to those caused by the cdc73Δ mutation, we sought to determine if the functional truncated Cdc73 proteins bound the Paf1 complex and if the defects associated with the minimal non-functional Cdc73Δ125–229 truncation were due to loss of Paf1 complex association or due to defects in other functions. We tested the ability of C-terminally Venus-tagged full-length Cdc73, Cdc73Δ230–393, Cdc73Δ2–124, Cdc73Δ125–229, or Cdc73:125–229 to co-immunopreciptate with C-terminally myc-tagged Paf1, Rtf1, Ctr9, or Leo1; all tagged proteins were expressed from the respective chromosomal loci. Cell lysates from doubly tagged strains were prepared from log-phase cells and immunoprecipitated with anti-myc antibodies, and then probed by Western blotting using anti-GFP antibodies. Full-length Cdc73 co-immunoprecipitated with Paf1, Rtf1, Ctr9, and Leo1 (Fig 6A), although the interaction with Rtf1 appeared to be weaker than the interaction with the other Paf1 complex subunits, consistent with previous observations [22,43,84]. The functional Cdc73 truncations, Cdc73Δ230–393, Cdc73Δ2–124, and Cdc73:125–229, all associated with Paf1, Ctr9, Leo1, and Rtf1 (Fig 6A). Reduced binding to Leo1 was observed with both the Cdc73Δ230–393 and Cdc73:125–229 truncations, suggesting that the C-terminus of Cdc73 may stabilize Leo1 in the complex. In contrast, the non-functional Cdc73 truncation, Cdc73Δ125–229, had substantially reduced binding to each of the other Paf1 complex subunits; a low level of residual binding was only detected with Ctr9 and Leo1. Thus residues 125–229 of Cdc73 appear to be necessary and sufficient for stable binding of Cdc73 to the Paf1 complex.
The Paf1 complex has been localized to the nucleus in wild-type cells by immunofluorescence [85], so we monitored the cellular localization of Cdc73 truncations. The wild-type and truncated forms of Cdc73 were C-terminally tagged with Venus, and functional versions of Cdc73, including the minimal construct Cdc73:125–229, localized to the nucleus (Fig 6B), with a high ratio of nuclear to cytoplasmic fluorescence (Fig 6C). In contrast, the non-functional Cdc73 truncation Cdc73Δ125–229, which did not stably associate with the Paf1 complex, had diffuse localization in both the nucleus and the cytoplasm, but was still expressed at normal levels based on total cellular fluorescence (Fig 6D). Thus, residues 125–229 of Cdc73 either include a nuclear localization signal or are necessary for binding to a Paf1 complex that is imported into the nucleus. Single mutant strains with deletions of PAF1, CTR9, RTF1, and LEO1 appeared to have normal nuclear localization of a Cdc73-Venus fusion protein (S22A Fig), although these mutations resulted in enlarged cells and abnormally elongated buds, as previously described [18,86]. Similarly, deletion of CDC73 did not prevent the nuclear localization of C-terminally Venus tagged Paf1, Rtf1, Ctr9, or Leo1 (S22B Fig), indicating that defects caused by the cdc73Δ mutation were not due to defects in the nuclear localization of other Paf1 complex subunits. Finally, the cdc73Δ mutation did not cause major changes in the cellular levels of the other Paf1 complex subunits as measured by Western blot (S22C Fig). These localization data are consistent with the observation that all Paf1 complex subunits other than Cdc73 are predicted to contain nuclear localization signals (S23 Fig); this is different from that seen with human Cdc73, which possesses a functional N-terminal nuclear localization signal [87]. Together, these data suggest that Cdc73 does not regulate the cellular localization of the Paf1 complex, but instead mediates the suppression of genome instability once the complex is already in the nucleus, potentially through contributions to overall complex stability or conformation.
Transcription, and defects in transcription including those that lead to the accumulation of R-loops, are becoming an increasingly well-appreciated source of genome instability [10,11]. Using a screen to identify genes that suppress the accumulation of GCRs, we found the loss of CDC73 results in increased rates of accumulating GCRs in three different GCR assays. We also found that a cdc73Δ mutation resulted in synergistic increases in GCR rates and in increased levels of telomere dysfunction when combined with either tel1Δ or yku80Δ mutations. This is reminiscent of the observation that tlc1Δ tel1Δ double mutants have synergistic increases in GCR rates relative to the respective single mutants, although they show delayed senescence and delayed loss of telomeres [88]; analysis of GCR rates and other telomere-related phenotypes in tlc1Δ yku80Δ double mutants was not possible as these double mutants cannot be propagated [59,89]. The fact that the cdc73Δ tel1Δ yku80Δ triple mutant appears to be highly defective in telomerase function and also shows a large synergistic increase in the rate of accumulating GCRs further suggests that telomere dysfunction is likely a hallmark of genome instability in cdc73Δ strains and that cdc73Δ, tel1Δ, and yku80Δ mutations all cause different defects that contribute to increased rates of accumulation of GCRs. A role for CDC73 in contributing to telomerase function is also consistent with our inability to observe GCRs formed by de novo telomere additions relative to the large increase in the levels of different translocation GCRs among the GCRs selected in the sGCR assay in cdc73Δ mutants. Consistent with the observation that cdc73 defects result in reduced levels of the TLC1 RNA component of telomerase [32], overexpression of TLC1 partially suppressed the increased GCR rate of the cdc73Δ tel1Δ double mutant. In contrast, over-expression of RNase H1, which degrades R-loops, did not suppress the increased GCR rate of the cdc73Δ tel1Δ double mutant.
The absence of telomerase in S. cerevisiae results in shortening of telomeres and reduced rates of cell growth until telomerase negative cells undergo crisis and survivors emerge in which telomeres are maintained by one of two different HR-mediated telomere maintenance pathways [61,63]. These surviving cells do not have increased rates of accumulating GCRs, although additional genetic defects can result in synergistic increases in GCR rates in these telomerase negative cells [55]. One exception where telomerase defects alone result in increased GCRs is telomerase negative cells that have been stabilized by re-expression of telomerase during crisis before the shortened telomeres have started to be maintained by HR [90]. In addition, tel1Δ mutations, which by themselves result in shortened telomeres and small increases in GCR rates, can result in large increases in GCR rates when combined with mec1 or other mutations [91]. Under all of these conditions, the telomeres with altered structures fuse to the ends of broken chromosomes, and the resulting fusion chromosomes then appear to undergo breakage and additional rearrangement events [92]; these altered telomeres can also undergo telomere to telomere fusion [93]. The structural analysis presented here showed that the cdc73Δ mutation that causes an increased GCR rate and the cdc73Δ tel1Δ and cdc73Δ yku80Δ double mutation combinations that cause synergistic increases in GCR rates did not appear to cause the accumulation of either de novo telomere addition-mediated GCRs or GCRs mediated by fusion of altered telomeres to broken chromosome ends. Telomerase activity is likely reduced but not absent in cdc73Δ mutants [32], which would explain the presence of telomeres that are shorter than normal and this is likely sufficient to result in modest increases in the rate of accumulating GCRs as well as the absence of de novo telomere addition-mediated GCRs. When a cdc73Δ mutation is combined with other mutations like tel1Δ and yku80Δ, which affect telomere maintenance in different ways and also cause shortened telomeres, there is an increased defect in telomere maintenance and an increased alteration of telomere chromatin structure as indicated by synergistic increases in TPE defects and a synergistic increase in the rate of accumulating GCRs. A hypothesis that explains the increased GCR rates and the spectrum of GCRs observed is that in these mutants reduced telomere maintenance combined with alterations in telomere chromatin structure results in a fraction of chromosome ends that escape protection and undergo extensive degradation (Fig 7). These degraded chromosome ends can then be processed by end joining to other DSBs, hairpin formation or short sequence-mediated HR resulting in the GCRs selected in the sGCR assay or longer sequence non-allelic HR resulting in the translocations selected in the dGCR assay. This mechanism also accounts for the increased GCR rates seen in paf1Δ and ctr9Δ single and double mutants analyzed as these mutations also cause telomere maintenance and telomere chromatin structure defects as evidenced by reduced TLC1 levels, short telomeres and TPE defects [32]. The lack of or limited increased GCR rates seen in rtf1Δ and leo1Δ single and double mutants is also accounted for by this mechanism as these latter mutations have smaller effects on TLC1 levels and telomere shortening [32], and in the case of leo1Δ mutations, no defect in TPE reflective of alterations in telomere chromatin structure.
The Paf1 complex promotes transcription elongation, 3’-end mRNA maturation, and histone modification [16,21–24]. Our results demonstrate that different subunits of the Paf1 complex subunits promote different Paf1 complex functions: (1) suppression of GCRs primarily requires Paf1, Ctr9, and Cdc73; (2) resistance to 6-azauracil inhibition of transcriptional elongation primarily requires Paf1 and Ctr9; and (3) silencing of telomere-proximal genes requires Cdc73, Paf1, Ctr9 and Rtf1 to differing degrees. The rather disparate effects of deleting genes encoding different Paf1 complex subunits observed here mirrors previous observations of different requirements for individual Paf1 complex subunits under different stress conditions [37]. Using an assay that detected chromosome loss and GCRs but did not distinguish between the two, a previous study showed that deletions of CDC73 and LEO1, but not PAF1, resulted in increased genome instability that could be suppressed by increased expression of RNase H1 [12]. The relatively important role of Paf1 in all of the Paf1 complex functions (our results as well as in previous studies [37]) is consistent with the idea that Paf1 functions by mediating recruitment of the other Paf1 subunits to the different processes they functions in. Alternatively, Paf1 may provide the major function of the Paf1 complex and may be recruited to different processes by different Paf1 complex subunits: Leo1 and Rtf1 bind RNA [94]; Rtf1 binds phosphorylated Spt5, which is a component of TFIIS and binds the elongating RNA polymerase II complex [95,96]; and the Cdc73 C-terminal domain mediates binding to the phosphorylated C-terminal domain (CTD) of RNA polymerase II [97].
The importance of the interaction of Cdc73 with Paf1 is demonstrated by the deletion analysis of Cdc73. The C-terminal domain and the N-terminal regions of Cdc73 were found to be dispensable for CDC73 function. However, the central 105 amino acid region (residues 125–229) was necessary and sufficient to: (1) suppress the defects of cdc73Δ strains studied here; (2) mediate incorporation into the Paf1 complex; and (3) promote nuclear localization of Cdc73. Remarkably, the C-terminal region, which binds the phosphorylated RNA polymerase II CTD [97] and contributes to suppression of Ty element expression [41], was not required for any of the functions analyzed here. The dispensable nature of the Cdc73 C-terminal domain could be consistent with the redundancy of recruitment of the Paf1 complex to RNA polymerase II by either Cdc73 binding to the phosphorylated RNA polymerase II CTD or by Rtf1 binding to phosphorylated Spt5 [97]. This redundancy also explains the synergistic defect in 6-azauracil sensitivity caused by combining a deletion of the C-terminal domain of Cdc73 with loss of Rtf1 [41]. Moreover, the available data suggest that the central 105 amino acid region (residues 125–229) of Cdc73 plays some previously unappreciated function in the Paf1 complex. Extensive chemical crosslinking between this region of Cdc73 and the TPR domain containing protein Ctr9 [81] and the requirement for Ctr9 for coimmunoprecipitation of Cdc73 with Paf1 suggest that the Ctr9-Cdc73 interaction recruits Cdc73 to the Paf1 complex. The fact we were unable to computationally predict a function-associated motif or domain structure within the central 105 amino acid region and that the N-terminus of S. cerevisiae Cdc73 up to residue 236 is highly sensitive to partial proteolysis [41] suggests the central 105 amino acid region of Cdc73 is likely unstructured in the absence of the Paf1 complex. This is consistent with the role of TPR domains in binding alpha-helices and unstructured peptides [83]. Together these data are also consistent with the fact that ctr9Δ mutations, like cdc73Δ mutations, also cause increased GCR rates, cause synergistic increases in GCR rates when combined with yku80Δ and tel1Δ mutations, have TPE defects, and defects in TLC1 expression [32].
Mutations in human CDC73 (also called HRPT2) identified in cases of sporadic and hereditary parathyroid carcinomas appear to primarily be loss-of-function mutations including frameshifts, premature stop codons, and deletions that result in truncated proteins. In many cases, the heterozygous germline mutations observed are associated with events leading to loss-of-heterozygosity in tumors; however, some tumors appear to have amplification of the mutant copy of CDC73, suggesting a dominant genetic phenotype [28–30,98–100]. The region of human Cdc73 (also called parafibromin) responsible for Paf1 complex binding [29] is in a region that is similar to the central 105 amino acid region in S. cerevisiae Cdc73 identified here, and at least some mutant versions of human Cdc73 seen in parathyroid carcinomas have lost their ability to interact with the Paf1 complex [101]. The 3 mutations in CTR9 found in Wilms tumor families comprised a nonsense mutation and 2 splice site mutations, all of which were consistent with causing loss-of-function [31]. Our results suggest that the CDC73 mutations seen in sporadic and hereditary parathyroid carcinomas and the CTR9 mutations found in Wilms tumor could cause increased genome instability; however, it is not currently known if these defects in human CDC73 and CTR9 cause genome instability and telomere dysfunction in human cells as observed here for the S. cerevisiae cdc73Δ mutation. Given the ability of the Paf1 complex to affect transcriptional elongation, RNA 5’ end maturation, and histone modification, inherited and sporadic CDC73 mutations and inherited CTR9 mutations in human cancers could have pleiotropic effects in which increased genome instability might not play the only role in carcinogenesis.
All S. cerevisiae strains used in this study were derived from S288c and were constructed by standard PCR-based gene disruption methods or by mating to strains containing mutations of interest (S7 Table; [102,103]). GCR assays were performed using derivatives of RDKY7635 (dGCR assay), RDKY7964 (sGCR assay), and RDKY6677, (uGCR assay) (S7 Table; [6,51]). The Venus, mCherry and 9myc tags were amplified from pBS7, pBS35, and pYM19, respectively [102,104], inserted at the 3’ end of the indicated genes using standard methods. For determining GCR rates of strains transformed with the RNase H1 tet-off overexpression plasmid pCM184RNH1 (a gift from Andrés Aguilera, Universidad de Sevilla, Seville, Spain [105]) or the ADH1 promoter TLC1 overexpression plasmid pVL2679 (a gift from Victoria Lundblad, Salk Institute), transformants were cultured overnight in complete synthetic medium (CSM)–Trp liquid media and plated onto either CSM–Trp medium or CSM–Arg–Trp medium supplemented with 1 g/L 5FOA and 60 mg/L canavanine. To test for transcription elongation defects, 6-azauracil (Sigma-Aldrich) was added to synthetic complete medium at a final concentration of 50 μg/ml.
CDC73, including 998 bp upstream and 536 bp downstream, was amplified by PCR using the primers 5’-CAC CGA ATT GCA AGC GCT TGC AAC TTG TTC TTT CTG TGC -3’ and 5’-GAA TTG CAA GCG CTC CCA TGG AAA TGA GAG AAG C-3’ (AfeI cut site underlined) and cloned into the pENTR/D-TOPO vector (Thermo Fisher Scientific) to generate pRDK1705. The hygromycin B resistance gene was amplified from the plasmid pFA6a-hphNT1 with the primers 5’-GAA TTG CAA AGC TTC GGA TCC CCG GGT TAA TTA A-3’ and 5’-GAA TTG CAA AGC TTT AGG GAG ACC GGC AGA TCC G-3’ (HindIII cut site underlined) and inserted into pRDK1705 at a HindIII cut site located 693 bp upstream of the CDC73 start codon to make plasmid pRDK1706. The cdc73 alleles were made in pRDK1706 using the GeneArt Site-Directed Mutagenesis kit (Life Technologies) to generate pRDK1708 (cdc73Δ230–393), pRDK1770 (cdc73Δ2–229), pRDK1771 (cdc73Δ92–229), pRDK1772 (cdc73Δ2–91), pRDK1781 (cdc73Δ92–147), pRDK1782 (cdc73Δ148–229), pRDK1784 (cdc73:92–229), pRDK1788 (cdc73Δ2–124), pRDK1789 (cdc73Δ125–229), and pRDK1790 (cdc73:125–229). These plasmids were integrated at the endogenous CDC73 locus by transformation with AfeI-digested plasmid DNA. Integrants were confirmed by PCR and Sanger sequencing.
We crossed a strain containing the dGCR assay and a cdc73Δ or rtf1Δ mutation against 638 strains from the S. cerevisiae deletion collection and obtained haploid progeny by germinating spores generated from the resulting diploids, as previously described [6].
Systematically generated cdc73Δ double mutants and control haploid and diploid strains were screened by flow cytometry for DNA content to exclude diploid isolates. Briefly, 10 μL aliquots of overnight cultures grown in YPD were added to 190 μL of fresh YPD, and the cells were incubated in a 30°C shaker for 3 hours. Cells were washed, resuspended in 60 μL of dH2O, and fixed with 140 μL of cold absolute ethanol. Fixed cells were sonicated and resuspended in 150 μL of 50 mM sodium citrate with 1 mg/mL Proteinase K (Sigma-Aldrich) and 0.25 mg/mL RNase A (Sigma-Aldrich) and incubated at 37°C overnight. Treated cells were washed, resuspended in 100 μL of 50 mM sodium citrate containing 1 μM Sytox Green (Life Technologies), and analyzed using a BDS LSR II flow cytometer at The Scripps Research Institute flow cytometry core facility. Data were analyzed using FlowJo v10 [106].
Patch tests for identifying systematically generated double mutants with increased GCR rates were performed as described [6]. GCR rates were determined using at least 14 independent cultures from 2 independent biological isolates of each strain using the fluctuation method as previously described [107]. Significantly different GCR rates were identified through analysis of the 95% confidence intervals.
The t(V;XIV) and t(V;IV or X) homology-mediated translocation GCRs were identified by PCR, as previously described [51].
Multiplexed paired-end libraries were constructed from 5 μg of genomic DNA purified using the Purgene kit (Qiagen). The genomic DNA was sheared by sonication and end-repaired using the End-it DNA End-repair kit (Epicentre Technologies). Common adaptors from the Multiplexing Sample Preparation Oligo Kit (Illumina) were then ligated to the genomic DNA fragments, and the fragments were then subjected to 18 cycles of amplification using the Library Amplification Readymix (KAPA Biosystems). The amplified products were fractionated on an agarose gel to select 600 bp fragments, which were subsequently sequenced on an Illumina HiSeq 2000 using the Illumina GAII sequencing procedure for paired-end short read sequencing. Reads from each read pair were mapped separately by bowtie version 2.2.1 [108] to a reference sequence that contained revision 64 of the S. cerevisiae S288c genome [109], hisG from Samonella enterica, and the kanMX4 marker (S3 Table). Reads are available from National Center for Biotechnology Information Sequence Read Archive under accession number: SRP107803.
GCR structures were determined using mapped reads using version 0.6 of the Pyrus suite (http://www.sourceforge.net/p/pyrus-seq) [52]. Rearrangements relative to the reference S288c genome were identified by analyzing the read depth distributions (S5–S8 Figs), the discordantly mapping read pairs (S2–S4 Figs; S4 Table), and/or extracting the sequences of the novel junctions (S9–S13 Figs). Associated junction-sequencing reads, which were reads that did not map to the reference but were in read pairs in which one end was adjacent to discordant reads defining a junction, were used to sequence novel junctions. Most hairpin-generated junctions (S12 Fig) could be determined using alignments of junction-sequencing reads. For junctions formed by HR between short repetitive elements (S9–S11 Figs) and for problematic hairpin-generated junctions (S12 Fig), the junction sequence could be derived by alignment of all reads in read pairs where one read was present in an “anchor” region adjacent to the junction of interest and the other read fell within the junction to be sequenced.
Junctions indicated by copy number changes, discordant read pairs, and junction sequencing were identified with a high degree of confidence; however, previous analyses have indicated that even junctions inferred from only copy number changes can be experimentally verified at high frequency [52,92,110,111]. Analysis of the sequencing data identified all of the genetic modifications introduced during construction of the starting strains, such as the his3Δ200 deletion, (S2–S4 Figs) as well as the molecular features associated with the selected GCRs (S5–S13 Figs; S4 Table). Several inverted duplications (isolates 307, 324, and 331) with a YCLWdelta5/YELWdelta1 junction copied very little sequence in the vicinity of YELWdelta1, and had an additional HR-mediated translocation between YELWdelta1 and an unannotated delta sequence on chrV R, which we term here “YERWdelta27” (S14 Fig).
Telomere Southern blots were performed using a modified version of a previously described protocol [112]. Genomic DNA was purified from 50 mL overnight cultures using the Purgene kit (Qiagen). 5 μg of DNA was digested with XhoI (New England Biolabs) in a 50 μL reaction for 2 hr at 37°C. The reaction was stopped by adding 8 μL of loading buffer, and the samples were run on a 0.8% agarose gel in 0.5X TBE for 16 hr at 50 V. The DNA in the gel was transferred to Amersham Hybond-XL membranes (GE) by neutral capillary blotting, allowed to run overnight. The DNA was crosslinked to the membrane by UV irradiation in a Stratalinker (Stratagene) apparatus at maximum output for 60 seconds. Biotinylated TG probes were purchased from ValueGene. Probe hybridization was performed with ULTAhyb oligo hybridization buffer (Life Technologies) at 42°C for 1 hr. The membrane was then washed extensively and developed with a chemiluminescent nucleic acid detection kit (Life Technologies) and imaged with a Bio-Rad Imager.
DNA plugs for PFGE were prepared as described [113]. Strains were grown to saturation in 50 mL of YPD at 30°C for 3 days. Cell counts were measured by optical density at 600 nm, and 7.5 x 108 cells from each strain were washed and resuspended in 200 μL of 50 mM EDTA, then mixed with 70 μL of 1 M sorbitol, 1 mM EDTA, 100 mM sodium citrate, 0.5% β-mercaptoethanol, 8 U/mL of zymolase. The cells were then mixed with 330 μL of liquefied 1% ultrapure agarose (Bio-Rad) to prepare multiple 80 μL plugs. The plugs were incubated in 15 mL conical tubes in 750 μL of 10 mM Tris pH 7.5, 500 mM EDTA pH 8, 1% β-mercaptoethanol for 16 hr at 37°C. The plugs were then incubated in 750 μL 10 mM Tris pH 7.5, 500 mM EDTA pH 8, 1% sodium N-lauryl sarcosine, 0.2% sodium dodecyl sulfate containing 2 mg/ml Proteinase K (Sigma-Aldrich) for 6 hr at 65°C. Finally, the plugs were washed in 50 mM EDTA pH 8 prior to resolving the chromosomes in a 1% agarose gel run in a CHEF (clamped homogeneous electric field electrophoresis) apparatus in chilled (14°C) 0.5x TBE (89 mM Tris-borate, pH 8.3, 25 mM EDTA). Electrophoresis was performed using a Bio-Rad CHEF-DRII apparatus at 6 V/cm, with a 60 to 120 s switch time for 24 h. The gels were stained with ethidium bromide and imaged.
The TPE assay was constructed by transforming BY4742 (MATalpha leu2Δ0 his3Δ1 ura3Δ0 lys2Δ0) with pADH4UCA ([38], a gift from Virginia Zakian, Princeton University) digested with SalI and EcoRI. Integration of URA3 into ADH4, which was verified by PCR, generated the strain RDKY8230, and mutant derivatives were constructed by PCR-mediated gene disruption (S7 Table). TPE was assayed by culturing strains overnight in YPD at 30°C followed by spotting 1.5 μL of 10-fold serial dilutions onto CSM, and CSM supplemented with 1 g/L of 5FOA (CSM+5FOA). Plates were incubated at 30°C for 3 days before imaging. In some experiments, the plates also contained either 10 mM or 30 mM HU [42].
RNA isolation and qRT-PCR for TLC1 and TERRA RNA levels were performed using published techniques [114,115]. Cells were grown in YPD to an OD600 of 0.6 to 0.8. 1 mL samples were used for RNA isolation with the RNeasy kit (Qiagen), with on-column DNase I treatment using the RNase-Free DNase Set (Qiagen). 1 μg RNA was reverse transcribed with the iScript cDNA Synthesis Kit (Bio-Rad), which uses random primers. cDNA was diluted 1:10 with distilled H2O. qPCR was performed with 2 μL of the dilution in a final volume of 20 μL using the iTaq Universal SYBR Green Supermix (Bio-Rad) in a Bio-Rad CFX96 Touch Real-Time PCR Detection System. Reaction conditions: 95°C for 10 min, 95°C for 15 sec, 50°C for 1 min, 40 cycles. Primer concentrations and sequences were the same as previously described [115].
The μMACS anti-c-myc magnetic bead IP kit (Miltenyi Biotec) was used in immunoprecipitation experiments. Lysates were generated from strains in which one or two Paf1 complex genes in the S. cerevisiae strain BY4741 (MATa leu2Δ0 his3Δ1 ura3Δ0 met15Δ0) were tagged with Venus or c-myc. Strains were grown to mid-log phase in 50 mL YPD, harvested, resuspended in 1 mL of the supplied lysis buffer, and incubated on ice for 30 minutes. Cells were lysed with the addition of 100 μL of glass beads and vortexed four times for 1 minute with cooling. Lysates were clarified at 14,000 rpm for 10 minutes at 4°C. Protein concentrations were determined using the DC Protein Assay (Bio-Rad). For the input analysis, 500 μg of protein was trichloroacetic acid (TCA) precipitated, resuspended in 100 μL of 2x SDS gel loading buffer (100 mM Tris-Cl (pH 6.8), 4% SDS, 20% glycerol, 200 mM DTT, 0.2% bromophenol blue) and 10 μL was used for Western Blotting. For the immunoprecipitation, 1000 μg of protein was incubated with 50 μL anti-c-myc MicroBeads (Miltenyi Biotec) for 30 minutes on ice, then passed through the μMACS separator column. The column was washed twice with 200 μL of lysis buffer, washed twice with 200 μL of wash buffer 1, then washed once with 100 μL of wash buffer 2. The column was then incubated with 20 μL of heated elution buffer for 5 minutes, before the proteins were eluted with 50 μL of heated elution buffer. Of the eluted volume, 12 μL was used for Western Blotting.
Proteins were resolved on a 4–15% SDS-PAGE gel (Bio-Rad) and transferred overnight onto nitrocellulose membrane (Bio-Rad). Venus-tagged proteins were detected with the rabbit monoclonal antibody ab290 (Abcam, 1:2000) and myc-tagged proteins were detected with 71D10 rabbit monoclonal antibody (Cell Signaling, 1:1000). Horseradish peroxidase-conjugated goat anti-rabbit secondary antibody (Jackson Laboratories, 1:5000) was used, followed by chemiluminescence detection with SuperSignal Femto Sensitivity Substrate (Life Technologies) and imaged with a Bio-Rad Imager. Venus-tagged protein levels were also detected using mouse monoclonal antibody B34 (Covance, 1:1000) and mouse monoclonal anti-Pgk1 antibody (ab113687, Abcam, 1:5000) was used to detect Pgk1 as a loading control.
Exponentially growing cultures were washed and resuspended in water before being placed on minimal media agar pads, covered with a coverslip, and sealed with valap (a 1:1:1 mixture of Vaseline, lanolin, and paraffin by weight). Cells were imaged on a Deltavision (Applied Precision) microscope with an Olympus 100X 1.35NA objective. Fourteen 0.5 μm z sections were acquired and deconvolved with softWoRx software. Further image processing, including intensity measurements were performed using ImageJ. Intensity levels were quantified by taking the mean intensity in the nucleus, the cytoplasm, and a background measurement outside of the cell using a 3-pixel diameter circle. The ratio of background-subtracted nuclear fluorescence to background-subtracted cytoplasmic fluorescence was then calculated per cell. The total fluorescence was estimated by taking the background-subtracted nuclear fluorescence and adding it to 12.5 times the background-subtracted cytoplasmic fluorescence as an approximation of the ratio cytoplasmic to nuclear volume.
|
10.1371/journal.ppat.1005879 | Multi-dose Romidepsin Reactivates Replication Competent SIV in Post-antiretroviral Rhesus Macaque Controllers | Viruses that persist despite seemingly effective antiretroviral treatment (ART) and can reinitiate infection if treatment is stopped preclude definitive treatment of HIV-1 infected individuals, requiring lifelong ART. Among strategies proposed for targeting these viral reservoirs, the premise of the “shock and kill” strategy is to induce expression of latent proviruses [for example with histone deacetylase inhibitors (HDACis)] resulting in elimination of the affected cells through viral cytolysis or immune clearance mechanisms. Yet, ex vivo studies reported that HDACis have variable efficacy for reactivating latent proviruses, and hinder immune functions. We developed a nonhuman primate model of post-treatment control of SIV through early and prolonged administration of ART and performed in vivo reactivation experiments in controller RMs, evaluating the ability of the HDACi romidepsin (RMD) to reactivate SIV and the impact of RMD treatment on SIV-specific T cell responses. Ten RMs were IV-infected with a SIVsmmFTq transmitted-founder infectious molecular clone. Four RMs received conventional ART for >9 months, starting from 65 days post-infection. SIVsmmFTq plasma viremia was robustly controlled to <10 SIV RNA copies/mL with ART, without viral blips. At ART cessation, initial rebound viremia to ~106 copies/mL was followed by a decline to < 10 copies/mL, suggesting effective immune control. Three post-treatment controller RMs received three doses of RMD every 35–50 days, followed by in vivo experimental depletion of CD8+ cells using monoclonal antibody M-T807R1. RMD was well-tolerated and resulted in a rapid and massive surge in T cell activation, as well as significant virus rebounds (~104 copies/ml) peaking at 5–12 days post-treatment. CD8+ cell depletion resulted in a more robust viral rebound (107 copies/ml) that was controlled upon CD8+ T cell recovery. Our results show that RMD can reactivate SIV in vivo in the setting of post-ART viral control. Comparison of the patterns of virus rebound after RMD administration and CD8+ cell depletion suggested that RMD impact on T cells is only transient and does not irreversibly alter the ability of SIV-specific T cells to control the reactivated virus.
| Antiretroviral therapy (ART) does not eradicate HIV-1 in infected individuals due to virus persistence in latently infected reservoir cells, despite apparently effective ART. The persistent virus and can rekindle infection when ART is interrupted. The goal of the “shock and kill” viral clearance strategy is to induce expression of latent proviruses and eliminate the infected cells through viral cytolysis or immune clearance mechanisms. Latency reversing agents (LRAs) tested to date have been reported to have variable effects, both on virus reactivation and on immune functions. We performed in vivo reactivation experiments in SIV-infected RMs that controlled viral replication after a period of ART to evaluate the ability of the histone deacetylase inhibitor romidepsin (RMD) to reactivate SIV and its impact on SIV-specific immune responses. Our results suggest that RMD treatment can increase virus expression in this setting, and that it does not markedly or durably impair the ability of SIV-specific T cells to control viral replication.
| Viral reservoirs are infected cells that persist even in the face of seemingly effective suppressive antiretroviral therapy (ART) and can give rise to recrudescent infection when ART is stopped. Reservoir cells include latently infected resting, memory CD4+ T cells, as well as other cells, such as T memory stem cells (TSCM) or T follicular helper cells (Tfh) [1–8]. Cells harboring latent proviruses carry the virus for the duration of their lifespan. As the half-life of central memory T helper cells is estimated at 44 months [9], and even longer for the TSCM and Tfh [6,10,11], and latently infected cells that do not express viral antigens are invisible to immune clearance mechanisms, such cells can persist for decades, even in patients successfully treated with ART [12–16]. Upon stochastic reactivation, perhaps in connection with homeostatic proliferation or antigen specific stimulation, these quiescent cells can revert their status and start producing new virions [5,17]. Even if expression of viral antigens results in immune clearance, the virus will persist as long as proliferation equals or exceeds clearance. ART may suppress most de novo infections of susceptible cells by virions derived from reactivated cells, but viral rebound occurs after variable delays at the cessation of ART, with plasma viral load (PVLs) typically rebounding to pretreatment levels [18,19]. With an estimated 1 latently infected cell per 1x106 CD4+ T cells [20], current paradigms predict that the latent viral reservoir is unlikely to be naturally eliminated over the lifetime of an HIV-infected individual on ART [21,22].
With the reports of the Berlin patient, the Boston patients and the Mississippi baby, there has been renewed interest in the prospect of achieving viral eradication, or at least sufficient reduction of the reservoir to allow extended viral remission in the absence of continuous ART, both considered as acceptable “HIV cure” strategies [23–26]. Current cure approaches include ART intensification studies [27–31], infusion of CCR5-gene modified CD4+ T cells following chemotherapy [24,32], enhancement of host HIV-specific immune responses to remove reactivated cells, and variations on the “shock and kill” approach [33–40]. The “shock and kill” strategy has seen the most emphasis. This approach consists of the administration of latency reversing agents (LRAs) to induce expression of latent proviruses, with the cells in which virus expression is induced being eliminated by viral cytopathic effects or host immune responses. De novo infection of susceptible cells by LRA-induced virus is prevented by ongoing ART [41,42]. This strategy represents a logical approach that theoretically should eventually result in the curbing/elimination of the reservoir. Several types of LRAs have been tested and have shown at least a limited success in activating the latent reservoir, including histone deacetylase inhibitors (HDACis) [38,43,44], protein kinase C (PKC) agonists, such as bryostatin-1 and prostratin [35,45] and the bromodomain-containing protein 4 inhibitor JQ1 [34,36]. Of these, HDACis have shown the most promise in reactivating the virus from the reservoirs. Valproic acid, givinostat, entinostat, vorinostat (suberoylanilide hydroxamic acid, SAHA), panobinostat and romidepsin (RMD) are the most studied HDACis; their effects on reactivating the virus from the reservoir are variable [38,42,46–48]. RMD has been shown to be among the most potent inducers of HIV in both in vitro and ex vivo models [38]. Recent studies have also documented modest activity of RMD to increase plasma viremia in both HIV-infected patients [49] and SIV-infected macaques [50] on ART. However, RMD has also been reported in vitro to inhibit host CTLs that target infected cells [51].
Elimination of reactivated resting CD4+ T cells in which expression of latent proviruses is induced by LRAs is proving to be more challenging than originally proposed [52]. Initially, it was believed that LRAs capable of potent and extensive induction of latent proviruses could be identified and that upon reactivation, infected cells would be killed by cytopathic effect (CPE) and/or host immune responses. Though logical, experience to date has not fulfilled these expectations. Specifically, studies have demonstrated that: only a small fraction of proviruses present in resting CD4+ T cells are reactivated by a single round of HDACi treatment in vitro [53]; reactivation of infected, resting CD4+ T cells by vorinostat did not result in viral CPE-mediated cell death [42]; impairment of host HIV-specific CTLs’ functions is specifically associated with immunodeficiency characteristic to HIV infection even in treated patients [54]; viral clearance by the CTLs only occurs in elite controllers [55]; the effect of various LRAs on latently infected cells, both in vivo and ex vivo, is highly variable [45,56,57]; in vitro treatment of cells with HDACi can impair CTL function [51]; and, finally, much of the virus present in latently infected cells in individuals that started ART in the chronic phase of infection contains escape mutations for immunodominant epitopes [58].
Therefore, research is currently refocusing on improving the efficiency of viral induction, for example through combinatorial approaches with LRAs targeting different mechanisms that contribute to the establishment and maintenance of viral latency, and boosting the killing of cells expressing induced virus through enhancing host immune responses, particularly the CD8+ CTL responses, by way of therapeutic vaccines, monoclonal antibodies and immune checkpoint inhibitors [39,59–62]. Note, however, that complex combinatorial strategies, in which LRAs are combined with immunomodulators, may add difficulty to data interpretation (as was the case with the Berlin patient [63]).
In this context, to evaluate different strategies for targeting viral reservoirs, animal models are needed that reproduce key features of HIV infection, while representing experimentally tractable systems that allow a mechanistic evaluation of different interventions. In addition to permitting testing of strategies for which lack of proof of concept or safety concerns may preclude clinical evaluation, animal models, particularly NHP models, offer the opportunity for extensive tissue sampling, to assess what is, after all, a tissue-based disease [64,65]. Yet, current animal models have their own limitations: (i) For years, SIVmac infection in RMs was more difficult to control than HIV-1 infection, requiring the use of complex and expensive ART regimens [50,66–69], and only recently, a simplified coformulated regimen was reported to effectively control SIVmac infection in RMs [70,71]; (ii) Humanized mice models, though suitable for addressing some questions relevant to cure research [65,72–74], do not allow detailed assessment of potential virus reservoirs due to limitations on samples obtainable from individual animals and on the feasible durations of treatment.
The “Visconti cohort” is a group of French HIV-infected individuals in which prolonged ART was initiated early in infection but eventually halted, and in which a fraction of patients managed to control virus rebound in the absence of continued ART, despite the lack of protective MHC alleles or other known factors that might lead to such an outcome [75]. We developed a NHP model to replicate post-ART control of viral replication in RMs infected with an infectious molecular clone (SIVsmmFTq). This model permits characterization of both the host factors associated with post-treatment control of viral replication and of the dynamics of the viral reservoir in post-treatment controllers. Furthermore, our model can be used to test virus reactivation strategies in the presence of apparently effective immune responses (in a “shock and effective kill” approach). Thus, our model permits the study of LRA efficacy on viral induction without confounding factors such as ART and immunotherapy. We used this model to assess the ability of the HDACi RMD to reactivate virus. We report that RMD can effectively increase virus expression in this model of post-treatment viral control and that RMD administration did not induce a marked or durable alteration of the cellular immune responses in vivo.
To establish a macaque model of post-ART control of virus for studies of approaches to target viral reservoirs that persist in this setting, we identified a SIVsmm strain (FTq) [76] that replicates in RMs at levels that are similar to those observed in chronically HIV-1-infected patients. We developed a transmitted/founder (TF) infectious molecular clone (IMC) of SIVsmmFTq, using the methodology reported in Gnanadurai et al. [77] and used this new IMC to intravenously infect ten Indian RMs. Six RMs were used as controls, in which SIVsmmFTq infection followed its natural course in the absence of any therapeutic intervention, while the remaining RMs served as a study group and received ART, followed by virus reactivation with RMD, as illustrated in Fig 1.
During >1 year follow-up, we show that SIVsmmFTq closely reproduced the patterns of virus replication observed in HIV-1 infection, with high PVL peaks [107−108 viral RNA (vRNA) copies/mL] and a robust, but relatively controlled replication during chronic infection (104−105 copies/mL) (Fig 2a). Furthermore, this robust virus replication resulted in a significant depletion of peripheral CD4+ T cells during the acute SIVsmmFTq infection and a partial CD4+ T cell restoration during chronic infection (Fig 2b).
The set-point levels of chronic SIVsmmFTq replication being in the range of HIV-1 infection, and lower than those observed with SIVmac239 [78], we reasoned that, similar to HIV-1, SIVsmmFTq can be readily controlled with ART. Four RMs intravenously infected with SIVsmmFTq received a combination of tenofovir (PMPA), emtricitabine (FTC) and integrase inhibitor (L-870812) for over nine months, starting at 65 days postinfection (dpi). ART resulted in a multiphased decay of plasma viremia with PVLs decreasing to <10 copies/ml in all RMs receiving ART by 30 days on ART, at 95 dpi (Fig 2a). During the follow-up, our priority was to assess the robustness of viral control and assess whether or not viral blips occurred under ART. To this end, RMs on ART were sampled every three days. This very frequent sampling schedule limited the amount of plasma available for viral quantification, preventing us from lowering the detection limit below 10 copies/mL. However, at this detection limit, no detectable vRNA blips were recorded in the plasma over the following 8 months of treatment in any of the RMs receiving ART, confirming our hypothesis that the administered ART regimen successfully controlled SIVsmmFTq replication in these RMs.
During treatment, one of the RMs on ART (RM177) died of unrelated conditions (complications of anesthesia), at 112 dpi (52 dpt). At the time of death, plasma viremia was < 10 copies/mL in this monkey (Fig 2a).
The robust control of viral replication in RMs receiving ART impacted the recovery of the CD4+ T cells, which, in spite of being similarly depleted from circulation in both groups of SIVsmmFTq-infected RMs during acute infection, recovered to nearly preinfection levels in the RMs treated with ART (9 months of ART, with 8 months of plasma viremia < 10 copies/mL). Comparatively, the control RMs showed a more limited restoration at the end of the follow-up (Fig 2b).
Furthermore, the fractions of CD4+ and CD8+ T cells expressing immune activation markers were lower in RMs receiving ART compared to controls (Fig 2c). Note, however, that, similar to HIV-infected patients on ART [79,80], and other models of ART-treated SIV-infected RMs [66], a low level of residual immune activation persisted during antiretroviral therapy in SIVsmmFTq-infected RMs, despite of a robust viral control with ART. These features of the SIVsmmFTq-infected RM model more closely recapitulate key aspects of HIV infection of humans compared to the highly pathogenic SIVmac239 infection.
After demonstrating that conventional ART can robustly control SIVsmmFTq replication, we next attempted to reactivate the virus from the reservoir through administration of RMD. One dose of 7 mg/m2 of RMD was administered to the three RMs on ART in a slow perfusion over four hours. Blood samples were obtained during and after RMD treatment to assess the pharmacological effects of the RMD, as well as virus reactivation.
We first monitored RMD activity by measuring acetylated histone (H3 and H4) levels in both CD4+ and CD8+ T cells [66]. Histone acetylation increased during the RMD treatment peaked at 6 hours post-RMD treatment initiation and returned to nearly pretreatment levels by 5 days post treatment (dpt), confirming that we had delivered a bioactive dose of the drug (Fig 3). However, in spite of the documented increase of the levels of acetylated histones, we did not observe any measurable increase in plasma viremia after RMD administration to RMs on ART (Fig 4).
Therefore, one week after completion of RMD treatment, ART was stopped in all RMs. Cessation of ART was followed by rapid and robust rebound of plasma viremia in all three RMs. Viral rebound: (i) occurred very rapidly, with SIVsmmFTq being detected in plasma only three days after ART cessation (Fig 4); (ii) was higher than expected, reaching peak levels of 105−107 copies/ml (Fig 4), higher than the set-point PVL established during chronic infection, which is often the level of virus rebound observed at the cessation of ART [81]; and (iii) was controlled to < 30 copies/ml (below the limit of detection of our conventional assay) within 50 days after discontinuation of ART (Fig 4). We continued to closely monitor PVLs in these RMs using a more sensitive assay and observed that PVLs fluctuated between ≤10 copies/ml and 30 copies/ml, but no animal lost control of the virus over 150 days of observation. Based on the characteristics of the post-treatment dynamics of viremia, these three RMs were labelled as post-treatment controllers [75] (Fig 4).
Due to the unexpected characteristics of the PVL rebound leading to eventual post-treatment control of infection, we designed a new strategy to test whether or not the transient “excess of viral rebound” that followed ART interruption can be attributed to RMD administration. To assess the ability of RMD to induce viral expression in the setting of post-ART spontaneous viral control, we administered three doses of RMD at 35–50 day intervals to the three post-treatment controller RMs.
After each treatment, RMD had detectable in vivo activity, as illustrated by increased levels of acetylated histones which peaked at 6 hours postadministration, and returned to nearly pretreatment levels by 5 dpt, as illustrated in Fig 3. Notably, repeated RMD administration did not result in changes in the levels of acetylated histones, which would have suggested tolerance. As this clear impact on histone acetylation was consistently observed after each RMD administration, we next assessed the ability of RMD to induce increased viral expression.
Virus reactivation was monitored by measuring the levels of vRNA in plasma with a single copy PCR assay (SCA) specifically developed for SIVsmmFTQ, similar to other previously described assays [82,83]. While no detectable rebound of PVLs could be documented in the plasma samples collected at 4, 6, 24 and 48 hours post-RMD administration, detectable PVLs were observed starting from 5 dpt in the RMD-treated RMs. Rebounding PVLs peaked at up to 104 vRNA/ml by 13 dpt. After each RMD administration, this consistent virus rebound was gradually controlled to less than 10 copies/ml by 34 dpt (Fig 5a). With repeated RMD administrations, we observed increasingly robust virus rebounds, as documented by increased PVL peaks and longer delays to virus control (Fig 5a). We concluded that RMD has the ability to reactivate the controlled virus in vivo in post-treatment controller RMs. The delays in control of virus rebound, as well as their relative robustness may be due to the fact that, in the absence of ART, new cycles of replication occurred, permitting viral detection in plasma. As such, our study design allowed us to both confirm the efficacy of RMD in reversing latency and document that the virus reactivated after RMD administration is replication-competent. Yet, our results also point to the key observation that the levels of reactivated virus seen with RMD in the presence of ongoing ART are relatively low (as PVLs were below the SCA limit of detection immediately after RMD administration) and suggest that only amplification by de novo rounds of infection in the absence of ART allowed us to observe the effect.
In the absence of ART, RMD administration did not have a significant impact on the levels of total vDNA from CD4+ memory T cells. Thus, after each round of RMD administration, the levels of vDNA transiently increased.This was due to the study design, which allowed the virus to complete cycles of replication, resulting in the seeding of short-lived memory cells (i.e., effector memory cells). As these short-lived cells are productively infected and thus rapidly eliminated, the levels of vDNA in memory cells rapidly returned to pretreatment levels between RMD administrations (Fig 5b).
To analyze the effect of RMD in more detail, we developed a simple dynamical model of virus production (see Methods), which shows that the slopes of increase of (the logarithm of) virus after each cycle of RMD in the absence of ART are related to the enhancement in viral production due to RMD. We estimated the slope of increase in PVLs using a linear-mixed effects model. We found that this slope was not significantly different across the three cycles of RMD treatment in the absence of ART, nor across RMs. The estimated slope was 0.418 log10/day (s.e. 0.037). The dynamical model indicates that the increase in viral production over baseline is proportional to this estimated slope (see Methods). Therefore, we can estimate that the increase in viral production due to RMD was between 1% and 5% of the baseline production before RMD, depending on how fast virus is cleared (100 day-1 or 20 day-1, respectively–see Methods for details). Assuming that at baseline production is in balance with viral clearance (P0 = cV0), these percentages allow us to estimate that the average increase in total body virus production attributable to RMD was only between ~150 and ~8000 virions per day.
Due to the nature of RMD administration by slow perfusion and the potential complications of prolonged anesthesia, all the animals treated with RMD received fluids throughout the time of drug administration. They also received Boost/Ensure via gavage at the last bleed on day of infusion, then again at 1, 2 and 3 dpt in order to compensate for the effects of prolonged sedation and extensive bleeding which could have reduced their appetite. In these conditions, we did not observe any major adverse effects of RMD in any RM, with the exception of a slight weight loss (probably due to frequent anesthesia), that recovered by 15 dpt. There were only minimal signs of toxicity after RMD administration, as suggested by the chemistry tests, which were normal at 5 dpt, with the exception of decreases in creatine kinase in all three animals and fluctuations in urea and total protein levels, as illustrated in S1 Fig. Similarly, complete blood counts (CBCs) did not show any major change in the blood cell populations indicative of drug toxicity (S2 Fig). Importantly, drug toxicity did not increase with repeated RMD administration. However, due to sample limitations, CBCs and chemistries were only performed prior and 5–7 days after RMD administration and these results should be treated with caution. Therefore, in an additional effort to assess potential RMD related toxicity, we closely monitored samples collected at multiple time points after each of the RMD treatments for levels of plasma lactate dehydrogenase (LDH), a marker of cell injury and death [84]. As illustrated in S3 Fig, RMD administration did not result in a significant increase in the levels of LDH in RMs (p = 0.459) (S3 Fig). These results suggest that RMD is effective and safe in RMs at the dose administered in our study.
However, in each of the treated RMs, after every RMD administration, a massive, but transient leukopenia was observed, with the overall levels of lymphocytes being reduced by an average of 76% (range: 55–86%) (Fig 6). Leukopenia occurred within 24 hours after RMD administration and lasted less than three days, with the lymphocyte levels being very rapidly restored to pretreatment levels within 5 dpt (Fig 6a and S2 Fig). This pattern was observed in all three RMs, and after every RMD administration, yet the total lymphocyte populations dramatically fluctuated in RM178 with more limited variations in the remaining two RMs (Fig 6a).
As a result, both CD4+ and CD8+ T cell counts were drastically reduced upon RMD administration (Fig 6b and 6c, respectively), but similar to the overall lymphopenia, they rebounded to pretreatment levels within one week (Fig 6b and 6c). Since T cell recovery after depletion is typically much slower, this rapid rebound suggests that the apparent reduction in T cell counts observed after RMD administration is not due to real cell depletion. We therefore monitored the CD3 expression on the surface of gated lymphocytes and identified a significant downregulation of CD3 following RMD administration (Fig 6d). The frequency of the CD3+ T cells in the lymphocyte gate decreased after RMD administration, with a concomitant increase in the frequency of CD3-negative cells (Fig 6d). As such, our results suggest that the apparent lymphopenia observed after RMD administration is due to downregulation of lymphocyte surface markers rather than a direct depletion of cells due to drug toxicity.
We next monitored levels of activation and proliferation after RMD administration by assessing the fraction of CD4+ and CD8+ T cells expressing the immune activation markers CD25 (Fig 7a), HLA-DR, CD38 (Fig 7b) and CD69 (Fig 7c), which increased only transiently in RMs treated with RMD. Increases in the levels of immune activation markers always preceded increases in PVLs suggesting that RMD can activate resting cells (S4 Fig).
The fraction of CD4+ and CD8+ T cells expressing the proliferation marker Ki-67 also increased significantly, but this increase tended to be slower than seen for CD69 [85] (Fig 7d). Thus, the fraction of CD4+ and CD8+ T cells expressing Ki-67 peaked at 12 dpt, paralleling viral replication (Fig 7d). The frequency of T cells expressing both Ki-67 and immune activation markers returned to pretreatment levels prior to subsequent RMD administration. Altogether, the dynamics of immune activation and proliferation markers, in combination with findings when the animals were administered RMD while on ART [50], suggested that these changes were due to RMD administration rather than a response to viral replication, at least in the initial stages after RMD administration.
A recent ex vivo study attributed the ability of RMD to reactivate HIV from the reservoir to a major effect exerted by this drug (similar to other HDACi) on immune cell effectors, through elimination of CD8+ T cells and a reduction of the cytolytic capabilities of CTLs [51]. As we also observed a major, albeit transient, lymphopenia in RMs after administration of RMD, we next assessed the impact of RMD on SIV-specific T cells in vivo.
Functional activity of both CD4+ and CD8+ T cells were monitored by intracellular cytokine staining (ICS) measurements of IL-2, TNF-α, IFN-γ, CD107α and MIP-1β production in response to stimulation with SIVmac239 Gag or Env peptide pools, measured in samples obtained various time points prior to and after RMD administration (Figs 8 and 9 and S5 and S6 Figs). ICS showed that RMD had only a transient impact on the absolute counts of Gag and Env-specific CD4+ and CD8+ T cells (Fig 10). Thus, while combined cytokine production was transiently hindered after RMD administration, SIV-specific T cell function was rapidly regained, before the viral control was reestablished (Figs 8 and 9 and S5 and S6 Figs). Furthermore, polyfunctionality of the CD4+ and CD8+ T cells was maintained or even boosted after RMD administration, probably as a result of the virus rebound representing a sufficient antigenic stimulus (Figs 8 and 9 and S5 and S6 Figs). The majority of the SIV-specific T cells were positive for the degranulation marker CD107α, which in ICS assays is considered a correlate of cytotoxic potential. Furthermore, the frequency of SIV-specific CD107α positive cells increased after RMD administration (Figs 8 and 9 and S5 and S6 Figs). The same pattern was observed after all RMD treatments. Together with the pattern of viral replication demonstrating control of the rebounding virus after each administration of RMD, our results suggest that in the post-treatment controller RMs that have functional immune responses, the virus reactivated through RMD administration can be effectively cleared by CTLs and that the impact of RMD on SIV-specific T cells is only transient and modest in vivo.
Both the dynamics of the immune activation markers and their correlation with PVLs (Fig 7 and S4 Fig), as well as testing of the specific SIV responses, strongly suggested that virus rebound in the RMD-treated RMs was due to virus reactivation after LRA administration and not to loss of viral control through a major ablation of the CTL functions by HDACi. However, to further discriminate between the loss of control and virus reactivation, we modeled in vivo the ablation of CTL responses through direct experimental depletion of CD8+ cells. Post-treatment RM controllers received the M-T807R1 monoclonal antibody (mAb) to deplete CD8+ cells, after which plasma viremia, and the number and activation status of CD4+ T cells were compared and contrasted with the results observed after RMD administration. The anti-CD8 mAb successfully depleted peripheral CD8+ cells (Fig 11a) and loss of immune control was associated with a dramatic rebound of plasma viremia in all CD8-depleted RMs (Fig 11b). PVLs peaked at up to 107 vRNA copies/ml by 10 days post M-T807R1 administration. As such, the PVLs observed after CD8+ cell depletion were orders of magnitude higher than those observed after RMD administration. PVLs were then slowly controlled over 5 weeks, much slower than after RMD administration, but mirroring the recovery of CD8+ T cells (Fig 11b). This massive viral replication resulted in a significant depletion of the CD4+ T cells in CD8+-depleted post-treatment controller RMs (Fig 11c). CD8+ cell depletion was also associated with a steady increase in the frequency of CD4+ T cells expressing Ki-67 (Fig 11d), which returned to predepletion levels after the rebound of CD8+ cells and control of PVLs.
When the levels of CD4+ T cell immune activation and proliferation markers were plotted on the PVLs in the CD8+-depleted post-treatment controller RMs, the viral rebound clearly preceded the increase in the levels of CD4+ T cell immune activation and proliferation markers (S7 Fig). This suggests that the observed virus rebound resulted from the ablation of the immune responses rather than from activation of the reservoir cells, as observed after RMD administration (S4 and S7 Figs).
Based on these results clearly documenting different patterns of viral rebound and control after RMD administration and CD8+ T cell depletion, we concluded that RMD administration does not trigger a permanent hindrance on CTL function and that virus rebound after RMD administration is due to the drug administration rather than to an ablation of CTL responses by RMD.
As research for a cure for HIV/AIDS gathers momentum, so does the use of animal models that can be employed to answer multiple key questions related to HIV infection pertinent to potential curative strategies, such as the location and structure of viral reservoirs, the impact of various therapeutic approaches on these reservoirs, as well as the toxicity of candidate LRAs [64]. These questions cannot be addressed without very invasive sampling and without major risks that can be achieved in animal models, but not in a clinical setting where the standard of care for HIV-infected individuals on ART means that in spite of being on chronic medication, they are otherwise able to have a virtually normal life, with a life expectancy that nears that of HIV-uninfected patients [86].
Here, we developed a model of post-treatment virus control of SIV infection that recapitulates features of the human post-treatment controllers [75]. RMs were intravenously infected with a new transmitted founder infectious molecular clone [77] derived from the strain SIVsmmFTq. This strain, which was identified during our previous surveys of SIVsmm diversity in Primate Centers in the US [76,87], has never been passaged in vitro and displays a lower pathogenicity in RMs than the highly adapted SIVmac/SIVsmm strains. We reasoned that since the set-point PVLs of this strain are lower than those of the reference SIVmac strains, SIVsmmFTq may be more readily controlled with ART. At 65 dpi, when the set-point viremia was achieved, but before major immune suppression occurred, RMs received an ART regimen consisting of the NRTIs PMPA and FTC and the integrase inhibitor L-812820, which is similar to the NRTI/Integrase inhibitor ART regimens containing Raltegravir or Dolutegravir used in combination of antiretrovirals recommended as first line therapy in HIV-infected patients [88]. ART was given continuously for over nine months, which we reasoned should ensure both completion of the first three stages of SIV decay [89,90], as well as a significant decay of the central memory T cells, the major component of the viral reservoir [91]. As shown by our results, our approach was effective, with PVLs controlled to <10 vRNA copies/ml for the duration of treatment, without any blips. Furthermore, at the end of treatment, the biological parameters improved in treated RMs compared to controls, with a trend to better preservation of CD4+ T cells and a partial control of T cell immune activation. Such profiles are characteristic to HIV-infected patients on ART [86,91].
We next assessed the usefulness of this new model for testing virus reactivation strategies. For these experiments, we chose RMD (94). The rationales for our choice were that HDACi are the most advanced class of LRAs, they are less toxic than other classes of LRAs (i.e., PKC agonists or the JQ1) [34–36,45], and that RMD is one of the most active HDACi [45,49,50]. SIVsmmFTq-infected RMs on ART received RMD at a dose of 7 mg/m2, two-fold higher than the dose previously used in RMs [50,92], but closer to the dose employed in human patients [93,94]. During and after treatment, we collected multiple samples and monitored both the effect of RMD on the levels of acetylated histones and the PVLs. We also monitored the side effects of the drug and report that these side effects were minimal, with the exception of dramatic, but transient lymphopenia, the mechanisms of which are currently being investigated in subsequent studies. The tight sampling schedule limited the amounts of plasma available for the SCA, and increased our limit of detection from 1 to 5–10 vRNA copies/ml. However, this is still a very high sensitivity and we could not detect any increase in PVLs after administration of RMD, in agreement with studies in HIV-infected patients, which suggested that RMD has only a limited effect on the reservoir [56]. Therefore, we decided to stop ART seven days after RMD administration.
A very rapid and massive virus rebound was observed upon ART interruption in all the SIVsmmFTq-infected RMs. Virus rebound was not unexpected, as in HIV-infected patients, rebound is nearly universal at the cessation of ART. However, detectable levels of SIVsmmFTq were quantifiable in plasma of RMs as early as three days after cessation of ART. This was more rapid than expected, considering that in patients in whom ART is initiated early during infection, similar to our RMs, the average time to detectable virus rebound is ~8 weeks [95,96], and is correlated with the duration on ART [96], the levels of HIV-1 DNA [97] and with the expression of T-cell exhaustion markers [98]. Even in patients starting ART during the chronic infection and in whom the immune system is exhausted, virus rebound occurs after an average of 2 weeks after ART interruption [99–101], longer than in the RMs in this present study.
Furthermore, while in HIV-infected patients the magnitude of virus rebound at the cessation of ART is generally similar to the set-point levels of viral replication prior to initiation of ART [81], in our study, the virus rebound was massive (up to 107 vRNA copies), orders of magnitude higher than the set point PVLs established prior to treatment (i.e., 104 vRNA copies/ml). Therefore, based on the characteristics of the virus rebound, we concluded that the excess of SIVsmmFTq replication observed at the cessation of ART was likely due to RMD, which had been administered only one week before. This was a first indication that RMD successfully contributed to virus reactivation.
Unexpectedly, in spite of its massive nature, likely to result in large-scale reseeding of the reservoir, the initial rebound was followed in all RMs by virus control below 50 copies/ml (the limit of detection of conventional assays). Such post-treatment control may raise questions relative to the relevance of our model, considering that ART interruption is associated with permanent loss of virus control in the vast majority of HIV-infected patients [99–101]. Note, however, that there is an important difference between the majority of HIV-infected patients, for whom ART is generally initiated during chronic infection, when they present with a significant immune suppression [99–101], and our RMs, in which ART was initiated during the initial stages of chronic infection. As such, our model of post-treatment control should be compared with HIV-infected patients in whom ART is initiated early and maintained for prolonged periods of time and for whom post-treatment control was reported to occur [75,96,102]. The most prominent case of post-treatment control is the Mississippi baby, in whom a very early initiation of ART (at 30 hours postdelivery) for a relatively long period of time (>18 months) resulted in post-treatment control and delayed virus rebound for 27 months [25]. Similarly, in the ANRS-Visconti cohort of patients, ART was initiated during acute infection for an average of 36 months and post-treatment control was reported to occur in 15% of subjects [75]. Moreover, while clinical trials using short course ART did not report post-treatment control, an impact on the reservoir size has been observed in these studies, resulting in a delayed virus rebound [95,96]. Finally, while a <6 month ART regimen initiated early in infection in SIVmac-infected RMs did not result in post-treatment control, a delay of virus rebound was associated with the early ART administration [71]. As such, the overall conclusion from these studies is that a sufficient reduction of the reservoir leading to a delayed virus rebound requires both early initiation and long duration of ART to allow the decay of central memory cells [60]. Here, we fulfilled both these requirements for the post-treatment control, with ART both initiated early in infection and maintained for duration roughly similar to the half-life of memory cells, and, as such, post-treatment control should not be completely unexpected. Also, note that in our newly developed moderately pathogenic SIVsmmFTq model we obtained a more robust control of viral replication with ART than during infection with the highly pathogenic SIVmac239, in which more aggressive ART regimens are frequently associated with less robust control of the virus and blips of viral replication [50,69,70].
In this context, it is tempting to speculate that the observed pattern of virus replication at the cessation of ART (i.e., massive rebound followed by control) is due to the fact that early and prolonged administration of ART before the total destruction of the immune system, together with the complete control of a moderately pathogenic virus, contributed to reservoir curbing. An alternative explanation is that preservation of effective cell-mediated immune responses (through a lower pathogenicity of the virus and an early initiation of ART) permitted effective control of the virus rebound. Future studies in animals on and off ART will allow us to detail the mechanism(s) of the post-treatment control in this model.
We next investigated whether or not RMD can reactivate the virus in vivo. We already had indications that RMD could have been at least partly responsible for the excess of viral replication compared to the pre-ART levels observed at treatment interruption. However, at the time of these experiments, the ability of HDACi to reactivate the latent virus and reduce the size of the reservoir was downplayed by ex vivo studies [45], as well as in vivo studies in humans and macaques on ART [66,103]. Therefore, it was critical to design a study which could unequivocally confirm the efficacy of RMD in reactivating the latent virus. There was also a debate in the field regarding the strategy of choice for measuring the reservoir. While authors were arguing that the inducible virus, which would be the main source of virus rebound in patients interrupting ART, can only be assessed by employing the quantitative virus outgrowth assay (Q-VOA) [104], other authors were arguing that Q-VOA underestimates the levels of inducible virus [105]. The major argument was that Q-VOAs were negative in both the Mississippi baby [26] and in the Boston patients [23], while the virus eventually rebounded in all these patients [23,25]. It was also argued that PCR-based methods overestimate the size of the reservoir and do not always correlate with the levels of inducible virus, as they detect defective viral genomes [106,107]. Since all these arguments were valid and no gold standard for the assessment of the viral reservoir is yet available, we reasoned that for our study, it would be best to use a conventional method for viral quantification to monitor the effects of RMD in vivo (i.e., PVL quantification). Such an approach would not only permit us to bypass controversies in the field, but would also allow us to simultaneously assess both the ability of RMD to reactivate the virus, and the replicative abilities of the reactivated virus, thus representing a valuable strategy to monitor LRA efficacy in preclinical in vivo screenings of new LRAs.
Our rationale was that without ART, the virus reactivated after RMD administration could reinfect new target cells, replicate and amplify, thus increasing the likelihood of detection with conventional PVL assays. We acknowledge that the main limitation of this strategy is that, in the absence of ART, the reactivated replicating virus will reseed the reservoir and prevent direct assessment of whether or not RMD reduces the reservoir. However, the most important question that we wanted to address was whether or not RMD can reactivate replication-competent virus from the reservoir and, thus, whether it has any future in cure research.
We performed three additional RMD administrations to RMs off ART, which resulted in an immediate and transient increase of T cell immune activation, returning to pretreatment levels within 24 hours. This T cell activation was followed by virus rebound in all the RMs treated with RMD. PVLs became detectable at 5 days post-RMD administration and peaked at 103−104 vRNA copies/ml by 13 dpt, thus demonstrating that RMD can indeed activate replication-competent virus from the reservoir. PVLs always followed the increases in the T cell immune activation levels, and therefore we concluded that virus rebound is most likely a result of the reservoir cell activation by RMD and not to loss of viral control through immune cell impairment after RMD administration. Meanwhile, we cannot discount the alternative explanation that the virus rebound might have resulted from the induction of an increased number of target cells, with the implication that RMD-induced SIV transcription may have occurred in only very few cells. In this second scenario, RMD impact on viremia might have been mostly indirect, through homeostatic effects.
To address these issues, using a simple model, we estimated that virus reactivation corresponded to between 1% and 5% of the pre-RMD viral production. Note that viral production may be underestimated, because we assumed that the drug has immediate effect, whereas there likely is a delay of at least a couple of hours post-treatment. Accounting for this delay would make the estimated slope of viral increase larger and hence production would also be larger. Our results suggesting a demonstrable, albeit limited, efficacy of RMD in reactivating the latent virus are supported by recent studies in both humans [49] and RMs [50].
To address an alternative explanation for the observed rebound, i.e., RMD-induced rapid suppression of cytokine production from viable T-cells and selective death of activated T cells, with the net result of impairing the activity of cytotoxic T-lymphocytes, as previously reported [51], we monitored the percentage of SIV-specific T cells in RMs after RMD administration. We report that, while documenting a reduction in the SIV-specific T cells immediately after RMD administration, we did not observe a substantial long-term impact of RMD on cell-mediated immune responses. However, we observed several interesting features that could help explain the results of the studies reporting the deleterious effects of RMD on CTLs. Immediately after RMD administration, we observed a massive, but very transient, reduction in the T cell counts, with the CD3+ T cell counts recovering in less than 5 days post-RMD administration. Due to the extremely transient nature of this reduction, we reasoned that it does not result from a real depletion of CD3+ T cells, but rather points to the downregulation of the surface markers used for cell counts. An alternative explanation for the observed kinetics of T cells might have been their redistribution to tissues, but in the absence of adequate tissue sampling, we could not assess this alternative. However, we showed that the CD3+ T cell reduction was mirrored by a dramatic increase of the negative cells in the lymphocyte gate. As such, it is possible that the effects of RMD on the cellular immune responses may be artefactual and are likely not the factor behind the observed virus rebound.
Finally, to understand whether or not the impact of RMD on cell-mediated immunity contributes to the observed results, we modeled the damage of cellular immunity through experimental depletion of CD8+ cells. RMs received the M-T807R1 mAb and the impact of CD8+ cell depletion was monitored by assessing both the levels of viral replication and those of immune activation. This experiment clearly showed that ablation of CD8+ cells contributed to a massive virus rebound, which was orders of magnitude higher than that observed after RMD. One may argue that CD8+ cell depletion is not comparable with the changes observed following RMD administration which occurs through a different mechanism and only partially impacts the CD4+ and CD8+ T cell populations. Further, the M-T807R1 monoclonal antibody impacts the NK cells, which may also contribute to the observed effects of CD8+ cell depletion. However, our major focus was on comparing the patterns of viral reactivation after RMD and CD8+ cell depletion. We report that, contrasting with RMD administration, in which virus reactivation was a result of increased T cell activation, virus rebound following CD8+ cell depletion preceded the increases in immune activation. Due to these clearly different patterns of virus rebound after RMD or CD8+ cell depletion, we concluded that the observed virus reactivation in our studies is due to RMD and not to impairment of CTL responses.
Our study design with RMD in the absence of ART, did not prevent the induced virus from reinfecting susceptible cells, leading to additional cycles of viral replication, reseeding the tissues, and altering the final size of the inducible viral reservoir. Further, due to sample size limitations, we could not perform a thorough characterization of the reservoir changes for the different rounds of RMD treatment. Thus, trying to assess in detail the impact of RMD on the reservoir would be an exercise in futility. However, quantification of the total memory cell-associated vDNA levels revealed that, while the size of the viral reservoir apparently did not significantly change between RMD administrations, a transient increase in the levels of memory cell-associated vDNA occurred after each RMD treatment. This might explain the trend to higher levels of viral reactivation after each additional round of RMD: virus seeding of the short-lived effector memory cell population leads to alterations in the composition of the viral reservoir. For example, we may speculate that, due to virus control during prolonged ART, in the initial viral RMD-induced reactivation, the rebounding virus likely originated nearly exclusively from long-lived resting central memory CD4+ T cells. However, with every round, the reactivated virus infects mostly susceptible cells (i.e., activated CD4+ memory cells) which could contribute to plasma virus in the subsequent rounds.
In conclusion, our results demonstrate that RMD may be successfully used to reactivate the latent virus and one may expect that, in the presence of an effective immune response, this intervention may curb the reservoir. Studies of virus reactivation in a background of ART, which are currently ongoing, will enable us to assess whether or not RMD administration can significantly reduce the size of the reservoir. Since the drug efficacy in reactivating the virus is rather modest, combination between different LRA classes or with immune modulators might be a strategy of choice for our attempts to reduce/eliminate the reservoir.
All animals were housed and maintained at the University of Pittsburgh according the standards of the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC), and experiments were approved by the University of Pittsburgh Institutional Animal Care and Use Committee (IACUC). These studies were covered by the following IACUC protocol: 13011370. The animals were fed and housed according to regulations set forth by the Guide for the Care and Use of Laboratory Animals and the Animal Welfare Act [108]. All RMs included in this study were socially housed (paired) indoors in stainless steel cages, had 12/12 light cycle, were fed twice daily and water was provided ad libitum. A variety of environmental enrichment strategies were employed including housing of animals in pairs, providing toys to be manipulated, and playing entertainment videos in the animal rooms. Furthermore, the animals were observed twice daily and any signs of disease or discomfort were reported to the veterinary staff for evaluation. For sample collection, animals were anesthetized with 10mg/kg ketamine HCI (Park-Davis, Morris Plains, NJ, USA) or 0.7 mg/kg tiletamine (HCI) and zolazepan (Telazol, Fort Dodge Animal Health, Fort Dodge, IA) injected intramuscularly. At the end of the study, the animals were sacrificed by intravenous administration of barbituates.
Ten Indian RMs were included in the study. They were infected with plasma equivalent to 300 tissue culture infectious doses (TCID50) of SIVsmmFTq [76,109] transmitted-founder infectious molecular clone. Clone derivation and preparation was similar to that reported by our group previously [77]. None of the RMs included in this study harbored MHC genotypes associated with control of SIV virus replication (i.e. A*01, B*08, or B*17). Further, RMs were selected to be either homozygous or heterozygous for the TRP allele of Trim5α [110] as the infectious molecular clone was constructed to bypass TFP restriction [110].
Sixty days post-SIVsmmFTq infection, after the resolution of acute infection and establishment of the chronic viral setpoint, ART was initiated in four RMs. ART consisted of the reverse transcriptase inhibitors (R)-9-(2-phosphonylmethoxypropyl) adenine (PMPA; tenofovir; 20mg/kg; gift from Gilead Biosciences) and β-2’, 3’-dideoxy-3’-thia-5-fluorocytindine (FTC; emtricitabine; 50mg/kg; gift from Gilead Biosciences) by once-daily subcutaneous injection, the integrase inhibitor L-870812 (20mg/kg; gift from Merck) b.i.d. for nine months. PMPA and FTC were administered subcutaneously, while L-870812 was administered orally. During treatment, after all the RMs controlled viral replication, one animal was euthanized due to an unrelated clinical condition (complications of anesthesia).
After 9 months of ART, prior to treatment cessation, RMs were treated with the LRA RMD (Istodax, Celgene Corporation, Summit, NJ) at a dose of 7 mg/m2 in a slow perfusion over four hours. After cessation of ART, the RM controllers received three additional doses of RMD in similar conditions every 35–50 days.
Finally, 42 days after the last RMD administration, RMs received the CD8+ cell-depleting monoclonal antibody M-T807R1 (NIH Nonhuman Primate Reagent Resource, Boston, MA) at a dose of 50mg/kg. Animals were closely clinically monitored for physical and physiological changes during all stages of the study.
Blood was collected from all RMs as follows: three times prior to infection (-30, -15 and 0 dpi), biweekly for the first two weeks (4, 7, 10, and 14 dpi) and weekly thereafter. This sampling schedule was designed to monitor viral replication during the acute infection and establishment of the viral set point, at which time point ART was to begin after 5 consecutive similar PVL measurements. During ART treatment, a weekly sampling was designed to monitor for viral blips. Upon cessation of ART, blood was sampled every 3 days to monitor virus rebound.
After RMD and anti-CD8 mAb administration, the schedule of blood collection was as follows: 0, 4 and 6 hours post-treatment, followed by 1, 2, 5, 12, 14, 21, 28, 35 dpt.
Within one hour after blood collection, plasma was harvested and peripheral blood mononuclear cells (PBMCs) were separated from the blood using lymphocyte separation media (LSM, MPBio, Solon, OH).
Blood chemistries and complete blood counts (CBCs) were obtained from Marshfield Laboratories (Cleveland, OH) from serum and whole blood, respectively.
We monitored the levels of viral replication to assess treatment efficacy as well as the impact of RMD administration on the reservoir virus. Most samples were subject to a quantitative reverse-transcription PCR, as described previously [111]. For samples that achieved <50 copies/ml, a single copy assay (SCA) was performed, as described [83,112].
Large volumes of plasma (5–8 ml) were pooled and virus pelleted by ultracentrifugation at 170,000 x g for 30 min in a Sorvall T1270 rotor. To compensate for increased amounts of nonvirus materials in the plasma that can potentially interfere with accurate quantitation, a known amount of RCAS [83] was added to each sample prior to centrifugation. This serves as an internal control to monitor the overall efficiency of the assay. RNA was isolated as follows: virus pellets were suspended in 100 μl proteinase K for lysing and digestion; 400 μl of GuSCN/glycogen (glycogen acting as a carrier) was added and followed by 500 μl of isopropanol to precipitate RNA. The RNA samples were then resuspended in 65 μl Tris-HCl, pH 8.0, DTT, RNasin mixture. cDNA was first prepared in triplicate reaction mixtures for each RNA samples in a 96-well PCR plate. SIVsmmFTq gag standard (diluted to 1 copy/ml), two 7500 copy RCAS aliquots and corresponding water (negative control) were added the plate. Reaction mixtures contained 10 μl RNA and 12 μl cocktail [reverse transcriptase plus buffer, Superscript III First-strand synthesis Supermix for qRT-PCR kit (Invitrogen)]. cDNA was then synthesized under the following thermal conditions: 25°C for 10 min, 50°C for 50 min, 85°C for 5 min, and 4°C hold. Following reverse transcription, either SIVsmmFTq primers/probe for Gag (200 nM and 100 nM, respectively) or RCAS primers/probe and 30 μl Taqman Gene Expression Master (Applied Biosystems, Foster City, CA) were added to their respective wells. The primer and probe sequences are as follows: SIVFTqF: 5’-AAG TCC AAG AAC ACT GAA TGC ATG-3’; SIVFTqR: 5’-TAT AAT TTG CAT GGC TGC CTG ATG-3; SIVFTqProbe: 5’-/56-FAM/AGC GGA GGT/ZEN/AGT GCC AGG ATT CCA GGC/3IABkFQ/-3’; RCASF: 5’-GTC AAT AGA GAG AGG GAT GGA CAA A-3’; RCASR: 5’-TCC ACA AGT GTA GCA GAG CCC-3’; RCASProbe: 5’-/56-FAM/TGG GTC GGG/ZEN/TGG TCG TGC C/3IABkFQ/-3’. Real-time PCR and data assimilation were performed utilizing an ABI 7900 HT real-time machine under the following thermal profile: 95°C for 10min to activate the polymerase, followed by 50 cycles of 95°C for 15 seconds, 60°C for 1 min.
To monitor the impact of ART on major immune cell populations with emphasis on CD4+ T cell restoration and immune activation, the immune cells were immunophenotyped by flow cytometry. First, a two-step TruCount technique was used to enumerate CD4+ and CD8+ T cells in blood, as previously described [113]. The number of CD45+ cells was quantified using 50 μl of whole blood stained with antibodies in TruCount tubes (BD Biosciences) that contained a defined number of fluorescent beads to provide internal calibration. The numbers of CD4+ and CD8+ T cells were then calculated based on the ratio of CD4+ and CD8+ T cells to CD45+ cells in whole blood at the same time point. Whole peripheral blood was stained with fluorescently-labeled antibodies (all antibodies from BD Bioscience, San Jose, CA, USA unless otherwise noted), CD4 (APC), HLA-DR (PE-Cy7), CD45 (PerCP), CD25 (PE), CD69 (APC-Cy7), CD20 (APC-H7), CD8 (PE-Texas Red) (Invitrogen) and CD38 (FITC) (Stemcell). For intracellular staining, cells were fixed, permeabilized and stained for Ki-67 (PE). Flow cytometry acquisitions were performed on an LSR II flow cytometer (BD Biosciences) and flow data were analyzed with FlowJo software (Treestar, Ashland, OR, USA).
To monitor the dynamics of changes in the virus reservoir due to RMD administration, frozen PBMCs were thawed and CD4+ total memory (effector and central memory) cells were sorted using magnetic bead kits and an AutoMACs Pro Separator (Miltenyi Biotec Cambridge, MA). Briefly, CD4+ cells were sorted by staining the PBMCs with the NHP CD4+ T cell isolation kit (Miltenyi Biotech). The sorted cells were then stained with CD95 (PE) (BD Bioscience, San Jose, CA), and then with anti-PE microbeads (Miltenyi Biotech) for the sorting of total memory cells. Sorted cells were pelleted and dry frozen for DNA extraction. After each sort, we removed 105 cells for purity check.
The purity checks were performed as follows: cells from the first sort were stained with the following antibodies: CD3 (V450), CD8 (PE-C594), CD4 (PE), CD14 (FITC), CD20 (APC-H7), CD11c (APC), CD123 (Pe-Cy7), HLA-DR (PerCP). Cells from the second sort were stained with the following antibodies: CD3 (V450), CD4 (APC), CD95 (PE), CD28 (PE-Cy7). Stained cells were analyzed on a LSR II flow cytometer and flow data was analyzed using with FlowJo software. Purity checks confirmed that the purity of the sorted populations was higher than 90%.
Total DNA was extracted from the cell pellets using Qiagen DNeasy blood and tissue kit (Qiagen, Valencia, CA). Extracted DNA was then subjected to quantification using the same assays for the plasma vRNA quantification, but omitting the reverse transcription step. Simultaneous quantification of CCR5 was done to normalize sample variability and allow accurate quantification of cell equivalents. The CCR5 primer and probe sequences were: RMCCR5F: 5’- CCA GAA GAG CTG CGA CAT CC—3’; RMCCR5R: 3’- GTT AAG GCT TTT ACT CAT CTC AGA AGC TAA C—3’; RMCCR5Probe: 5’- /56-FAM/TTC CCC TAC/ZEN/AAG AAA CTC TCC CCG GTA AGT A/3IABkFQ—3’. Primers and probes were ordered from Integrated DNA Technologies (Integrated DNA Technologies (IDT), Coralville, IA), and the Taqman Gene Expression mix was from Applied Biosystems (Applied Biosystems, Foster City, CA). The detection limit of the viral DNA quantification assay was 30 copies/106 cells.
Treatment efficacy was determined by measuring the levels of histone acetylation in PBMCs using a flow cytometric assay, on samples collected prior to RMD administration and then at 6 hours, 1, 3, 5 dpt. Approximately 2 x 106 freshly isolated PBMCs were surface immunophenotyped for 20 min at room temperature in the dark by using the following flow panel: CD69-brilliant violet (BV) 421 (FN50; Biolegend), CD4-V450 (L200), CD14-BV570 (M5E2; Biolegend), CD8-PE (SK1), CD28-ECD (CD28.2; Beckman Coulter), CD95-PE-Cy5 (DX2), PD-1-PE-Cy7 (EH12.2h7; Biolegend) and CD3-APC-Cy7 (SP34-2). Cells were immediately treated with PhosFlow lyse/fix buffer and incubated for 30 min at 37°C, washed twice, permeabilized with 0.4% Triton X-100 buffer (Sigma) for 10 min at room temperature in the dark, and washed again. Permeabilized cells were then stained intracellularly for 30 min at 4°C in the dark with the following antibodies: acetylated histone (recognizes several residues on histones H3 and H4; 3HHH4-2C2; Active Motif) and Ki-67-Alexa Fluor 647 (B56). Prior to use, the acetylated histone antibody was FITC labeled by using a Zenon reagent kit (Invitrogen), according to manufacturer’s instructions. After washing, stabilizing fixative was added, and approximately 200,000 CD3+ T cells were acquired for each sample by using a BD LSR-II flow cytometer. Population gating was performed using corresponding fluorescence minus one (FMO) and untreated negative control samples.
Frozen PBMC samples were thawed, counted and treated with the following antibody cocktail: CD107α (BD), CD28 (BD), and CD49d (BD) in R10 media. The cells were stimulated by four conditions: SIVmac239 Env peptide pool, SIVmac239 Gag peptide pool, positive control staphylococcal enterotoxin B (SEB), and negative control DMSO. SIVmac239 Env and Gag peptide pools were obtained through the AIDS reagent program, Division of AIDS, NIAID, NIH. The samples were incubated for 2 hours at 37°C. Cells were then treated with Brefeldin A (Sigma) and monensin (Sigma) for 4 hours at 37°C. Cells were then stained with Blue LIVE/DEAD (Invitrogen), CD4 (APC), CD8α/β (PE-Texas Red) and CD3 (V450) for surface and TNFα (AF700), IFNγ (FITC), IL-2 (PE), and MIP-1β (PE-Cy7) for intracellular. Samples were run on a LSR-II flow cytometer and analyzed with FlowJo software.
To investigate if leukopenia observed after RMD administration is due to a real destruction of leukocytes following RMD administration, LDH levels in plasma were quantified by ELISA according to the manufacturer’s protocol (NeoBioLab, Cambridge, MA, USA). Results were expressed in ng/ml plasma and the ranges of detection were 5–100 ng/ml. In healthy human individuals, the LDH levels range from 5.6 to 226 ng/mL.
Levels of LDH were compared between D0 and mean peak using paired t-test. Differences in the levels of leukocytes and immune activation markers were determined using Mann-Whitney U test. GraphPad Prism 6 (Graphpad software) was used for all statistical analysis except for mixed-effects models, which were determined using R.
To analyze the impact of RMD on activating cells into viral production we used a simple model, similar to a previously published analysis [114]. We assume that before each cycle of RMD after ART interruption, the PVL is in approximate steady state, as indicated by the data. Virus is produced at a constant rate P and is removed at rate c per virus, such that the change in viral load is described by dVdt=P−cV. Before RMD, the PVL is approximately constant, dV/dt = 0, and thus P0 = cV0, where we use the subscript to indicate time 0 of each RMD cycle. Note that at each cycle, the initial viral load (V0) can be different. At each RMD dose, we assume that the production of virus is increased by the recruitment of latent cells into productive infection. We model the early increase of viral production by dVdt=(P0+PR)−cV. This is only valid before new cycles of infection ensue, because then viral production (P0+PR) is no longer constant. This model can be solved to yield V(t)=P0c+PRc(1−e−ct). By taking the log10 of V(t) and expanding the result in a Taylor series to first order, we conclude that early on the PVL changes as
log10(V(t))≈ln(P0/c)ln(10)+cPRP0ln(10)t=ln(V0)ln(10)+PRV0ln(10)t,
where “ln” represents the natural logarithm, and we used that P0 = cV0 to obtain the last expression. These expressions show that the log10 of PVL should grow approximately linearly in time early on after RMD treatment, with slope given by the term multiplying time, t. Below we will use both of the above expressions.
We fitted a linear mixed effects model to the log10 of PVL between 0 dpt of each cycle of romidepsin treatment and the maximum PVL in each case, using the nlme package of R [115]. We used “Time” and “Cycle” of RMD treatment as fixed factors and RMs as a random factor. We check homogeneity of variance (plot of residuals) and normality of error (Normal qq-plot). We found that Cycle was a significant factor, but there was no interaction between Cycle and Time. We also found that the slope of increase over time was not significantly different among the three subjects. Thus, our final mixed-effects linear model that best fitted the initial increase in log10 PVL had a different intercept (i.e., estimated initial PVL, V0) for each macaque and Cycle, but the same slope (S) of increase in all cases, and can be represented by
log10(VMc)=V0,Mc+St,
where c = 1, 2, 3 represents the RMD cycle number after ART cessation and M corresponds to each RM. The fitting estimated Vc0,M and S in each case.
Putting the two approaches together, the linear mixed-effects model estimates with the dynamical model expressions, we see that S = PR/(V0ln(10)). Thus, we can estimate PR, which is the increase in viral production per ml and per unit time due to RMD. We multiply this number by the estimated total blood volume of about 500 ml to obtain the total production. In addition, we see also that S = cPR/(P0ln(10)), allowing estimation of PR/P0, which is the relative increase in production of virus over baseline due to treatment. In this last case, we need to know c for which we use a range of c≈20 day-1 [114] to c≈100 day-1 [116].
|
10.1371/journal.pgen.1007315 | The presence of rNTPs decreases the speed of mitochondrial DNA replication | Ribonucleotides (rNMPs) are frequently incorporated during replication or repair by DNA polymerases and failure to remove them leads to instability of nuclear DNA (nDNA). Conversely, rNMPs appear to be relatively well-tolerated in mitochondrial DNA (mtDNA), although the mechanisms behind the tolerance remain unclear. We here show that the human mitochondrial DNA polymerase gamma (Pol γ) bypasses single rNMPs with an unprecedentedly high fidelity and efficiency. In addition, Pol γ exhibits a strikingly low frequency of rNMP incorporation, a property, which we find is independent of its exonuclease activity. However, the physiological levels of free rNTPs partially inhibit DNA synthesis by Pol γ and render the polymerase more sensitive to imbalanced dNTP pools. The characteristics of Pol γ reported here could have implications for forms of mtDNA depletion syndrome (MDS) that are associated with imbalanced cellular dNTP pools. Our results show that at the rNTP/dNTP ratios that are expected to prevail in such disease states, Pol γ enters a polymerase/exonuclease idling mode that leads to mtDNA replication stalling. This could ultimately lead to mtDNA depletion and, consequently, to mitochondrial disease phenotypes such as those observed in MDS.
| Mitochondria are essential for energy production, and defects in the maintenance of mitochondrial DNA (mtDNA) lead to a variety of human diseases including mtDNA depletion syndrome (MDS). Certain forms of MDS are caused by imbalances in the mitochondrial deoxyribonucleoside triphosphate (dNTP) pool, which have also been shown to lead to altered levels of the ribonucleotides (rNMPs) that are embedded in mtDNA. In this study, we address the impact of these rNMPs on the mitochondrial DNA polymerase Pol γ at nucleotide concentrations that resemble those found inside a cell. We demonstrate that embedded rNMPs do not impair DNA synthesis by Pol γ even at the lowest concentrations of dNTPs tested. Based on these results, an increase in mtDNA rNMPs is unlikely to explain the mitochondrial defects in MDS. However, we find that Pol γ is inhibited by physiological levels of free ribonucleoside triphosphates (rNTPs). When combined with a dNTP pool imbalance, the presence of rNTPs leads to DNA replication stalling by Pol γ. These characteristics of Pol γ may help to explain the mtDNA depletion in forms of MDS.
| The replication of DNA is a highly accurate process where free deoxyribonucleoside triphosphates (dNTPs) are incorporated opposite their complementary base. In general, DNA polymerases are good at discriminating between ribonucleoside triphosphates (rNTPs; the units that constitute RNA) and dNTPs by virtue of a steric gate residue that clashes with the 2′-OH group of rNTPs [1]. However, because the rNTP concentration in the cell is several orders of magnitude higher than that of dNTPs, DNA polymerases will occasionally erroneously incorporate rNTPs instead of dNTPs [2]. The presence of embedded ribonucleoside monophosphates (rNMPs) in DNA induces structural and chemical changes [3,4] that contribute to unwanted effects such as genome instability and replication stress [5,6].
Due to their negative influence on DNA stability, rNMPs are actively removed from nuclear DNA (nDNA) by the ribonucleotide excision repair (RER) pathway that is initiated by cleavage at the incorporated rNMP by the enzyme RNase H2 [6,7]. Mutations in RNase H2 lead to an increased rNMP frequency in nDNA and can in humans give rise to rare autoinflammatory disorders [8,9]. However, RER-mediated rNMP removal is absent in mitochondria [10,11]. Accordingly, mammalian mitochondrial DNA (mtDNA) has for decades been known to be rich in rNMPs [12–14]. Recent studies using fibroblast cell lines show that rNMPs are present at a frequency of approximately 54 rNMPs per 16 kb mammalian mtDNA molecule [10]. It is currently unclear whether the rNMPs embedded in mtDNA have a functional significance or if they merely are relatively well-tolerated in the mitochondria and therefore do not undergo the prompt removal observed in nDNA. Nonetheless, mutations in the gene coding for RNase H1, an endonuclease implicated in the removal of longer stretches of rNMPs, cause adult-onset mitochondrial encephalomyopathy marked by multiple mtDNA deletions [15], underscoring the importance of at least a certain level of rNMP removal from mtDNA.
MtDNA is a 16.5 kb circular, double-stranded DNA molecule that encodes for key subunits of the oxidative phosphorylation (OXPHOS) system. OXPHOS is responsible for the majority of the ATP production in eukaryotic cells, and malfunctions in this process can lead to neuromuscular disorders, emphasizing the importance of mtDNA integrity [16]. The duplication of mtDNA is performed by a set of dedicated replication proteins that are nuclear-encoded and post-translationally imported into mitochondria. These include the replicative mtDNA polymerase Pol γ, the replicative helicase Twinkle, mitochondrial single-stranded DNA-binding protein (mtSSB) and the mitochondrial RNA polymerase that primes mtDNA replication [17].
Pol γ discriminates rigorously between dNTPs and rNTPs [18], but the frequency of rNMP incorporation into the genome also depends on the ratio between the free dNTPs and rNTPs available in the cell. The mtDNA is especially vulnerable in this respect, since it is replicated independently of cell cycle phase [19]. Outside S phase, dNTP levels are low, and since rNTP levels show little fluctuation over the cell cycle, the rNTP/dNTP ratio is expected to be high [20,21]. The rNTP/dNTP ratio might be particularly high in certain tissues of patients that suffer from defects in the mitochondrial dNTP supply. Mutations in e.g. thymidine kinase 2 [22] or deoxyguanosine kinase [23] are expected to lead to decreased mitochondrial pools of certain dNTPs, which in turn leads to mtDNA instability by a still uncertain mechanism. A recent study by Berglund et al. suggested that an increased rNTP/dNTP ratio could cause the elevated levels of mtDNA rNMPs observed in fibroblasts derived from patients with defects in mitochondrial dNTP metabolism [10]. It was further proposed that increased rNMP accumulation in mtDNA might impair consecutive rounds of replication and could therefore contribute to the mtDNA instability and disease phenotypes observed in these patients. However, no studies have so far addressed how the fidelity and efficiency of Pol γ are affected by incorporated rNMPs at physiologically relevant dNTP and rNTP levels.
Using highly purified recombinant proteins we show that human Pol γ bypasses single rNMPs with an unprecedentedly high efficiency and fidelity. Additionally, our data indicate that human Pol γ—alone or in the context of the reconstituted mitochondrial replisome—has a striking ability of discriminating against rNMPs during incorporation, and displays several-fold lower frequency of rNMP incorporation compared to nuclear replicative polymerases. However, we show that free rNTPs can negatively impact mtDNA replication, likely by competing with the dNTPs for binding to the active site during DNA synthesis by Pol γ, and can lead to Pol γ stalling. In certain diseases, dNTP pool disturbances lead to compromised mtDNA stability without any major effect on the nDNA. We propose that this differential outcome is at least partly due to the sensitivity of the mitochondrial replication fork to high rNTP/dNTP ratios.
Mammalian mtDNA is known to contain embedded rNMPs [10,13,24]. We confirmed the uniform distribution of rNMPs in mouse liver mtDNA by Southern blot analysis and found rNMPs to be embedded on average every 500 nucleotides on either strand (S1 Fig). Due to the relative frequency of rNMPs in mature mtDNA, Pol γ is expected to encounter several rNMPs in the template strand during replication of the ~16 kb mtDNA molecule. If these rNMPs impair replication, they could have pathological consequences for patients that, as a result of a shortage of specific dNTPs, have increased levels of rNMPs in mtDNA [10]. The exonuclease deficient variant of Pol γ has previously been reported to be able to bypass rNMPs [18], however, we wanted to examine the bypass of wild type Pol γ as well as the efficiency of bypass at a range of physiologically relevant dNTP concentrations. These in vitro polymerization reactions were carried out on synthetic DNA templates where Pol γ encountered the ribonucleotide at the 5th position after the initiation of DNA synthesis (Fig 1A); the control templates contained the corresponding deoxyribonucleoside monophosphate (dNMP) in place of rNMP in an identical sequence context. We found that bypass of a single embedded rNMP by Pol γ was not obviously reduced relative to an all-dNMP template, even at the lowest dNTP concentration tested (0.01 μM; Fig 1B, compare dNMP/rNMP pairs). This result indicates that DNA synthesis by the Pol γ holoenzyme is not strongly inhibited by a single rNMP in the DNA template at the range of dNTP concentrations present in vivo. Similar results were observed with the exonuclease-deficient D274A Pol γ variant (S2 Fig).
The above reactions were performed with a 2.5-fold excess of polymerase over DNA template, whereby re-initiation of DNA synthesis could mask a moderate reduction in bypass efficiency. We therefore performed primer extension assays in reaction mixtures containing a large excess of template-primer over DNA polymerase, so that once a primer is extended, the probability that it will be used a second time is negligible, and the products therefore derive from a single cycle of synthesis. Band intensities were used to calculate termination probabilities at specific nucleotide positions surrounding the rNMP, as previously described ([25], also see Materials and Methods). Under these conditions, the presence of rNMP in the template led to a relatively moderate increase in termination probability, especially at positions -2 and -1 relative to the embedded rNMP (Fig 1C). In addition, the rUMP-containing template showed an increase in termination probability at position +1.
Taken together, the results of Fig 1 show that even at low dNTP concentrations the efficiency of Pol γ is only slightly affected by single rNMPs present in the DNA template, which is in agreement with the reverse transcriptase activity of Pol γ [26,27]. Notably, the observed effect of single rNMPs on the termination probability of Pol γ is considerably lower than that reported for yeast and human nuclear replicative DNA polymerases [28–30].
We next addressed the fidelity of rNMP bypass by sequencing the DNA products of a primer extension assay in order to determine the base inserted by Pol γ opposite a single template rNMP present at position +5 relative to the primer terminus. Sequencing analysis revealed that Pol γ incorporates the correct base opposite a rNMP with similar fidelity as opposite a dNMP (98.3% versus 98.2%, which are expected values for this PCR based assay). This shows that a single rNMP in the template has no detectable adverse effects on insertion fidelity. In contrast, Pol γ incorporated the correct base, dC, opposite 8-oxo-7,8-dihydroguanine (8-oxo-G) in only 68% of the sequenced products, which is comparable to the 73% found in an earlier report [31], validating the experimental set-up. The fidelity of the proofreading-deficient D274A mutant of Pol γ was found to be slightly lower opposite an rNMP (correct base in 96.4% of products), but a similar drop in fidelity was observed on an all-dNMP template (96.0%). These data suggest that unrepaired rNMPs in the mitochondrial genome are well-tolerated by Pol γ, not only when it comes to bypass efficiency (Fig 1), but also in terms of replication fidelity.
Pol γ, the replicative mitochondrial polymerase, has been suggested to be the main source of the rNMPs embedded in the mitochondrial genome, and it has been shown to incorporate one rNMP for every 2.0 ± 0.6 × 103 bases on a short 70 nt template in vitro [10]. However, earlier work on nuclear DNA polymerases has shown that in vitro analysis of rNMP insertion frequency is greatly affected by the sequence context of the short, defined oligonucleotides used in these studies [32]. To circumvent this problem, we used the 7.3 kb M13 ssDNA as a template to estimate the propensity of Pol γ to incorporate rNMPs (Fig 2A). Use of a long DNA template allows many sequence contexts to be tested simultaneously and the results should therefore give a closer estimate of the rNMP incorporation frequency occurring in vivo. To further increase the relevance of our data, we aimed to perform the reactions at physiologically relevant rNTP and dNTP concentrations. In an attempt to simulate the conditions that prevail during different stages of the cell cycle, two different sets of dNTP concentrations were tested: “normal” that mimic the concentrations present during S phase, and “low” that represent concentrations found during the rest of the cell cycle [33–35]. The latter dNTP concentrations are also expected to be similar to those found in non-dividing cells where nuclear DNA is not being replicated, but mtDNA replication still occurs. As a reference, we carried out reactions using the S. cerevisiae lagging strand DNA polymerase, Pol δ, of which the rNMP incorporation frequency has been reported using an identical assay set-up [7]. Unfortunately, comparison to human nuclear replicative polymerases is prevented by the lack of data on their rNMP incorporation frequencies using a similar long-template assay, and the general discrepancy between incorporation frequencies determined using long vs. short templates [7]. To enable comparison to literature values, some reactions with Pol γ and yeast Pol δ were performed using the nucleotide concentrations measured from logarithmically growing S. cerevisiae cells [2]. In all reactions, the DNA template was coated with the relevant single-stranded DNA binding protein (Fig 2B, lanes 1–6 with RPA; lanes 7–14 with mtSSB) to avoid stalling of DNA synthesis due to the formation of secondary DNA structures.
Radioactively labelled products of the primer extension reactions on M13 ssDNA were treated with NaOH to hydrolyse the phosphodiester bond on the 3′ side of incorporated rNMPs and analysed by agarose gel electrophoresis under mildly denaturing conditions (Fig 2B). In the absence of rNTPs, Pol γ synthesized long products that were only moderately affected by alkaline treatment (Fig 2B, compare lanes 7–8 and 11–12). In contrast, the DNA products synthesized in the presence of rNMPs were alkali-sensitive and a range of smaller DNA products was observed (Fig 2B, compare lanes 9–10 and 13–14) indicating rNMP incorporation. The distribution of radioactive signal in individual lanes was quantified and transformed to a size distribution as previously described [7] in order to determine the median length of alkali-stable DNA fragments. The median length values were further used to determine the frequency of rNMP incorporation (see Materials and Methods for details). Please note that because the signal in Fig 2B derives from incorporation of α-32P dCTP, longer products that contain a larger number of radioactive nucleotides appear stronger in intensity than shorter products. This causes the apparent product length on the gel to appear longer than the actual median length determined from the size distribution plot (in which the signal has been corrected for length). At “normal” dNTP levels, the median length of untreated DNA products synthesized by Pol γ was 5.2 kb (Fig 2B, lane 9) and this value dropped to 1.6 kb after NaOH treatment (Fig 2B, lane 13). These median lengths correspond to an average rNMP incorporation frequency of 1 rNMP per 2300 nt at “normal” dNTPs levels, while at “low” dNTP concentrations, the average rNMP incorporation frequency was 1 rNMP per 1400 nt (Fig 2B, lane 10 vs. lane 14). The calculations take into account that Pol γ was unable to fully replicate the 7.3 kb template in the presence of “low” dNTPs (Fig 2B, lane 10) due to a strong reduction in replication rate in the presence of rNTPs.
In comparison, the rNMP incorporation frequencies obtained for Pol δ were 1 rNMP per 770, 650 and 400 nt at “S. cerevisiae”, “normal” and “low” dNTP concentrations, respectively (Fig 2B, lanes 1–6). The rNMP incorporation frequency at “S. cerevisiae” dNTP concentrations is in excellent agreement with the previously reported value of 1 rNMP per 720 nts obtained using an identical approach [7], thus validating our experimental set-up. In conclusion, the results of Fig 2B show that Pol γ incorporates rNMPs far less frequently than its yeast homolog Mip1 (1 rNMP per 600 nts; [11]) or the yeast nuclear replicative polymerases Pol δ or Pol ε (1 rNMP per 720 or 640 nts, respectively [7]).
For processive replication of double-stranded mtDNA, Pol γ requires the activity of the mitochondrial DNA helicase Twinkle. To examine the impact of Twinkle on rNMP incorporation by Pol γ, we constructed a DNA substrate consisting of a primed single-stranded mini-circle with a 40 nt 5′-overhang to allow Twinkle loading (Fig 2C). Once initiated, leading-strand DNA synthesis, coupled to continuous unwinding of the DNA template, can in theory continue indefinitely (rolling circle replication). This replication system requires ATP, since DNA unwinding by Twinkle is an ATP-dependent process. As the concentration of ATP normally employed in our in vitro assays (4 mM) would interfere with the rNMP incorporation studies, we made use of a creatine phosphokinase-based ATP regeneration system that maintained the ATP concentration at a lower concentration of approximately 300 μM. This way, the influence of free ATP in the assays without rNTPs was minimized while ensuring that efficient rolling circle replication could occur (S3 Fig).
In the absence of rNTPs, the reconstituted human mitochondrial replisome consisting of Pol γ, Twinkle and mtSSB synthesized long DNA fragments at all dNTP concentrations tested (Fig 2D, lanes 1–3). Alkali treatment of the reactions lacking rNTPs resulted in a small drop in DNA fragment size, which was ascribed to the presence of the 300 μM ATP required for Twinkle function (Fig 2D, compare lanes 1–3 with lanes 7–9). In contrast, the DNA products synthesized in the presence of rNTPs showed a substantial reduction in length upon alkali treatment, consistent with rNTP incorporation (Fig 2D, compare lanes 4–6 with lanes 10–12). The median length of DNA products synthesized in the presence of rNTPs was 5.2 kb using “normal” and 1.9 kb using “low” dNTP concentrations (Fig 2D, lane 4 and 5). NaOH treatment reduced the median length of the DNA products to 1.5 kb and 0.86 kb under “normal” and “low” dNTPs concentrations, respectively (Fig 2D, lane 10 and 11). Based on these median length values, the average rNTP insertion frequency of Pol γ holoenzyme in the presence of the helicase Twinkle and mtSSB is 1 rNMP per 2200 and 1600 nts at “normal” and “low” nucleotide concentrations, respectively. As expected, at the somewhat higher dNTP concentrations prevailing in logarithmical-growing S. cerevisiae cells, the rNMP incorporation frequency was the lowest, 1 rNMP per 2400 nts. These incorporation frequencies are similar to the values obtained in the reactions without Twinkle (Fig 2B), showing that the rNMP incorporation propensity of Pol γ is not affected by the presence of other core mitochondrial replication factors.
Numerous studies have shown that nuclear replicative DNA polymerases proofread incorporated rNMPs poorly, if at all [29,30,36]. Because proofreading of rNMPs could explain the relatively low rNMP incorporation frequency by Pol γ observed in Fig 2B, we tested the contribution of its exonuclease activity to rNMP incorporation both in vitro and in vivo.
First, we carried out in vitro primer extension assays on M13mp18 ssDNA with either wild type (WT) or exonuclease-deficient (exo-) Pol γ (Fig 3A). An extended reaction time (120 min) and addition of extra polymerase after half the time ensured that the majority of DNA products synthesized were full-length (7.3 kb) (Fig 3B, lanes 1–4). Analysis of NaOH-treated products synthesized in the presence of rNTPs revealed a comparable distribution of DNA fragment sizes from reactions catalysed by wild type or exo- Pol γ (Fig 3B, lane 7 vs. 8; and Fig 3C, “WT + rNTPs” vs “exo- + rNTPs”). This observation demonstrates that the proofreading activity of Pol γ does not influence the frequency with which it incorporates rNMPs in vitro.
To confirm our in vitro findings, we next isolated mtDNA from mice homozygous for an exonuclease-deficient mutant of the catalytic subunit of Pol γ (PolgAD257A /D257A, hereafter referred to as PolgAD275A). Due to the lack of proofreading activity by the PolgAD275A variant, these mice accumulate mtDNA mutations and show signs of premature ageing [37]. Linearized mtDNA isolated from the liver of WT PolgA+/+ and the proofreading-deficient PolgAD275A mutant mice migrated as expected as a 16 kb mtDNA molecule (Fig 3D, lanes denoted with “C”). Consistent with the results shown in S1C Fig, alkali treatment resulted in shorter DNA fragments (Fig 3D, lanes denoted with “A”). The size distribution plots of alkali-treated wild type and PolgAD275A mtDNA were indistinguishable from each other, indicative of a comparable rNMP incorporation profile (Fig 3E). The results of the in vivo and in vitro experiments therefore both indicate that the exonuclease activity of Pol γ does not influence the amount of rNMPs embedded in the mtDNA.
Our findings presented so far indicate that rNMPs incorporated in the DNA template do not constitute a considerable impediment to replication by Pol γ. On the other hand, the lower amount of synthesis products observed in experiment in Fig 2B (lanes 7–10) suggested that the presence of free rNTPs in the reaction mix in addition to dNTPs may have an adverse effect on DNA synthesis by Pol γ in vitro, especially when dNTP concentrations are low. To better study the effect rNTPs on the rate of replication by Pol γ, we performed DNA replication experiments with the WT and exo- Pol γ variants in the presence and absence of rNTPs using the 3 kb pBluescript DNA template (Fig 4A). As the reaction progressed, both the WT and exo- Pol γ variants were able to efficiently produce a full-length 3 kb DNA product when only dNTPs were present in the reaction mix (Fig 4B, lanes 2–5 and lanes 11–14). However, the addition of rNTPs into the reaction substantially reduced the efficiency of replication catalysed by WT Pol γ, and only a faint full-length band was visible even after a 90-min reaction (Fig 4B, lanes 6–9). In contrast, the length of DNA products synthesized by the exo- Pol γ variant was unaffected by the presence of rNTPs in the reaction (Fig 4B, compare lanes 11–14 with lanes 15–18). Quantification of full-length reaction products confirmed that the processivity of Pol γ was only affected by the presence of free rNTPs when its proofreading ability was intact (Fig 4C). This suggests that the presence of rNTPs forces the wild type enzyme to slow down and/or stall in a manner that depends on a functional exonuclease domain. However, as shown in Fig 3, the proofreading activity of Pol γ is unable to selectively remove incorporated rNMPs. Taken together, these findings are consistent with a scenario where the presence of free rNTPs forces WT Pol γ to idle between polymerase and exonuclease modes, leading to slower replication.
The above reactions were performed in the presence of 10 μM equimolar dNTPs. However, the rNTP-dependent decrease in Pol γ replication speed was even more pronounced at physiologically relevant (“normal”) dNTP concentrations (S4A Fig, lanes 6–9). Similar reactions with the S.cerevisiae homolog of Pol γ, Mip1, showed that also the replication rate of this polymerase was affected by the presence of the rNTPs, albeit to a lesser extent than its human counterpart (S4B Fig). Taken together, these results indicate that free ribonucleotides can impair the activity of mitochondrial polymerases, with human Pol γ being more affected than its yeast homolog Mip1.
We next compared the effect of different dNTP concentrations on synthesis by Pol γ when rNTPs were present at constant, physiological concentrations. At high dNTP concentrations (25 and 50 μM), there was no striking difference in the size of the products from reactions carried out in the absence or presence of rNTPs (Fig 4D, compare lanes 5–6 with lanes 11–12). However, as dNTP levels decreased, the adverse effect of rNTPs on replication became increasingly evident. At 10 μM dNTPs, approximately half of the products of the dNTP-only reaction were full-length, while the rNTP-containing reactions yielded no full-length product (Fig 4D, lanes 4 and 10). Therefore, the apparent inhibition of synthesis by Pol γ in the presence of rNTPs was especially pronounced at low dNTP concentrations.
Finally, we simulated conditions of imbalanced dNTP pools by lowering one dNTP at a time to 1 μM, while the other dNTPs were kept at “normal” concentrations. As expected, limiting the concentration of one dNTP led to a visible reduction in the size of reaction products (Fig 4E, compare lane 1 to lanes 2–5 and lane 6 to lanes 7–10). However, while the reactions containing only dNTPs still yielded fairly long products (Fig 4E, lanes 1–5), the size of the rNTP-containing reaction products was far below full-length (Fig 4E, lanes 6–10). Decreasing the concentrations of dCTP or dTTP had a more striking effect than limiting dATP or dGTP. The reason that limiting pyrimidines had a greater effect than limiting purines is unclear, and could not be explained by sequence bias in the DNA template where all four bases are represented at equal frequencies. Taken together, the data in Fig 4 suggest that the combined effect of a limiting dNTP together with the presence of rNTPs can lead to severe stalling of replication by Pol γ.
The observation that mtDNA is rich in rNMPs has intrigued researchers for over four decades [10,13,18,24,38]. Although rNMPs are erroneously incorporated during DNA synthesis both in the nucleus and in mitochondria, RER efficiently removes the incorporated rNMPs from nuclear DNA. Unlike the nucleus, mitochondria appear to lack rNMP removal pathways [5,10], meaning that incorporated rNMPs persist and could therefore potentially interfere with mtDNA replication, although this phenomenon has not been studied in detail.
Our analysis of mouse mtDNA fragmentation by alkaline and RNase H2-treatment confirmed the relative abundance of embedded rNMPs in the mammalian mitochondrial genome, as well as their uniform distribution (S1 Fig). However, we show that the presence of rNMPs in the replication template does not negatively impact the efficiency or fidelity of replication by Pol γ, even at the lower dNTP concentrations that mimic the conditions in cycling or non-dividing cells (Fig 1). Therefore, the 3- to 4-fold increase in mtDNA rNMPs that was recently reported in fibroblasts derived from patients with disturbed mitochondrial dNTPs pools [10] is, based on our experiments, unlikely to be problematic for the mtDNA replication fork thanks to the high fidelity and efficiency of Pol γ when bypassing single embedded rNMPs.
Interestingly, we found that the frequency of rNMP incorporation by Pol γ was at least three-fold lower than that of the yeast nuclear Pol δ and Pol ε (Fig 2 and [7]). Based on our in vitro data, Pol γ is expected to incorporate about 14 rNMPs during the synthesis of one 16.5 kb dsDNA molecule of mtDNA in cycling cells (Fig 2B lane 13). This value is likely to be higher (24 rNMPs) in post-mitotic cells in which the dNTP levels are thought to be substantially lower (Fig 2B lane 14). However, our Southern blot analysis of mouse liver mtDNA indicated an in vivo rNMP frequency of approximately 1 rNMP per 500 nucleotides (S1 Fig; corresponds to 65 rNMPs per ds mtDNA), which is in good agreement with the values recently reported using a genome-wide next-generation sequencing approach (54 rNMPs per mtDNA molecule in human fibroblasts and 36 in HeLa; i.e. a frequency of 1:613 and 1:920, respectively [10]). The in vivo rNMP frequency of mtDNA is therefore 2- to 5-fold higher than expected from the rNMP incorporation rates observed in our in vitro experiments (Fig 2B and 2D). This difference may partly be explained by lower than estimated dNTP concentrations or higher than estimated rNTP concentrations inside mitochondria. We found that decreasing dNTP concentrations to as low as 0.5–1 μM in the in vitro replication assay gave rise to an rNMP incorporation frequency that is comparable to the in vivo frequency (1:620; S5B Fig) using exo- Pol γ. However, such low dNTP concentrations in the presence of rNTPs do not support DNA synthesis by the WT Pol γ enzyme (S4A Fig lane 6–9) and are therefore not likely to be the sole explanation to the higher in vivo rNMP frequencies observed by us and others. An alternative, or contributing, explanation for the discrepancy between in vivo and in vitro rNMP levels is that additional polymerases in addition to Pol γ could contribute to incorporation of rNMPs into mtDNA, which then persist due to the absence of ribonucleotide excision repair inside mitochondria [10,11]. For instance, Pol β was recently detected in mitochondria [39]. Interestingly, Pol β levels can greatly affect rNMP incorporation in nDNA opposite oxidative DNA lesions [40]. Given the presumably high levels of oxidatively damaged nucleotides in mtDNA [41], incorporation by Pol β could significantly contribute also to rNMP incorporation frequencies in mtDNA. Similarly, the primase/polymerase PrimPol can synthesize nucleic acids using both rNTPs and dNTPs, and has been shown to be involved in mtDNA maintenance [42–44]. Together, these and possibly other polymerases could influence the rNMP levels in mtDNA [45].
We conclusively show that the comparably low frequency of rNMP incorporation by Pol γ is not due to efficient proofreading of rNMPs, as liver mtDNA from WT and proofreading-deficient Pol γ mice exhibited an indistinguishable profile and frequency of incorporated rNMPs (Fig 3D and 3E). Furthermore, in vitro incorporation frequencies were similar for WT and exonuclease-deficient Pol γ (Fig 3B and 3C). Therefore, the low rNMP frequency of Pol γ is expected to be due to efficient discrimination against rNTPs during the insertion step of DNA synthesis.
Finally, we find that the presence of free rNTPs significantly decreases the replication speed by Pol γ (Fig 4). A similar phenomenon has been described for family B polymerases [7,46], but to our knowledge, this is the first report of such inhibition in a family A polymerase. The negative effect of rNTPs on replication by Pol γ is dependent on the proofreading-activity of the polymerase and is therefore likely due to idling between the polymerase/exonuclease modes in the presence of rNTPs. The inhibition of Pol γ by rNTPs was most pronounced at low dNTP concentrations; at 10 μM dNTPs, a concentration that is comparable to that found in cycling cells, the negative effect of rNTPs impeded synthesis of a full-length product in our assay (Fig 4D). It is likely that at high rNTP/dNTP ratios, Pol γ requires more time to find the correct dNTP as the abundant corresponding rNTP acts as a competitive inhibitor of the enzyme. We show that especially in combination with the decreased level of a single dNTP, the presence of rNTPs in the reaction causes severe replication stalling (Fig 4E). These findings lead us to speculate that the high rNTP/dNTP ratio normally found in cells can be especially challenging in combination with a dNTP pool imbalance such as the ones found in patients suffering from mutations in thymidine kinase 2 (TK2) or deoxyguanosine kinase (DGUOK). Frequent replication stalling under such conditions might prevent the mtDNA replisome from completing replication of the entire mitochondrial genome, potentially leading to mtDNA depletion as it has been reported in patients affected by mtDNA depletion syndromes [47].
MtDNA mutator mouse samples were obtained from the Stewart lab at the Max Planck Institute for Biology of Ageing, Cologne Germany, where the mice are raised and handled in strict accordance to the guidelines of the Federation of European Laboratory Animal Science Associations (FELASA). Breeding and sacrifice protocols were approved by the “Landesamt für Natur, Umwelt und Verbraucherschutz Nordrhein-Westfalen" (84–02.04.2015 & 84–02.05.50.15.004)."
Wild type C57BL/6J mice were euthanized at the age of 2 months and livers were frozen in liquid nitrogen. Liver samples from mtDNA mutator mice were gifted from J.B. Stewart of the Max Planck Institute for Biology of Ageing, Cologne, Germany. These mice carried the PolgAD257A allele [37], but were backcrossed onto the C57Bl/6J nuclear genetic background, and used to generate the mice for this study. Three homozygous mtDNA mutator mice (1 male, 2 females), and two wild type sibling controls (1 male, 1 female) at 50–53 weeks of age were used. Mice were bred to limit female-transmitted mtDNA mutations [48]. All animal procedures were conducted in accordance with European, national and institutional guidelines and protocols, and were approved by local government authorities.
Mouse liver was minced into pieces and homogenized using a glass teflon Dounce homogenizer in homogenization buffer (10 mM HEPES pH 7.8 with 225 mM mannitol, 75 mM sucrose and 10 mM EDTA) followed by centrifugation at 800 × g for 10 min at 4°C. The supernatant was then centrifuged at 12 000 × g for 10 min at 4°C to pellet mitochondria that were resuspended in homogenization buffer and overlaid on a 1.5 M/1M sucrose gradient in 10 mM HEPES pH 7.4 and 10 mM EDTA. After ultracentrifugation at 40 000 × g for 1 h at 4°C in a Beckman SW60Ti rotor, the mitochondrial layer was recovered and diluted in 4 volumes of 10 mM HEPES pH 7.4, 10 mM EDTA. The mitochondria were pelleted at 12 000 × g for 10 min at 4°C in a JA25.5 rotor, resuspended in homogenization buffer, treated with Proteinase K, lysed in 20 mM HEPES pH 7.8, 75 mM NaCl, 50 mM EDTA, 1% SDS and treated again with Proteinase K. The mtDNA was extracted once with (25:24:1) phenol:chloroform:isoamyl alcohol and twice with chloroform. The DNA was precipitated and resuspended in 20 mM HEPES pH 7.2.
1 μg of isolated mtDNA was linearized with SacI, precipitated and dissolved in 10 mM Tris HCl pH 7.5. The DNA was hydrolyzed with 0.3 M NaOH at 55°C or digested with RNase H2 (New England Biolabs) at 37°C for 2 h. Samples were run on a 0.8% agarose alkaline gel (30 mM NaOH, 1 mM EDTA) at 25 V, 4°C for 20 h and blotted onto Hybond-N+ membrane (Amersham, GE Healthcare). Single-stranded probes were end-labelled with γ-32P ATP using T4 polynucleotide kinase (Thermo Scientific) following the manufacturer’s protocol. Double-stranded probes were generated by labelling an approximately 500 bp PCR product with α-32P dCTP using Prime-It II Random Primer Labeling kit (Agilent Technologies). Hybridization was for 16 h at 42°C for ssDNA probes and at 65°C for dsDNA probes. The membrane was exposed to a PhosphoImager screen and scanned in a Typhoon laser scanner (GE Healthcare). The radioactive intensity was quantified using ImageJ software and plotted on a distribution plot. The median size of alkali-treated products was determined from the distribution of the radioactivity intensity and related to the size marker that was run in parallel [7].
Human mitochondrial DNA polymerase γ catalytic subunit A (Pol γ A), processivity subunit B (Pol γ B), helicase Twinkle and mitochondrial single-stranded binding protein (mtSSB) were expressed and purified as previously described [49,50]. For the exonuclease-deficient Pol γ A, a D274A mutant was prepared as previously described [51]. Saccharomyces cerevisiae RPA [52], PCNA [53] and RFC [54] were purified from Escherichia coli overexpression systems, while S. cerevisiae Pol δ [55] was purified from a yeast overexpression system. Wild type S. cerevisiae mitochondrial DNA polymerase Mip1 was purified as previously described [11].
Single-stranded M13mp18 DNA (7.3 kb) was purchased from New England Biolabs. Single-stranded pBlueScript SK+ was prepared as previously described [56]. Primers were either 5′-end labelled with γ-32P using T4 polynucleotide kinase (Thermo Scientific) or purchased with 5′-TET fluorescence label. A 36-mer primer 6330 (Figs 2B and 3B) or primer 682 (Fig 4 and S4 Fig) were annealed in a 1:1 ratio to M13mp18 or pBluescript SK+, respectively. For the linear 70 nt templates, a 25 nt primer (Fig 1B) was annealed to the template containing either a dNTP or a rNTP at position 30 (S1 Table) by heating to 80°C and slowly cooling to room temperature. The 70 nt mini-circle template (Fig 2C) was prepared as described in [51]. All oligonucleotides used in this study are listed in S1 Table.
Primer extension was performed using 2.5–10 nM primed circular ssDNA with 12.5 nM of WT or exo- Pol γ A and 18.75 nM of Pol γ B (as dimer). Additional protein was added after half the incubation time when indicated. MtSSB was added to a final concentration of 750 nM. The following reaction conditions were used: 25 mM Tris-HCl pH 7.6, 10 mM MgCl2, 1 mM DTT and 100 μg/ml BSA; dNTPs and/or rNTPs were added at indicated concentrations and run at 37°C. The reactions with Pol δ were performed essentially as described [7] at 30°C with the following protein concentrations: 3 nM Pol δ, 375 nM RPA, 15 nM PCNA (as trimer), and 3 nM of RFC. Mip1 reactions were performed with 5 nM of Mip1 in the same buffer conditions as Pol γ, but at 30°C. The reactions were performed in the presence of what we termed”normal” (5 μM dATP, 5 μM dCTP, 3 μM dGTP and 10 μM dTTP), “low” (2 μM dATP, 1 μM dCTP, 1 μM dGTP and 2 μM dTTP) or “very low” (1 μM dATP, 0.5 μM dCTP, 0.5 μM dGTP and 1 μM dTTP) dNTP concentrations. rNTP concentrations were kept constant (3 000 μM ATP and 500 μM of CTP, GTP and UTP). As the relatively high concentration of rNTPs could lead to sequestering of divalent cations, additional magnesium (4.5 mM) was included in the reactions containing rNTPs in order to maintain a constant concentration of magnesium throughout the experiment. As a reference, some reactions were performed with the concentrations of nucleotides found in logarithmically-growing S. cerevisiae cells: 16 μM dATP, 14 μM dCTP, 12 μM dGTP, 30 μM dTTP, 3 mM ATP, 0.5 mM CTP, 0.7 mM GTP and 1.7 mM UTP [2]. To follow the reaction, [α-32P]-dCTP (Perkin Elmer) was added. The reactions were incubated at 37°C for 1–120 min, stopped with 0.5% SDS, 25 mM EDTA and cleaned over G-25 columns to remove excess [α-32P]-dCTP. For incorporation assays, the sample was divided in two; one control was treated with 0.3 M NaCl and one sample was treated with 0.3 M NaOH. Before gel loading both samples were incubated for 2 h at 55°C. For use as a size reference, 1 kb GeneRuler (ThermoScientific) was end-labeled with γ-P32 ATP using T4 polynucleotide kinase (ThermoScientific). Samples and size marker were separated on a 1.5% alkaline agarose gel in buffer with 30 mM NaOH and 1 mM EDTA at 17 V for 16 h in 4°C. Visualization and quantification was performed by phosphoimaging of the dried gel on a Typhoon 9400 system (GE Healthcare).
The mini-circle substrate (5 nM) was added to a 10 μl reaction mixture containing 25 mM Tris HCl pH 7.5, 75 mM NaCl, 10 mM magnesium acetate, 1 mM DTT, 100 μg/ml BSA, 4 mM ATP, 12.5 nM Pol γ A, 18.75 nM Pol γ B (as dimer), 250 nM mtSSB and 12.5 nM Twinkle. To keep the ATP concentration as low as possible in the reactions without rNTPs, we used an ATP regeneration system consisting of 400 ng creatine kinase and 5 mM creatine-phosphate-Tris. The reactions included the addition of “low”, “normal” or “S. cerevisiae" dNTPs (concentrations listed above) and rNTPs where indicated. The reactions were performed at 37°C and started by addition of polymerase. At the indicated time points, the reactions were stopped by the addition of 1.1 μl of termination mixture (5% SDS, 250 mM EDTA) and analyzed on an 8% polyacrylamide gel containing 8 M urea. Quantification was performed by phosphoimaging of the dried gel on a Typhoon 9400 system (GE Healthcare).
The average rNMP incorporation rate of Pol γ in vitro (Fig 2B and 2D) was determined as previously described [7]. Briefly, we determined the median size of DNA fragments in the reaction with only dNTPs (a) and in the reaction with both dNTPs and rNTPs (b). These values were used in the following formula to calculate the incorporation frequency: rNTP incorporation frequency = a/(a/b—1).
5–10 nM of template was used in primer extension reactions with 25 mM Tris HCl pH 7.5, 10 mM MgCl2, 1 mM DTT, 100 μg/ml BSA, 12.5 nM Pol γ A and 18.75 nM Pol γ B (calculated as dimer), with an increasing amount of dNTPs. For single hit conditions, reactions contained 50 nM template and 1 nM protein. Reactions were stopped after 2–20 min with 0.5% SDS, 25 mM EDTA and incubated at 50°C for 10 min. The samples were run on a 10–12% polyacrylamide urea gel, dried and exposed to a phosphoimager screen and scanned in a Typhoon 9400 system. Termination probability was calculated as previously described [25]. Briefly, at template position N, the termination probability was determined by the intensity of the band at N and divided by the intensity at position N plus the intensity at bands for longer products, at N, = [N] / ≥ [N].
The sequencing of in vitro synthesized DNA was performed as described [51] except that the primer extension reactions were performed on a 70 nt template containing either a dNMP or rNMP at position 30. As control template, a 70 nt oligo with 8-oxo-G at position 30 was used. The template was primed with a biotinylated oligonucleotide with a HindIII site (S1 Table). Reactions were stopped by incubation at 70°C for 1 h. Following the manufacturer’s instructions, the bypass products were immobilized on Dynabeads M-280 Streptavidin for 15 min at room temperature. The two strands were denaturated two times in 0.1 M NaOH for 5 min. The single-stranded product was washed according to the manufacturer’s instructions and amplified by high fidelity PCR using Phusion High-fidelity DNA polymerase (NEB) to generate a double-stranded 104 bp product. The fragment was cleaved using BfaI and HindIII restriction enzymes and cloned into pUC19. The ligation was transformed into E. coli TOP10 and colonies were sequenced with the M13 (-49) primer.
|
10.1371/journal.pcbi.1007290 | Estimating information in time-varying signals | Across diverse biological systems—ranging from neural networks to intracellular signaling and genetic regulatory networks—the information about changes in the environment is frequently encoded in the full temporal dynamics of the network nodes. A pressing data-analysis challenge has thus been to efficiently estimate the amount of information that these dynamics convey from experimental data. Here we develop and evaluate decoding-based estimation methods to lower bound the mutual information about a finite set of inputs, encoded in single-cell high-dimensional time series data. For biological reaction networks governed by the chemical Master equation, we derive model-based information approximations and analytical upper bounds, against which we benchmark our proposed model-free decoding estimators. In contrast to the frequently-used k-nearest-neighbor estimator, decoding-based estimators robustly extract a large fraction of the available information from high-dimensional trajectories with a realistic number of data samples. We apply these estimators to previously published data on Erk and Ca2+ signaling in mammalian cells and to yeast stress-response, and find that substantial amount of information about environmental state can be encoded by non-trivial response statistics even in stationary signals. We argue that these single-cell, decoding-based information estimates, rather than the commonly-used tests for significant differences between selected population response statistics, provide a proper and unbiased measure for the performance of biological signaling networks.
| Cells represent changes in their own state or in the state of their environment by temporally varying the concentrations of intracellular signaling molecules, mimicking in a simple chemical context the way we humans represent our thoughts and observations through temporally varying patterns of sounds that constitute speech. These time-varying concentrations are used as signals to regulate downstream molecular processes, to mount appropriate cellular responses for the environmental challenges, or to communicate with nearby cells. But how precise and unambiguous is such chemical communication, in theory and in data? On the one hand, intuition tells us that many possible environmental changes could be represented by variation in concentration patterns of multiple signaling chemicals; on the other, we know that chemical signals are inherently noisy at the molecular scale. Here we develop data analysis methodology that allows us to pose and answer these questions rigorously. Our decoding-based information estimators, which we test on simulated and real data from yeast and mammalian cells, measure how precisely individual cells can detect and report environmental changes, without making assumptions about the structure of the chemical communication and using only the amounts of data that is typically available in today’s experiments.
| For their survival, reproduction, and differentiation, cells depend on their ability to respond and adapt to continually changing environmental conditions. Environmental information must be sensed and often transduced to the nucleus, where an appropriate response is initiated, usually by selectively up- or down-regulating the expression levels of target genes. This information flow is mediated by biochemical reaction networks, in which concentrations of various signaling molecules code for different environmental states or different response programs. This map between environmental input or response output and the internal chemical state is, however, highly stochastic, because typical networks operate with small absolute copy numbers of signaling molecules [1]; moreover, different environments can be encoded by the same signaling molecule, by differentially regulating the dynamics of its concentration [2]. This raises two fundamental questions: first, how much information the cells could, even in principle, encode in the combinatorial and possibly time-varying concentrations of multiple signaling molecules and how such information could be plausibly read out during “downstream” processing; and second, how can we quantify, in an unbiased and model-free fashion, the amount of information available to the cells from limited experimental data.
Information theory provides a framework within which the theoretical study of limits to communication as well as the empirical study of actual information flows can be addressed [3]. Applications of information theory to questions in biology and, in particular, neuroscience started already in the 1950s and continue to this day, with the main focus to understand how—and with what accuracy—neural activity encodes information about the environment [4–6]. Applications of analogous techniques to biochemical signaling only started recently and represent an active area of research at the interface of physics, biology, statistics, and engineering [7–10].
Recent theoretical work analyzed the reliability of information transmission through specific reaction systems in the presence of molecular noise, e.g., during ligand binding [11], in chemotaxis [12], gene regulation [13–19], biochemical signaling networks [20], etc., and asked how such transmission can be maximized by tuning the reaction rates. Generally, these studies focused on steady state, by considering the information encoded in a single temporal snapshot of the reaction network at equilibrium given the input signals. Rigorous extensions to dynamical signals have been either rare and only possible for simple cases, like the BIND channel [11], or required specific operating regimes that permitted linearization and Gaussianity assumptions [12, 21, 22]. At its core, the analysis of signal transduction through nonlinear noisy chemical systems requires one to have control over the distribution of concentration trajectories given the (possibly) time-varying inputs; even if it were possible to calculate this distribution in principle, the curse of dimensionality puts strong limits to the manipulations required to compute the information transmission. Consequently, problems of this kind are currently considered intractable in their full generality.
Empirical estimates of information transmission in biochemical networks similarly focused on the steady state [23, 24], or considered only specific, hand-picked dynamical features, such as the amplitude or the frequency of the response, as information carriers [25]. Recent developments of fluorescent reporters and microfluidics have enabled the characterization of dynamical responses at a single cell resolution using large (> 104) numbers of sampled response trajectories, thereby permitting direct information estimates using generic estimators like the k-nearest-neighbors (knn) [26]. Existing approaches, however, suffer from severe limitations: they still require a prohibitive number of samples, especially when the response is distributed over multiple chemical species; or they necessitate uncontrolled assumptions about trajectory features that are supposed to be “relevant”. We recently proposed and applied decoding-based information estimators [27] as an alternative that draws on the past experiences in neuroscience [28–30] to dissect the yeast stress-response network. In this paper we provide a detailed account of the new methodology, show that it alleviates the most pressing problems of existing approaches, and benchmark it against synthetic and real data.
At their core, cellular processes consist of networks of chemical reactions. A chemical reaction network consists of a set of m molecular species {X˜1,X˜2,…,X˜m} that interact through K coupled reactions of the form:
ν1k′X˜1+…+νmk′X˜m→θkν1k″X˜1+…+νmk″X˜m,k=1,…,K (1)
where ν1k′,…,νmk′ and ν1k″,…,νmk″ are coefficients that determine how many molecules of each species are consumed and produced in the k-th reaction. θ1…θk∈R+ determine the rates at which the reactions occur and depend on binding affinities of chemical species, temperature and possibly the external conditions.
If we assume that the system is well-stirred, in thermal equilibrium and the reaction volume is constant, it can be shown that the probability that a reaction of type k takes place in an infinitesimal time interval [t, t + dt] can be written as ak(x˜,θk)dt=θkgk(x˜)dt, where x˜=[x˜1,…,x˜m]T∈N0m contains the amounts of molecules of the m species that are present in the system at time t, and gk(x˜)=∏i=1m(x˜iνik′) counts all possibilities of choosing the required reaction molecules out of all available molecules [31, 32]. θk is a constant that depends on the physical characteristics of the cell but also on the environmental conditions.
Let us denote the probability that x˜ molecules of the m species are present in the system at time t∈R+ by p(x˜,t) and define the stoichiometric change vectors νk=[ν1k,…,νmk]T∈Zm,k=1,…,K, as the net changes in the amount of molecules in the reactions, i.e. νik=νik″−νik′,i=1,…,m,k=1,…,K. Then it can be shown [32] that the chemical master equation (CME) can be written as:
p˙(x˜,t)=−p(x˜,t)∑k=1Kak(x˜,θk)+∑k=1Kp(x˜−νk,t)ak(x˜−νk,θk), (2)
or in a more compact form [32]
p˙(t)=Mp(t), (3)
where p(t) is a vector with components p(x˜,t), which is, in principle, infinite dimensional, and M contains the transition rates between all possible states, e.g. the transition rate from state x˜k′=x˜−νk to state x˜ is given by
Mx˜,x˜k′=ak(x˜k′,θk)−δx˜,x˜k′∑qaq(x˜,θq), (4)
where δ is the Kronecker delta, which is 1 when x˜=x˜k′ and 0 otherwise.
The CME given in Eq (3) is an instance of a continuous-time discrete-state-space Markov Chain for a random process X that can be solved exactly only for a few simple cases. It is nevertheless possible to efficiently generate samples x of the random process X, which we will refer to as “trajectories” or “paths”, for a selected time interval, t ∈ [0, T], according to the correct probability distribution p, by the Stochastic Simulation Algorithm (SSA, or the Gillespie algorithm) [33].
To study information transmission through the biochemical networks described by the CME, we need to define the input and output signals. In the simplest setup considered here, the input U is a discrete random variable that can take on one of the q ≥ 2 possible values, U ∈ {u(1), u(2), …, u(q)}. Each input in general corresponds to a distinct set of reaction rate constants θ, but in models of real biological networks, changing input often modulates only one or a few rates in the system, e.g., by representing the change in a key external ligand concentration, receptor activity, etc. Changes in the input are reflected in changes in the resulting trajectories of chemical species amounts, x. Typically, only a subset of chemical species could be considered as biologically-relevant “outputs” that encode the information about the environmental change: this would correspond to marginalizing p in Eq (3) over the unobserved (non-output) chemical species for the purposes of information transmission. While this is an interesting theoretical problem in its own right, here we work with simple toy examples where the output will be the trajectory, x, over the complete state space, i.e., we assume that all chemical species in the reaction network can be fully and perfectly observed. As we explain below, this allows us to define and compute the mutual information between a discrete input, U, and the output random process X given by the CME in a straightforward fashion. We later show that this computation can be carried out also when the continuous-time process X is sampled at uniform discrete times, as would be the case with experimental measurements.
Information theory introduces the mutual information as the measure of fidelity by which changes in one random variable, e.g., the input U, can effect changes in another random variable, e.g., X. In this sense, mutual information is simply a measure of statistical dependency (i.e., any correlation, be it linear or not) between U and X, and can thus be written as a functional of the joint probability density function p(x, u):
I(X;U)=∫X∫Up(x,u)log2(p(x,u)pX(x)pU(u))dudx (5)
where pU and pX are the marginal density functions for U and X, respectively, and we have generically written u and x as continuous variables; if they are discrete, integral signs are replaced by summations over the support for the corresponding probability distributions, as appropriate.
Mutual information is a non-negative symmetric quantity that is measured in bits, and is zero only if X and U are statistically independent. When studying information transmission through a channel U → X specified by p(x|u), for which U serve as inputs drawn from an input distribution pU(u), it is common to rewrite Eq (5) as
I(X;U)=H(U)−H(U|X)=H(X)−H(X|U), (6)
where H(X) is the differential entropy of X (and analogously for H(U)), defined as
H(X)=−∫XpX(x)log2pX(x)dx, (7)
and the conditional entropy, H(X|U), is
H(X|U)=∫UH(X|u)pU(u)du=−∫U∫XpU(u)p(x|u)log2p(x|u)dxdu. (8)
Eq (6) can be interpreted in two ways: information is either the difference between the total variability in the repertoire of responses X that the biochemical network can generate (measured by the response entropy, H(X)) and the average variability at fixed input that is due to noise in the network (measured by the noise entropy, H(X|U)); alternatively, information is also the entropy of the inputs, H(U), minus equivocation H(U|X), or the average uncertainty in what input was sent given that a particular response was observed. These interpretations make explicit the dependence of information both on the properties of the channel (the biochemical reaction network), as well as on the distribution of signals pU that the network receives. In this work, we will consider discrete inputs and will assume uniform pU. It is, however, also possible to compute the channel capacity C by maximizing the information flow at given p(x|u) over all possible input distributions,
C=maxpUI(X;U); (9)
Shannon’s classic work then proves that error-free transmission at rates higher than those given by capacity is impossible, while error-free transmission at rates below capacity can be achieved with the optimal use of the channel. Contrary to engineering, where the focus is on finding encoding and decoding schemes that best utilize a given channel, in biophysics and systems biology mutual information is used as a tool to quantify the limits to biological signal processing due to noise without needing to make assumptions about possible biochemical encoding and decoding mechanisms.
The setup we consider here is one in which inputs U are drawn independently from a uniform distribution and change rarely, i.e., at a rate that is much lower than the (inverse) timescale on which the reaction network in Eq (1) relaxes to its steady state. We assume that after an input change, we observe a fixed-time segment of the complete network dynamics, x, which is a sample path in m-dimensional discrete space, making direct calculation of information, I(X; U), by integrating / summing over all possible trajectories as implied by Eq (5) intractable. We will nevertheless show that estimates of exact information are possible if the reaction network is known, by explicitly using the transition matrix M of the Markov Chain from Eq (3) and generating exact sample paths, that is, realizations of X, using SSA. We call this model-based approach exact Monte Carlo approximation and contrast it to uncontrolled model-free estimations such as those obtained by using Gaussian approximations or k-nearest-neighbors methodology. We then introduce various decoding estimators and establish a hierarchy through which these estimates upper and lower-bound the true information, as shown in Fig 1.
In the absence of a full stochastic model for the biochemical reaction network, mutual information estimation is tractable only if we make assumptions about the distribution of response trajectories given the input. We briefly summarize two approaches below: in the first, k-nearest-neighbor procedure, the space in which the response trajectories are embedded is assumed to have a particular metric; in the second, Gaussian approximation, we assume a particularly tractable functional form for the channel, p(x|u).
Here and in the next section we introduce a class of decoding-based calculations that lower-bound the exact information, I(X; U), and can tractably be used as information estimators over realistically-sized data sets. Let D consist of a set of N labeled paths, typically represented in discretely sampled time, D={(u1,x1),(u2,x2),…,(uN,xN)}, where ui and xi, for i = 1, …, N, are realizations of the random variables U ∈ {u(1), …, u(q)} and X∈Rm×d, respectively. Here, D can represent either real data (typically containing N ∼ 102 − 103 trajectories) in case of model-free information estimates, or trajectories generated by exact simulation algorithms (in which case the sample size, N, is not limiting) from the full specification of the biochemical reaction network in case of model-based approximations.
The procedure of estimating the input u^ from x, such that the estimated u^ is “as close as possible” to true u for a given trajectory x, is known as decoding in information theory and neuroscience, and can equivalently be viewed as a classification task in machine learning or as an inference task in statistics. This procedure is implemented by a decoding function,
u^=Fω(x); (22)
F is typically parametrized by parameters ω that need to be learned from data for model-free approaches, or derived from biochemical reaction network specification in case of model-based approaches. F assigns to every xi in the dataset a corresponding “decode” u^i from the same space over which the random variable U is defined; formally, these decodes are instances of a new random variable U^. The key idea of using decoding for information estimation starts with the observation that random variables
U→X→TdX→FωU^, (23)
where Td represents time discretization, form a Markov chain. In other words, the distribution of U^ is conditionally independent of U and only depends on X, p(u^|x,u)=p(u^|x), and so
p(u^,x,u)=pU(u)p(x|u)p(u^|x). (24)
The data processing inequality [43] can be used to further extend the bounds in Eq (18):
I(U;U^)≤Iexact(U;X)≤Iexact*(U;X), (25)
where equality between the first two terms holds only if I(U;X|U^)=0. Consequently, I(U;U^) is a lower bound to the information between trajectories X and the input U [44]. Note that analogous reasoning holds for decoding directly from continuous-time trajectories X. Better decoders which increase the correspondence between the true inputs and the corresponding decoded inputs will typically provide a tighter lower bound on the information.
To compute the information lower bound, we apply the decoding function to each trajectory in D in model-based approximations or to each trajectory in the testing dataset for model-free estimators that need to be learned over training data first. We subsequently construct a q × q confusion matrix, also known as an error matrix, where each element ϵij counts the fraction of realizations of x generated by an input u = u(i) that decode into u^=u(j). This matrix provides an empirical estimate of the probability distribution p(u^,u), which can thus be used to compute the information estimate:
I(U^;U)=∑u,u^p(u^,u)log2p(u^,u)pU(u)pU^(u^)≈∑i=1q∑j=1qϵijlog2ϵij(∑kϵkj)(∑lϵil), (26)
Crucially, in this estimation O(N) data points are used to empirically estimate the elements of a q × q matrix ϵ, and information estimation involves a tractable summation over these matrix elements; in contrast, direct estimates of I(U; X) would involve an intractable summation over (vastly undersampled) space for X. For typical applications where q is small, decoding thus provides an essential dimensionality reduction prior to information estimation: in a simple but biologically relevant case of two distinct stimuli (q = 2), information estimation only requires us to empirically construct a 2 × 2 confusion matrix. If required, one can apply well-known debiasing techniques for larger q [5].
We start by considering three simple chemical reaction networks for which we can obtain exact information values using the model-based approach outlined in Section Exact information calculations for fully observed reaction networks. This will allow us to precisely assess the performance of decoding-based model-free estimates, and systematically study the effects of time discretization, the number of sample trajectories, and the number of distinct discrete inputs, q.
The three examples are all instances of a simple molecular birth-death process, where molecules of X˜ are created and destroyed with rates α and β, respectively:
→α(U)X˜→β(U)⌀. (36)
The reaction rates, α and β, will depend in various ways on the input, U, and possibly time, as specified below. Given an initial condition, x(t = 0), the production and degradation reactions generate continuous-time stochastic trajectories, x(t), recording the number of molecules of X˜ at every time t ∈ [0, T], according to the Chemical Master Eq (3). These trajectories, or their discretized representations, are considered as the “outputs” of the example reaction networks, defining the mutual information I(X; U) that we wish to compute. In all three examples we start with the simplest case, where the random variable U can only take on two possible values, u(1) and u(2), with equal probability, pU(u(1)) = pU(u(2)) = 0.5.
Example 1. In this case, x(t = 0) = 0, β = 0.01, independent of the input U, and the production rate depends on the input as α(u(1)) = 0.1, α(u(2)) = 0.07. Here, the steady state is given by Poisson distribution with mean number of molecules 〈x(t → ∞)〉 = α/β. Steady-state is approached exponentially with the timescale that is the inverse of the degradation rate, β−1. These dynamics stylize a class of frequently observed biochemical responses where the steady-state mean expression level encodes the relevant input value. Even if the stochastic trajectories for the two possible inputs are noisy as shown in Fig 2A, we expect that the mutual information will climb quickly with the duration of the trajectory, T, since (especially in steady state) more samples provide direct evidence about the relevant input already at the level of the mean trajectories.
Example 2. In this case, x(t = 0) = 0, β = 0.01, independent of the input U, and the production rate depends on the input as α(u(1), t) = 0.1, α(u(2), t) = 0.05 for all t < 1000, while for t ≥ 1000 the production rate is very small and independent of input, α(u, t) = 5 ⋅ 10−4. In the early period, this network approaches input-dependent steady state with means whose differences are larger than in Example 1, but the difference decays away for t > 1000 as the network settles towards vanishingly small activity for both inputs, as shown in Fig 2B. These dynamics stylize a class of transient biochemical responses that are adapted away even if the input state persists. In this case, lengthening the observation window T will not provide significant increases in information.
Example 3. In this case, x(t = 0) = 10. All reaction rates depend on the input, α(u(1)) = 0.1, α(u(2)) = 0.05, β(u(1)) = 0.01, β(u(2)) = 0.005, and are chosen so that the mean 〈x(t)〉 = 10 is constant across time and equal for both conditions, as shown in Fig 2C. In this difficult case, inputs cannot be decoded at the level of mean responses but require sensitivity to at least second-order statistics of the trajectories. Specifically, signatures of the input are present in the autocorrelation function for x: the timescale of fluctuations and mean-reversion is two-fold faster for u1 than u2. While this case is not frequently observed in biological systems, it represents a scenario where, by construction, no information about the input is present at the level of single concentration values and having access to the trajectories is essential. Because there is no difference in the mean response, we expect linear decoding methods to provide zero bits of information about the input. This case is also interesting because of the recent focus on pulsatile stationary-state dynamics in biochemical networks [53]. These pulses, reported for transcription factors such as Msn2, NF-κB, p53, etc., occur stochastically and, when averaged over a population of desynchronized cells, can yield a flat and featureless mean response. Information about the stimulus could, nevertheless, be encoded in either the frequency, amplitude, or other shape parameters of the pulses. While a generative description of such pulsatile dynamics goes beyond a birth-death process considered here, from the viewpoint of decoding, both pulsatile signaling and our example present an analogous problem, where the mean response is not informative about the applied input.
Before proceeding, we note that our examples are not intended to be realistic models of intracellular biochemical networks, but are chosen here for their simplicity and analytical tractability, in order to benchmark model-free estimators against known “gold truth” standard. In particular, while our examples include intrinsic noise due to the stochasticity of biochemical reactions at low concentration, they do not include extrinsic noise or cell-to-cell variability which, in some systems, is known to importantly or even dominantly contribute to the total variability in the response [26, 54]. The presence of such additional sources of variation by no means makes the model-free estimators inapplicable, as we show in S1 Fig where we study estimator performance in the simplest Example 1 model that includes cell-to-cell variability; it solely prevents us from comparing their performance to a tractably-computable MAP decoder result.
To illustrate the use of our estimators in a realistic context, we analyzed data from two previously published papers. The first paper focused on the representation of environmental stress in the nuclear localization dynamics of several transcription factors (here we focus on data for Msn2, Dot6, and Sfp1) in budding yeast [27]. The second paper studied information transmission in biochemical signaling networks in mammalian cells (here we focus on data for ERK and Ca2+) [26]. In both cases, single-cell trajectory data were collected in hundreds or thousands of single cells sampled at sufficient resolution to represent the trajectories discretized at tens to hundreds of timepoints. Similarly, both papers estimate the information transmission in trajectories about a discrete number of environmental conditions: Ref [27] uses the linear SVM approach presented here, while Ref [26] uses the knn estimator. This makes the two datasets perfectly suited for estimator comparisons. We further note that in both datasets the trajectories can be divided into two response periods: the early “transient” response period when the external condition changes, and the late “near steady-state” response period. Typically, the transient dynamics exhibit clear differences in the trajectory means between various conditions, reminiscent of our Example 1 or early Example 2; in contrast, in the late period the response may have been adapted away, or the stimulus could be encoded only in higher-order statistics of the traces, reminiscent of the late period in Example 2 or Example 3.
Fig 8 shows the raw data and summarizes our estimation results for the early and late response periods for the three translocating factors in yeast that report on the change from 2% glucose rich medium to 0.1% glucose poor stress medium. Fig 9 similarly shows the raw data and estimation results for the early and late response periods for the signaling molecules in mammalian cells responding to multilevel inputs.
Consistent with the published report [27], transient response in yeast nuclear localization signal can be decoded well with the linear SVM estimator that yields about 0.6 bits of information per gene about the external condition. Kernelized SVM outperforms the linear method slightly by extracting an extra 0.1-0.2 bits of information, while knn underperforms the linear method significantly for Msn2 and Dot6 (but not for Sfp1). The Gaussian decoder estimate shows a mixed performance and the neural network estimate is the worst performer, most likely because the number of samples here is only N = 100 per input condition and neural network training is significantly impacted.
It is interesting to look at the stationary responses in yeast which have not previously been analyzed in detail. First, low estimates provided by linear SVM for Msn2 and Dot6 imply that information in the stationary regime, if present, cannot be extracted by the linear classifier. Second, the Gaussian decoder also performs poorly in the stationary regime, potentially indicating that the relevant features are encoded in higher-than-pairwise order statistics of the response (e.g., pulses could be “sparse” features as in sparse coding [58]); it is, however, hard to exclude small number of training samples as the explanation for the poor performance of the Gaussian decoder. Third, K-nearest-neighbor estimator also yields low estimates, either due to small sample number or low signal-to-noise ratio, the regime for which knn method has been observed to show reduced performance [40]. A particularly worrying feature of the knn estimates is their non-robust dependence on the length of the trajectory T. As S6 Fig shows, the performance of knn peaks at T ≈ 50 min and then drops, even well into unrealistic negative estimates for T ≈ 400 min (corresponding to the highest dimensionality d = 170 of discrete trajectories). While it is possible to make an ad hoc choice to always select trajectory duration at which the estimate peaks, the performance of kernelized SVM is, in comparison, extremely well behaved and increases monotonically with T, as theoretically expected. Finally, nonlinear SVM estimator extracts up to 0.4 bits of information about condition per gene, more than half of the information in the early transient period. This is even though on average the response trajectories for the two conditions, 2% glucose and 0.1% glucose, for Msn2 and Dot6 are nearly identical. For Sfp1 there is a notable difference in the mean response, which the linear estimator can use to provide a ∼ 0.15 bits of information, yet still significantly below ∼ 0.4 bits extracted by the nonlinear SVM. For both transient and stationary responses in yeast, our results are qualitatively in line with the expectations from the synthetic example cases—given the small number of trajectories, tightest and most robust estimates are provided by the decoding information estimator based on nonlinear (kernelized) SVM. Regardless of the decoding methodology and even without small sample corrections at N = 100 trajectories per input, our estimates are not significantly impacted by the well-known information estimation biases thanks to the dimensionality reduction that decoding provides by mapping high dimensional trajectories X back into the space for inputs U which is low dimensional; this is verified in S7 Fig by estimating the (zero) information in trajectories whose input labels have been randomly assigned.
Random pulses that encode stationary environmental signals have been observed for at least 10 transcription factors in yeast [53] and for tens of transcription factors in mammalian cells [59]. Recent studies investigated the role of the pulsatile dynamics in cellular decision-making [57, 60]. Nevertheless, methods for quantifying the information encoded in stochastic pulses are still in their infancy. Our nonlinear SVM decoding estimates convincingly show that there is information to be learned at the single cell level from the stationary stochastic pulsing. An interesting direction for future work is to ask whether hand-crafted features of the response trajectories (pulse frequency, amplitude, shape, etc) can extract as much information from the trajectories as the generic SVM classifier: for that, one would construct for each response trajectory a “feature vector” by hand, compute the linear SVM decoding bound information estimate from the feature vectors, and compare that to the kernelized SVM estimate over the original trajectories. This approach is a generic and operationally-defined path for finding “sufficient statistics” of the response trajectories—or a compression of the original signal to the relevant set of features—in the information-theoretic sense.
A different picture emerges from the mammalian signaling network data shown in Fig 9. The key difference here is the order of magnitude larger number of sample trajectories per condition compared to yeast data. Most of the information seems linearly separable in both the early and late response periods, as evidenced by the success of the linear SVM based estimator whose performance is not improved upon by the kernelized SVM (indeed, for early ERK response period linear SVM gives a slightly higher estimate than the nonlinear version). The big winner on this dataset is the neural-network-based estimator that yields the best performance in all conditions among the decoding-based estimators, likely owning to sufficient training data. As before, the Gaussian decoder shows mixed performance which can get competitive with the best estimators under some conditions. Lastly, knn appears to do well except on the late Ca2+ data (perhaps due to low signal-to-noise ratio). It also shows counter-intuitive non-monotonic behavior with trajectory duration T in S8 Fig (cf. with Fig 2C of Ref [27], where the analysis of information conveyed in dynamical signals as a function of trajectory duration was also very revealing about signaling in yeast). Once again it is worth keeping in mind that knn is estimating the full mutual information which could be higher than the information decodable from single responses.
Increasing availability of single-cell time-resolved data should allow us to address open questions regarding the amount of information about the external world that is available in the time-varying concentrations, activation or localization patterns, and modification state of various biochemical molecules. Do full response trajectories provide more information than single temporal snapshots, as early studies suggest? Is this information gain purely due to noise averaging enabled by observing multiple snapshots, or—more interestingly—due to the ability of these intrinsically high-dimensional signals to provide a richer representation of the cellular environment? Can we isolate biologically relevant features of the response trajectories, e.g., amplitude, frequency, pulse shape, relative phase or timing, without a priori assuming what these features are? How can cells read out the environmental state from these response trajectories and how close to the information-theoretic bounds is this readout process? More broadly, a framework for analyzing complete response trajectories in signaling or genetic regulatory networks at the single cell level could lead to architectural and functional constraints on the biological network [27], and allow us to further pursue the ideas of optimal information representation in biological systems [9].
Here, we made methodological steps towards answering these questions by focusing on two related problems: first, if we are given a full stochastic description of a biochemical reaction network, under what conditions can we theoretically compute information transmission through this network and various related bounds; second, if we are given real data with no description of the network, what are tractable schemes to estimate the information transmission. We show that when the complete state of the reaction network is observed and the inputs are discrete sets of reaction rates, there exist tractable Monte Carlo approximation schemes for the information transmission. These exact results that we compute for three simple biological network examples then serve to benchmark a family of decoding-based model-free estimators and compare their performance to the commonly-used knn estimator. We show that decoding-based estimators can closely approach the optimal decoder performance and in many cases perform better than knn, especially with typical problem dimensions (d ∼ 1 − 100) and typical number of sample trajectories (N ∼ 102 − 103). This is especially true when we ask about the combinatorial representation of the environmental state in the time trajectories of several jointly observed chemical species, as in our previous work [27], where alternative information estimation methods usually completely fail due to the high dimensionality of the input space.
It is necessary to emphasize the flexibility of the decoding approach: decoding-based information estimation is based directly on the statistical problems of classification (for discrete input variable, U) or regression (for continuous input variable, U), so any classification / regression algorithm with good performance can provide the basis for information estimation. Concretely, for problems in the low data regime (small N), linear or kernelized SVM approaches appear powerful, while at larger N neural-network-based schemes can provide a better performance and thus typically a tighter information lower bound. In contrast to information approximations for which it is often impossible to assess their precision or bias (or even its sign) when the dimension, d, of the problem is large, the decoding approach yields a conservative estimate of the true information. Statistical algorithms underlying decoding-based estimations have the extra advantage that, (i), we may be able to gain biological insight by inspecting which features of the response carry the relevant stimulus information (e.g., by looking at the linear kernels or features that neural networks extract in their various layers); (ii), pick a decoding algorithm based on features previously reported as relevant (e.g., the Gaussian decoder for second-order statistics as in Example 3); (iii), estimate the information as a function of trajectory duration; and (iv), gain confidence in our estimates by testing their performance on withheld data. While we tested these estimators on a very restricted set of toy examples in order to be able to compare to analytically computed results, model-free decoding-based approaches are applicable more generally, e.g., to complex, partially-observed reaction systems, or networks with significant contribution of cell-to-cell variability or extrinsic noise.
By construction, decoding-based estimators only provide a lower bound to the true information. This, however, could turn out to be a smaller problem in practice than it appears in theory, especially for biochemical reaction networks. First, our extension to the Feder-Merhav bound provides us with an estimate of how large the gap between the true information and the decoded estimation can be. The bound is not tight on our examples, and can only be applied when the optimal MAP decoder can be constructed [61, 62]. Second, and perhaps more importantly, information that can be decoded after single input presentations is the quantity that is likely more biologically relevant than the true channel capacity, if the organisms are under constraint to respond to the environmental changes quickly. Typically, organisms across the complexity scale operate under speed-accuracy tradeoffs [63]: faster decisions based on noisy information lead to more errors and, conversely, with enough time to integrate sensory information errors can be reduced. When speed is at a premium or relevant inputs are sparse, decisions need to be taken after single input presentations. In this case, decoding-based estimation should not be viewed as an approximate but rather as the correct methodology for the biological problem at hand. Of course, there is still the question of whether the model-free decoders that we use on real data can achieve a performance that is close to the optimal MAP decoder that represents the absolute performance limit. While there is no general way to answer this question, it appears that simple SVM decoding schemes work well when the response trajectories differ in their conditional mean, and neural networks as general approximators can be used to check for more complicated encoding features when data is plentiful. Unlike in neuroscience, there is much less clarity about what kind of read-out or decoding operations biochemical networks can mechanistically realize to mimic the functioning of our in silico decoders, and it may be challenging to biochemically implement even arbitrary linear classification of response trajectories. Until experimentally shown otherwise, it thus appears reasonable to proceed with the assumption that environmental signals can be read out from the time-dependent internal chemical state with a simple repertoire of computations.
We also mention a caveat when using decoding-based estimators that rely on classification or regression methods with large expressive power, such as neural networks. While it is possible to successfully guard against overfitting within the same dataset using cross-validation, scientific insights into biological function often require generalization beyond one particular dataset. Typically, we ask for generalization at least over independent experimental replicates, but sometimes even over similar (but not same) external conditions, strains, or experimental setups. This can present a serious issue if e.g., neural networks overfit to such systematic variations between replicates or conditions even when such variations are not biologically relevant. Regularization alone will not necessarily guard against this, unless the networks are actually trained over a subset of all data on which they will be tested. A pertinent recommendation here is to evaluate the difference in performance of expressive decoding-based estimators when trained over a subset or over all replicates, and to compare that to the generalization of less-expressive methods for which the sufficient statistics are known (e.g., linear or Gaussian decoders).
We conclude by emphasizing a simple yet important point. The decoding-based approach that we introduced here should also motivate us to look beyond methodological problems of significance and estimation, to truly biological problems of cellular decision making. Currently, data on biological regulatory processes is often analyzed by looking for “statistically significant differences” in the network response for, say, two possible network inputs. For example, one may report that the steady-state mean expression level of a certain gene is significantly larger in the stimulated vs unstimulated condition, with the statistical significance of the mean difference established through an appropriate statistical test that takes into account the number of collected population samples. While statistical significance is a necessary condition to validly report any difference in the response, it is very different from the question of whether a single cell could discriminate the two conditions given access only to its own expression levels. In caricature, population-level statistics tell us with what confidence we, as scientists having access to N samples, can discriminate between conditions given some biological readout; decoding based information estimates, on the other hand, are relevant to the N = 1 case of individual cells. We hope that further work along the latter path can clarify and quantify better the difficult constraints and conditions under which real cells need to act based on individual noisy readouts of their stochastic biochemistry.
|
10.1371/journal.ppat.1006423 | Cooperative inhibition of RIP1-mediated NF-κB signaling by cytomegalovirus-encoded deubiquitinase and inactive homolog of cellular ribonucleotide reductase large subunit | Several viruses have been found to encode a deubiquitinating protease (DUB). These viral DUBs are proposed to play a role in regulating innate immune or inflammatory signaling. In human cytomegalovirus (HCMV), the largest tegument protein encoded by UL48 contains DUB activity, but its cellular targets are not known. Here, we show that UL48 and UL45, an HCMV-encoded inactive homolog of cellular ribonucleotide reductase (RNR) large subunit (R1), target receptor-interacting protein kinase 1 (RIP1) to inhibit NF-κB signaling. Transfection assays showed that UL48 and UL45, which binds to UL48, interact with RIP1 and that UL48 DUB activity and UL45 cooperatively suppress RIP1-mediated NF-κB activation. The growth of UL45-null mutant virus was slightly impaired with showing reduced accumulation of viral late proteins. Analysis of a recombinant virus expressing HA-UL45 showed that UL45 interacts with both UL48 and RIP1 during virus infection. Infection with the mutant viruses also revealed that UL48 DUB activity and UL45 inhibit TNFα-induced NF-κB activation at late times of infection. UL48 cleaved both K48- and K63-linked polyubiquitin chains of RIP1. Although UL45 alone did not affect RIP1 ubiquitination, it could enhance the UL48 activity to cleave RIP1 polyubiquitin chains. Consistently, UL45-null virus infection showed higher ubiquitination level of endogenous RIP1 than HA-UL45 virus infection at late times. Moreover, UL45 promoted the UL48-RIP1 interaction and re-localization of RIP1 to the UL48-containing virion assembly complex. The mouse cytomegalovirus (MCMV)-encoded DUB, M48, interacted with mouse RIP1 and M45, an MCMV homolog of UL45. Collectively, our data demonstrate that cytomegalovirus-encoded DUB and inactive R1 homolog target RIP1 and cooperatively inhibit RIP1-mediated NF-κB signaling at the late stages of HCMV infection.
| Activation of NF-κB signaling leads to expression of proinflammatory cytokines and chemokines and plays a key role in regulating innate immune response and inflammation to virus infection. HCMV upregulates and downregulates NF-κB signaling during the course of infection. Upregulation of NF-κB signaling may promote viral gene expression or viral dissemination, but its downregulation may be necessary for suppression of excessive immune responses. Recently, it was demonstrated that viral late functions downregulate TNFα- and IL-1β-induced NF-κB activation. However, the viral proteins involved and the underlying mechanisms are not understood. In the present study, we demonstrate that two HCMV proteins, the largest tegument protein harboring deubiquitinase activity and the inactive homolog of cellular ribonucleotide reductase large subunit, cooperatively inhibit RIP1-mediated NF-κB signaling at the late stages of infection. This study for the first time identified RIP1 as a substrate of viral deubiquitinase and highlights the importance of the negative regulation of NF-κB during virus infection.
| Human cytomegalovirus (HCMV), which belongs to the β-herpesvirus subfamily, typically causes asymptomatic infections in immunocompetent individuals. HCMV is ubiquitously but latently distributed throughout the world. However, infection of pregnant women often causes congenital infection, and reactivation from latent infection in immunocompromised individuals can lead to life-threatening complications [1]. HCMV is a large, enveloped, double-stranded DNA virus and its 235 kb genome encodes for at least 165 proteins [2]. A structural feature unique to herpesviruses is the presence of a protein layer, called the tegument, between the capsid and the envelope. Upon initial fusion of the viral envelope with the host cell plasma membrane, many of these tegument proteins are delivered into the cytoplasm and the nucleus, where they perform diverse functions including activation of viral gene transcription and antagonization of host intrinsic and innate immunity. Viral tegument proteins are also thought to be involved in capsid transport and virion egress [1, 3, 4].
The herpesvirus deubiquitinase (DUB) was first discovered as an N-terminal fragment of the 336 kDa UL36 tegument protein (also known as VP1/2) of herpes simplex virus-1 (HSV-1) [5]. This DUB domain is conserved in the UL36 equivalents of other herpesviruses [6]. Interestingly, the herpesvirus DUBs bear no structural homology to known eukaryotic DUBs, although the key amino acid residues in the active site are highly conserved [7]. The herpesvirus DUBs appear to play a key role in regulating innate immune and inflammatory signaling. HSV-1 UL36 and Kaposi’s sarcoma-associated herpesvirus (KSHV) ORF64 interact with and deubiquitinate TRAF3 and RIG-I, respectively, inhibiting IRF3 activation [8, 9]. Epstein-Barr virus (EBV) BPLF1 targets TRAF6 and inhibits its ubiquitination, downregulating NF-κB activation during lytic infection [10]. BPLF1 is also known to deubiquitinate PCNA and inhibit trans-lesion synthesis at DNA damage sites [11].
In HCMV, the UL48-encoded tegument protein is a homolog of HSV-1 UL36 and possesses DUB activity. The UL48 DUB was identified using a suicide substrate probe specific for ubiquitin-binding cysteine proteases in virus-infected cells [12]. This DUB, mapped to the first ~350 amino acids of the N-terminal region of pUL48, contains both ubiquitin-specific carboxyl-terminal hydrolase activity and isopeptidase activity that cleaves ubiquitin K11, K48, and K64 linkages [13, 14]. Mutations in active site residues (C24 and H162) completely abolish DUB activity, and the virus containing the UL48 (C24S) gene shows moderately reduced growth in culture cells, demonstrating that the DUB activity of HCMV can influence viral replication [14]. Like the UL36 DUBs of HSV-1 and Pseudorabies virus (PRV) [15, 16], the UL48 DUB has autocatalytic activity that regulates its own stability [17].The N-terminal DUB-containing region from UL48 is required for virion stability and efficient virus entry, although the associated DUB activity appears not to be critical [17]. Despite accumulating evidence that DUB activity and the DUB-containing region of UL48 are required for efficient viral growth, whether the UL48 DUB, like the equivalents of α and γ-herpesviruses, plays a role in regulation of innate immune or inflammatory signaling is not known yet. UL48 contains the nuclear localization signal (NLS) that is essential for viral growth [17, 18], suggesting a critical role of UL48 in the nucleus or in routing of the capsid to the nuclear pore as seen in infection with HSV-1 UL36 mutant virus [19, 20].
The nuclear factor-kappa B (NF-κB) transcription factors are critical regulators of host cell’s early responses to viral infection [21]. HCMV infection can induce NF-κB’s transcriptional activity, which subsequently drives expression of a number of different proinflammatory cytokines and chemokines. There is increasing evidence that HCMV can inhibit NF-κB signaling during lytic infection. This negative regulation may be necessary to suppress excessive immune responses that are detrimental to viral infection. Although viral downregulation of NF-κB signaling in the early stages of infection has been confirmed, it is believed that late virus functions are critical to suppress NF-κB signaling. UL26 is an early viral protein that initially localizes to the nucleus, but becomes cytoplasmic as the infection progresses and eventually localizes to virion assembly sites [22, 23]. UL26 is known to antagonize NF-κB activation induced by TNFα by attenuating IKK phosphorylation and subsequent IκBα degradation [24].
The HCMV UL45 gene encodes an inactive homolog of cellular ribonucleotide reductase (RNR) large subunit (R1), which is contained in the tegument. The function of UL45 still remains largely unknown. Analysis of the VR1814-derived bacmid (FIX-BAC) reconstituted recombinant virus showed that the UL45-null mutant virus normally grows in endothelial cells [25] but is defective in viral particle accumulation at low multiplicities of infection (MOI) and in spreading in fibroblasts [26]. The growth defect of UL45-null virus did not appear to result from a reduced dNTP supply [26]. In mouse CMV (MCMV), M45, a homolog of HCMV UL45, interacts with mouse receptor-interacting protein kinase 1 (mRIP1) and inhibits mRIP1-mediated signaling pathways, including activation of NF-κB after stimulation of TNFR1 and TLR3 [27]. M45 also blocks mRIP1-independent NF-κB activation and cytokine production after stimulation of IL-1R and TLRs and binds to and induces lysosomal degradation of NEMO by targeting it to autophagosomes [28]. Furthermore, M45 is also known to inhibit cytosolic DNA sensor DAI-mediated NF-κB activation through RIP homotypic interaction motif (RHIM)-dependent interaction [29]. Meanwhile, M45 acts as a suppressor of virus-induced or TNFα-induced cell death [30, 31]. M45 inhibits a caspase-independent form of programmed necrosis (necroptosis), which depends on the adaptor kinase RIP3 and DAI [32, 33]. It is not known whether UL45 has activity similar to M45 in regulating the host’s immune responses.
In this study, we identified RIP1 as a cellular target of HCMV-encoded deubiquitinase UL48. We provide evidence that RIP1 is also targeted by UL45, an HCMV-encoded inactive homolog of cellular RNR R1, and that UL48 and UL45 cooperatively suppress RIP1-mediated NF-κB signaling at the late stages of viral infection.
Since RIP1 is a critical mediator of NF-κB signaling and K63-linked polyubiquitination of RIP1 is an important process involved in this signaling, we tested whether the HCMV UL48 deubiquitinase targets RIP1. When CoIP assays were performed in co-transfected 293T cells, UL48 was found to interact with RIP1 (Fig 1A and 1B). RIP1 has three domains: the N-terminal kinase domain (KD), the central intermediate domain (ID), and the C-terminal death domain (DD) (Fig 1C). Three RIP1 deletion constructs (ΔKD, ΔID, and ΔDD) were employed for similar CoIP assays to determine the UL48-binding region of RIP1. The results revealed that both the intermediate and death domains of RIP1 are required for efficient RIP1 binding (Fig 1D and 1E). The intermediate domain contains the TNFα-induced Lys63 (K63)-linked polyubiquitination site (K377) [34] and the RHIM. To further investigate the effect of the RHIM and ubiquitination of RIP1 on the pUL48-RIP1 interaction, RHIM-deleted (ΔRHIM) and K377R mutants were also used in CoIP assays. The results showed that pUL48 still interacts with these ΔRHIM and K377R mutant RIP1 proteins (Fig 1F and 1G). Taken together, these results demonstrate that UL48 interacts with RIP1 through the intermediate and death domains, but this interaction does not require presence of the RHIM and polyubiquitination at K377 of RIP1.
Luciferase reporter assays were performed to investigate the effect of UL48 on RIP1-mediated NF-κB activation. UL48 expression effectively inhibited RIP1-mediated NF-κB activation in 293T cells (Fig 2A). HCMV permissive HF cells were more sensitive to apoptotic cell death induced by RIP1 overexpression than other cell types. Therefore, similar reporter assays were performed in HF cells with co-transfection of CrmA, an inhibitor of caspases 1 and 8. The results showed that the NF-κB activity induced by RIP1 expression was effectively inhibited by UL48 in HF cells (Fig 2B). TNFα treatment of HF cells moderately increased NF-κB activity, but this elevation was also inhibited by UL48 (Fig 2C). We next investigated whether the DUB catalytic activity of UL48 is required for this suppression using the C24S mutant in which the Cys24 active site was replaced with Ser [14]. The results of reporter assays showed that the inhibitory effect of UL48(C24S) on RIP1-mediated and TNFα-induced NF-κB activation was reduced to 30% to 60% that of wild-type UL48 (Fig 2D to 2F). The expression levels of wild-type and C24S mutant UL48 proteins were comparable and the RIP1 modification (probably ubiquitination) was reduced by wild-type UL48 but not by C24S mutant in transfected cells (S1A Fig). These results suggest that the DUB activity of UL48 is largely required for inhibition of RIP1-mediated NF-κB activation.
UL48 was first shown to interact with UL45, an inactive homolog of cellular RNR R1, in yeast two-hybrid interaction assays [35]. We also observed this interaction in CoIP assays (Fig 3A and 3B) [17]. Given that UL48 interacted with both RIP1 and UL45 and that M45, an MCMV-encoded homolog of UL45, interacted with mRIP1 [27, 31], the potential for interaction of UL45 with RIP1 was also tested. The results of CoIP assays revealed that UL45 interacts with RIP1 (Fig 3C and 3D).
CoIP assays using the mutant RIP1 constructs revealed that UL45 interacts with RIP1 through the intermediate domain (Fig 3E and 3F). The RHIM was reported to mediate the interaction between M45 and mRIP1 [31]. However, the UL45-RIP1 interaction did not seem to require the RHIM because UL45 did not contain an apparent RHIM, and a RHIM-deleted RIP1mutant also effectively bound to UL45 (Fig 3G). Furthermore, the K377R mutant RIP1 still interacted with UL45, demonstrating that the UL45-RIP1 interaction does not require polyubiquitination at K377 of RIP1 (Fig 3H). When UL45, UL48, and RIP1 were co-expressed, both UL48 and RIP1 were co-precipitated when UL45 was immunoprecipitated (Fig 3I). Similarly, UL45 and RIP1 were co-precipitated with UL48 (Fig 3J), and UL48 and UL45 were co-precipitated with RIP1 (Fig 3K). These results suggest the possibility that both UL48 and UL45 simultaneously target RIP1 and that they may form a complex.
We investigated whether, like UL48, UL45 inhibits RIP1-mediated NF-κB activation. In reporter assays performed in 293T and HF cells, NF-κB activity induced by RIP1 expression was effectively inhibited by UL45 (Fig 4A and 4B). UL45 also inhibited NF-κB activation induced by TNFα treatment in HF cells (Fig 4C). We further investigated whether UL48 and UL45 cooperatively affect RIP1-mediated NF-κB activation. In reporter assays, co-expression of UL48 and UL45 more effectively inhibited NF-κB activation by RIP1 than UL45 alone, and this cooperative effect of UL48 and UL45 was largely diminished when the UL48(C24S) mutant was used (Fig 4D and 4E; S1B Fig). This inhibitory effect of UL48 and UL45 was also observed in TNFα-induced NF-κB activation in HF cells (Fig 4F). Collectively, our data demonstrate that UL48 and UL45 may cooperate to inhibit RIP1-mediated NF-κB activation, and that the deubiquitinating activity of UL48 contributes to this repression.
To address the significance of the RIP1 targeting by UL48 and UL45 during HCMV infection, recombinant Toledo viruses that do not express UL45 (UL45-null) or express HA-tagged UL45 (HA-UL45) were produced using bacmid mutagenesis (S2 Fig). The growth curves of wild-type, UL45-null, and HA-UL45 viruses were compared in HF cells at an MOI of 0.1 or 2. The viruses released in the culture supernatants and associated with the cells were collected and pooled at various time points after infection, and titers of the infectious progeny virions were determined by infectious center assays. At an MOI of 0.1, the peak titers of UL45-null virus on day 7 was seven-fold lower than those of wild-type and HA-UL45 viruses (Fig 5A, left), while, at an MOI of 2, UL45-null virus produced about three-fold fewer progeny virions on day 5 than wild-type and HA-UL45 viruses (Fig 5A, right). The results of immunoblotting assays showed that the reduced growth of the UL45-null virus at low MOI correlated with the low level accumulation of viral late proteins, such as the late forms encoded by UL44 and the pp28 (encoded by UL99) true late protein (Fig 5B, left), while this difference was reduced at an MOI of 2 (Fig 5B, right). We also observed the reduced accumulation of the SUMO-modified forms of IE2 in UL45-null virus infection, which was often observed in mutant viruses with a moderate growth defect [14, 36]. Collectively, the analysis of UL45-null and HA-UL45 virus growth patterns suggests that UL45 contributes to viral growth in cultured cells at low MOI by affecting the late stages of the virus life cycle and that HA-UL45 virus has a similar growth pattern as wild-type virus.
Since each of UL48, UL45, and RIP1 was co-precipitated with the other two proteins in co-transfected cells, the possible formation of a complex containing these three proteins during viral infection was investigated using HA-UL45 virus. When HF cells were infected with HA-UL45 virus, immunoprecipitation of HA-UL45 co-precipitated both UL48 and RIP1 (Fig 6A) and immunoprecipitation of UL48 co-precipitated both HA-UL45 and RIP1 (Fig 6B). In a control CoIP experiment, however, immunoprecipitation of abundant pp65 viral protein did not co-precipitated UL48 and RIP1, although it unexpectedly pull-downed HA-UL45 (Fig 6C). These results suggest that RIP1 may exist in a protein complex containing UL48 and UL45. Furthermore, the result of gel filtration chromatography demonstrated that in HA-UL45 virus-infected cells, the RIP1 fractions shifted to higher molecular mass fractions compared to those in mock-infected cells, and that these high molecular mass fractions (>400 kDa) contained HA-UL45 and UL48 (Fig 6D). These results suggest that both UL48 and UL45 interact with RIP1, and a complex containing UL48, UL45, and RIP1 may be produced during HCMV infection.
We also investigated the localization patterns of UL48, HA-UL45, and RIP1 in cells infected with HA-UL45 virus. Human γ-globulin was used as a blocking agent for HCMV Fc receptors, which are recognized by rabbit antibodies in virus-infected cells at the late stages of infection (S3 Fig). When triple-label IFA was performed to visualize UL48, HA-UL45, and pp28 at 96 h after infection, UL48 and HA-UL45 were largely colocalized with pp28 in the cytoplasmic compartments, while a small fraction was also detected in the nucleus as punctate forms (Fig 7A). Since pp28 is localized in the cytoplasmic virion assembly complex (cVAC) [37], the association of UL48 and HA-UL45 with the cVAC was further investigated by co-staining GM130, a Golgi-associated cellular marker for the cVAC [38]. The results demonstrated that UL48 and HA-UL45 are largely colocalized with GM130 in the cVAC (Fig 7B). We next performed similar triple-label IFA to visualize UL48, HA-UL45, and RIP1 in virus-infected cells. In mock-infected cells, endogenous RIP1 was detected at very low levels throughout the cells as a diffuse form. However, in virus-infected cells, RIP1 was slightly stabilized and accumulated in the UL48 and HA-UL45-containing cVAC (Fig 7C). In two-color merge images, RIP1 appeared to be more effectively colocalized with HA-UL45 than UL48 (S4 Fig). These IFA data indicate that RIP1 is colocalized with UL48 and UL45 in the cVAC, supporting that UL48 and UL45 form a complex with RIP1 during viral infection.
RIP1 plays a key role in NF-κB signaling after TNFα treatment and HCMV inhibits NF-κB activation in the late stages of infection. Therefore, we investigated the effects of UL48 and UL45 on TNFα-induced NF-κB activation in the late stages of infection. We previously produced the DUB-defective UL48(C24S) mutant virus using the UL/b’ region-deleted Towne strain [14], while the UL45-null mutant in this study was produced in the Toledo strain. As a preliminary experiment, we compared the levels of NF-κB activation after TNFα treatment in cells infected with HCMV Toledo and Towne for different time periods. We found that phosphorylation of the NF-κB p65 subunit at serine residue 536 peaked 48 h after infection and began to decrease from 72 h in both Toledo and Towne virus-infected cells, while the phosphorylated p65 levels were higher during Toledo virus infection than during Towne virus infection (Fig 8A). The latter is consistent with the earlier reports demonstrating that the UL/b’ region in the Toledo virus and other clinical isolates sensitizes cells to TNFα signaling by upregulating cell surface expression of TNFR [39, 40].
Since UL45 and UL48 are expressed with early-late kinetics, their effects on TNFα-induced NF-κB activation were assessed 72 h after infection with wild-type or mutant viruses. In mock-infected cells, TNFα treatment for 5 or 15 min increased phosphorylation of p65 and IKKα/β. In Toledo virus infection, HA-UL45 virus less effectively upregulated TNFα-induced p65 and IKKα/β phosphorylation compared to UL45-null virus, demonstrating that UL45 expression inhibits TNFα-induced NF-κB activation (Fig 8B). In the case of Towne virus infection, wild-type virus resulted in less TNFα-induced p65 and IKKα/β phosphorylation compared to UL48(C24S) virus (Fig 8C). When we introduced the UL48(C24S) mutation in HA-UL45 Toledo virus, similar higher levels of p65 and IKKα/β phosphorylation were observed in UL48(C24S)-containing HA-UL45 virus compared to HA-UL45 virus (S5A and S5B Fig). In the experiments using UL48(C24S) mutant viruses, the levels of UL48(C24S) were lower than those of wild-type UL48. Therefore, it is not clear whether increased p65 and IKKα/β phosphorylation in mutant virus infection was due to the lack of DUB activity or due to the reduced protein level. However, since UL48(C24S) mutant protein that was expressed at a comparable level to wild-type protein largely lost the NF-κB repressing activity in reporter assays (see Figs 2D and S1A; Figs 4D and S1B), it is likely that the lack of UL48 DUB activity rather than the reduced level of UL48(C24S) largely contributed to the increase of TNFα-induced NF-κB activation. Supporting this idea, both HA-UL45 and RIP1 were still localized in the cVAC in HA-UL45/UL48(C24S) virus-infected cells (S5C Fig). Collectively, our analysis of UL45 and UL48 mutant viruses demonstrate that both UL45 and UL48 are involved in inhibition of TNFα-induced NF-κB activation in the late stages of infection.
The effect of UL45 and UL48 on TNFα-induced NF-κB activation was further investigated in HF cells co-transfected with UL45 and UL48 expression plasmids (Fig 9A). TNFα treatment increased phosphorylation of p65 in cells transfected with empty vector. However, TNFα-induced p65 phosphorylation was significantly impaired in cells transfected with UL45 or UL48. UL26, which was previously reported to antagonize NF-κB activation [24], also inhibited TNFα-induced p65 phosphorylation, but UL45 and UL48 did so more effectively, suggesting that UL45 and UL48 are potent inhibitors of NF-κB signaling.
To investigate the molecular mechanisms by which UL48 and UL45 inhibit TNFα-induced NF-κB activation, we assessed the effect of UL48 and UL45 expression on the levels of K63-linked polyubiquitination of RIP1. Co-transfection/ubiquitination assays were performed using plasmids expressing ubiquitin (K63 only) or ubiquitin (K48 only) in which all lysine (K) residues except K63 or K48 within ubiquitin are mutated to arginine to allow for the formation of only K63-linked or K48-linked ubiquitin chains. We found that UL48 inhibited RIP1 polyubiquitination through the K63 or K48 linkages, while UL45 did not affect RIP1 polyubiquitination (Fig 9B). When UL48 and UL45 were co-expressed, UL45 increased the activity of UL48 to cleave K63-linked polyubiquitin chains of RIP1 (Fig 9C). To address the role of UL45 in RIP1 ubiquitination in the context of viral infection, HF cells were infected with UL45-null or HA-UL45 virus and ubiquitination levels of endogenous RIP1 were examined. The results showed that the ubiquitination level of RIP1 was higher in UL45-null virus infection than in HA-UL45 virus infection (Fig 9D). These results demonstrate that UL48 inhibits TNFα-induced NF-κB activation through deubiquitination of RIP1 and that UL45 promotes this UL48 activity.
We also found using CoIP assays that the interaction of UL48 with RIP1 in co-transfected cells was increased when UL45 was co-expressed, demonstrating that UL45 promotes the binding of UL48 with RIP1 (Fig 10A). Furthermore, when HF cells were infected with HA-UL45 virus, RIP1 was re-localized to the cVAC in all cells showing HA-UL45 in the cVAC; however, when cells were infected with UL45-null virus, this RIP1 re-localization to the cVAC was not observed (Fig 10B). Taken together, these results demonstrate that UL48 inhibits TNFα-induced NF-κB activation through deubiquitination of RIP1 and that UL45 promotes the binding of UL48 with RIP1 and re-localization of RIP1 to the UL48-containing cVAC.
Since the DUB-containing tegument proteins and R1 homologs are conserved in other herpesviruses, we further investigated whether RIP1 is targeted by DUB and R1 encoded by other human α- and γ-herpesviruses. The results of CoIP assays using HSV-1 proteins showed that UL36 (DUB) did not interact with RIP1, while UL39 (R1) bound to RIP1 as previously reported [41] (S6A and S6B Fig). In similar assays using KSHV proteins, neither ORF64 (DUB) nor ORF61 (R1) interacted with RIP1 (S6D and S6E Fig). Furthermore, the interaction between DUB and R1was not observed with HSV-1 and KSHV proteins in CoIP assays (S6C and S6F Fig). These results demonstrate that targeting RIP1 by viral DUB occurs in HCMV but not in HSV-1 and KSHV, whereas targeting RIP1 by viral R1 homolog is shared by both HCMV and HSV-1.
We further examined whether the RIP1 targeting by viral DUB and R1 homolog is conserved in β-herpesviruses. In CoIP assays with MCMV proteins, both M48 (DUB) and M45 (R1 homolog) interacted with mRIP1 (S7A and S7B Fig). Furthermore, M48 interacted with M45 (S7C Fig), suggesting that the activity of viral DUB and R1 homolog to target RIP1 and interact with each other may be preserved in β-herpesviruses (S7D Fig).
In this study, we identified RIP1 as a cellular substrate of the HCMV-encoded UL48 DUB. We showed that UL48 interacts with RIP1, cleaves its K48- and K63-linked polyubiquitin chains, and inhibits NF-κB activation induced by RIP1 overexpression or TNFα treatment. The inhibitory action of UL48 on NF-κB signaling largely required its DUB activity, although the catalytically inactive mutant protein still resulted in modest inhibition of NF-κB activation. Other herpesvirus-encoded DUBs, such as HSV-1 UL36, KSHV ORF64, and EBV BPLF1, have been shown to target TRAF3, RIG-I, and TRAF6, respectively [8–10]. Therefore, our findings support a notion that the herpesvirus-encoded DUBs target key regulators of host innate immune and inflammatory signaling. We found that MCMV M48 interacted with mRIP1, while HSV-1 UL36 and KSHV ORF64 did not interact with RIP1, and that the interaction between viral DUB and R1 homolog was observed with MCMV proteins but not with HSV-1 and KSHV proteins. These suggest that RIP1 targeting by viral DUB and the interaction between DUB and R1 homolog may be conserved in β-herpesviruses.
We also showed that UL45, an inactive viral homolog of cellular RNR R1, interacts with RIP1 and inhibits NF-κB activation mediated by RIP1 overexpression or TNFα treatment. In MCMV infection, M45, a homolog of UL45, was shown to interact with mRIP1 to inhibit mRIP1-mediated NF-κB activation or cell death. M45 interacted with mRIP1 in a RHIM domain-independent manner and inhibited TNFα-induced activation of NF-kB, p38 MAPK, and caspase-independent cell death [27]. A role of the RHIM-dependent interaction of M45 with RIP1 in suppressing cell death was also reported [31]. Furthermore, the M45 required RHIM-dependent interaction with DAI and RIP3 in order to block DAI-induced NF-κB activation [29] and RIP3-mediated necroptosis [32, 33]. Our data demonstrate that HCMV UL45 interacts with RIP1 independent of the RHIM since UL45, which does not contain an apparent RHIM, bound to a RHIM-deleted RIP1 mutant. Interestingly, unlike the R1 homologs encoded by α- and γ-herpesviruses, HCMV UL45 or other β-herpesvirus R1 homologs are catalytically inactive due to the absence of most residues known to have a catalytic role [42]. Why β-herpesviruses encode inactive R1 homologs is not fully understood. However, our findings, together with the above-mentioned previous works with MCMV M45, point to the possibility that the inactive R1 homologs of β-herpesviruses have been preserved to regulate RIP1-associated activities. M45 was also reported to target NEMO and induce its lysosomal degradation [28]. Whether UL45 also targets NEMO needs to be further investigated.
HCMV has been shown to suppress robust activation of NF-κB signaling during the late stages of infection and viral early and late proteins appear to be involved in this regulation [43–45]. Recently, UL26, an early protein, was found to inhibit NF-κB activation by attenuating IKK phosphorylation in TNFα-treated cells [24]. The results of the present study reveal that UL48 and UL45, expressed at delayed-early or late times after infection, are potent suppressors of NF-κB activation in the late stages of infection. In particular, the results of our transfection assays demonstrated that UL48 and UL45 more effectively suppress TNFα-induced NF-κB activation than UL26. Therefore, at least three viral proteins appear to downregulate NF-κB signaling in the late stages of infection. We observed that infection with the Toledo virus containing the UL/b’ region led to greater NF-κB activation after TNFα treatment than infection with the Towne virus lacking this region. It has been shown that the UL138 gene in the UL/b’ region is responsible for this greater sensitivity of Toledo by upregulating cell surface expression of TNFR and this UL138 activity may contribute to TNFα-induced reactivation of virus from latently infected cells [39, 40]. Notably, our results demonstrated that the lack of UL45 expression or inactivation of UL48 DUB still increased TNFα-induced NF-κB signaling in Toledo virus infection. HCMV appears to differentially regulate NF-κB activation during the early and late phases of infection and between lytic and latent infection. Therefore, it is not surprising at all that HCMV has been evolved to have several genes that positively or negatively regulate NF-κB activation [46].
We demonstrated that, although UL48 and UL45 are separately capable of targeting RIP1 and inhibiting RIP1-mediated NF-κB signaling, they have a cooperative effect on the regulation of NF-κB signaling. Although UL48 was previously shown to interact with UL45 in yeast two-hybrid interaction assays and co-transfection/CoIP assays [17, 35], we for the first time provide evidence that this interaction occurs in virus-infected cells. In co-transfection assays, UL45 increased the activity of UL48 to cleave K63-linked polyubiquitination of RIP1 and interact with RIP1. More importantly, the ubiquitination level of endogenous RIP1 was higher in the late stages of infection during UL45-null virus infection compared to infection with HA-UL45 virus. One possible mechanism for the cooperative activity of UL45 is that UL45 binding to RIP1 alters RIP1 conformation or localization, converting it to a better substrate for UL48 DUB activity. Indeed, we observed that UL45 promoted the RIP1 binging by UL48 in co-transfected cells and re-localization of RIP1 to the UL48-containing cVAC in virus-infected cells. It would also be intriguing to test whether UL45 targets other UL48 substrates and promotes UL48’s DUB activity toward them.
Accumulation of deubiquitinated RIP1 in cells is a key step for the switch from pro-survival to pro-apoptotic function of RIP1. Our observation that UL48 and UL45 cooperatively increase the level of deubiquitinated RIP1 raises a question whether the expression of UL48 and UL45 sensitizes cells to apoptosis. We investigated this possibility by comparing anti-Fas/cycloheximide-induced cleavage of poly (ADP-ribose) polymerase (PARP) between wild-type and UL48(C24S) mutant virus-infected HF cells. However, we did not observe higher level of PARP cleavage by wild type virus than mutant virus. We reason this to the expression of strong viral inhibitor of caspase-8-induced apoptosis (vICA) [47, 48] that works at downstream of RIP1.
Both UL48 and UL45 are viral tegument proteins that are present within virions and delivered into host cells upon virus entry. Therefore, these proteins may also regulate NF-κB signaling immediately after virus entry. The interaction of these proteins may also play a role in the nuclear egress of the capsid or virion maturation in the cytoplasm, but this awaits further investigation.
Primary human foreskin fibroblast (HF) (American Type Culture Collection; ATCC PCS-201-010) and human embryonic kidney (HEK) 293T cells (ATCC) were grown in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum, penicillin (100 U/ml), and streptomycin (100 μg/ml) in a 5% CO2 humidified incubator at 37°C. Stocks of the Towne wild-type and UL48(C24S) mutant viruses were prepared as described previously [14]. To produce UV-inactivated HCMV (UV-HCMV), the virus stock was irradiated with UV light three times at 0.72 J/cm2 using a CL-1000 cross-linker (UVP). For infection experiments, HF cells were infected with virus at specified multiplicities of infection (MOI). At the indicated times, the growth medium was collected and combined with lysates prepared from the cell layer by freezing and thawing three times, clarified by centrifugation, and stored at -70°C until assayed for infectivity.
The wild-type UL45 DNA (Towne strain) was cloned into pENTR vectors (Invitrogen). Plasmid expressing Myc-tagged UL45 was produced by transferring the DNA to a pCS3-MT (with a 6Myc tag)-based destination vector using LR Clonase (Invitrogen). Plasmids expressing Myc-tagged UL48 (wild-type or C24S mutant) were described previously [14]. Plasmid expressing hemagglutinin (HA)-tagged UL48 was generated on a pSG5 background using Gateway technology as previously described [17]. Plasmid for flag-tagged RIP1 was provided by Jaehwan Song (Yonsei University). The wild-type RIP1 DNA was cloned into pENTR vectors (Invitrogen). Plasmid for HA-tagged RIP1 was produced by transferring the DNA to a pSG5-based destination vector using LR Clonase (Invitrogen). Plasmids encoding HA-tagged ΔKD, ΔID, ΔDD, ΔRHIM, and K377R mutant RIP1were generated by PCR in the same background. Plasmids expressing HSV-1 UL36 (pUL36-EGFP-N2) and KSHV ORF64 (pcDEF-Flag-ORF64) were provided by Prashant Desai (Johns Hopkins University) and Kyung-Soo Inn (Kyung Hee University), respectively. The HSV-1 UL39 and KSHV ORF61 genes were PCR cloned from viral DNAs into pENTR vector and plasmids expressing myc-tagged proteins were produced on a pCS3-MT-based destination vector using LR Clonase. pENTR vectors containing mouse RIP1 and MCMV M45 were produced by PCR from pCMV2-Flag-mRIP1 and pcDNA-M45-HA, respectively, that were provided by Wolfram Brune (Leibniz Institute for Experimental Virology). pENTR vector containing MCMV M48 was also produced by PCR from pEGFP-N1-M48 that was gifted by Jason W. Upton (University of Texas at Austin). pSG5 plasmids expressing HA-mRIP1 and HA-M45 and pCS3-MT plasmids expressing Myc-M45 and Myc-M48 were produced by transferring the DNAs into appropriate destination vectors using LR clonase. Plasmids for HA-Ub and CrmA were provided by Hongtae Kim (Sungkyunkwan University) and Kyeong Sook Choi (Ajou University), respectively.
Mouse monoclonal antibody (MAb) 810R, which detects epitopes present in both IE1 and IE2, was obtained from Millipore. Chicken egg yolk anti-UL45 IgY antibodies were gifted from Andrea Gallina (University of Milano). Rabbit polyclonal antibody (PAb) raised against the synthetic peptide corresponding to UL48 residues 278 to 295 (anti-UL48-N) was provided by Wade Gibson (Johns Hopkins University School of Medicine) [14]. Mouse MAbs against p52 (encoded by UL44) and pp28 (UL99) were purchased from Advanced Biotechnologies. The anti-pp65 (UL83) mouse MAb was purchased from Virusys. The anti-HA rat MAb (3F10) either conjugated with peroxidase or labeled with fluorescein isothiocyanate (FITC), anti-HA mouse MAb (12CA5), anti-HA rat MAb (3F10), and anti-Myc mouse MAb (9E10) conjugated with peroxidase were purchased from Roche. The anti-flag mouse MAb M2 was obtained from Sigma. Mouse MAbs for RIP1 (C-12) and p65 NF-κB subunit were purchased from Santa Cruz Biotechnology and Ab Frontier, respectively. The anti-phospho-p65(S536) and anti-phospho-IKKα/β rabbit MAbs were obtained from Cell Signaling Technology. Mouse MAbs for α-tubulin and β-actin were purchased from Sigma. Rabbit PAbs against NF-κB p50 and HDAC2 were purchased from Upstate and Zymed, respectively. Secondary antibodies such as FITC-labeled donkey anti-mouse IgG, Rhodamine Red-X-conjugated donkey anti-mouse IgG, Rhodamine Red-X-conjugated donkey anti-rabbit IgG, and Cy5-conjugated donkey anti-rat IgG were obtained from Jackson ImmunoResearch Laboratories, Inc.
The HCMV (Toledo strain) bacterial artificial chromosome (BAC) (Toledo-BAC) clone was kindly provided by Hua Zhu (Rutgers-New Jersey Medical School). The Toledo BAC clones encoding the UL45stop (UL45-null) and HA-tagged UL45 (HA-UL45) genes were generated using a counter-selection BAC modification kit (Gene Bridges). Briefly, the rpsL-neo cassette DNA was PCR-amplified using LMV1708/1731 primers containing homology arms consisting of 50 nucleotides upstream and downstream of the target region plus 24 nucleotides homologous to the rpsL-neo cassette. The amplified rpsL-neo fragments with homology arms were purified and introduced into E. coli DH10B containing the wild-type Toledo-BAC clone for recombination by electroporation using a Gene Pulser II (Bio-Rad). The intermediate Toledo-BAC constructs containing the rpsL-neo cassette were selected on Luria broth (LB) plates containing kanamycin. Next, the UL45stop fragments for replacing the rpsL-neo cassette were amplified by PCR using LMV1732/1733 primers. The amplified fragments were recombined into the Toledo-BAC DNA containing the rpsL-neo cassette, and the UL45-null Toledo-BAC was selected on LB plates containing streptomycin. The HA-UL45 Toledo-BAC clone was also generated from the UL45-null Toledo-BAC by the same method using LMV1494/1495/1733 primers. LMV primers used for bacmid mutagenesis were as follows: LMV1708, 5’-TCACTTTATTGAAATCTACCTGATTTCTTTGTTATTTTCCTCGTAAACTTGGCCTGGTGATGATGGCGGGATCG-3’; LMV1731, 5’- GCCGTCGGGAGACGGCGACTCGGGACGCCAACTGACGACGCCGCCACCACTCAGAAGAACTCGTCAAGAAGGCG-3’; LMV1732, 5’- TCACTTTATTGAAATCTACCTGATTTCTTTGTTATTTTCCTCGTAAACTTATGAATCCGGCTGACGCGGA-3’; LMV1733, 5’-GCCGTCGGGAGACGGCGACTCGGGACGCCA-3’; LMV1494, 5’-ATGTACCCATACGATGTTCCAGATTACGCTATGAATCCGGCTGACGCGGACGAGGAACAGCGGGTGTCCT-3’; and LMV1495 5’-TCACTTTATTGAAATCTACCTGATTTCTTTGTTATTTTCCTCGTAAACTTATGTACCCATACGATGTTCC-3’.
Electroporation was used to introduce the Toledo-BAC DNA into HF cells. For each reaction, HF cells (2 x 106) in 400 μl of resuspension buffer were mixed with 5 μg of Toledo-BAC DNA, 2 μg of plasmid pCMV71 encoding pp71 to enhance activation of the major immediate-early (MIE) promoter, and 1 μg of plasmid pEGFP-C1 to monitor electroporation efficiency. Electroporation was done at 1300 V for 30 ms using a Microporator MP-100 (Digital Bio Technology), and the cells were transferred to T-25 plates. When the surviving cells reached confluence, cells were transferred at a dilution of 1:2 into new flasks.
293T cells were transfected via polyethylenimine (PEI) (Sigma). A mixture of plasmid DNA and serum-free media was mixed with PEI. The volume of PEI used was based on a 3:1 ratio of PEI (μg):total DNA (μg). The mixture was kept at room temperature for 20 min and then added dropwise to cells. HF cells were transfected via OmicsFect (Omics Bio) or via electroporation. Electroporation was performed at 1300 V for 30 ms using a Microporator MP-100 (Digital Bio). The transfection efficiency of HF cells using electroporation under these conditions was about 50 to 60% (S1C Fig).
Cell lysates were harvested and prepared by sonication in 1 ml CoIP buffer (50 mM Tris-Cl [pH 7.4], 50 mM NaF, 5 mM sodium phosphate, 0.1% Triton X-100, containing protease inhibitors [Sigma]) using a Vibra cell microtip probe (Sonics and Materials) for 10 sec (pulse on: l s, pulse off: 0.5 s). The clarified cell lysates were incubated for 16 h with the appropriate antibody at 4°C. A total of 30 μl of 50% slurry of protein A- and G-Sepharose (Amersham) was added. After incubation for 2 h at 4°C, the mixture were pelleted and washed several times with CoIP buffer. Each sample was analyzed by SDS-PAGE and immunoblotting with the appropriate antibody.
The DNA-transfected cells or virus-infected cells were harvested with phosphate-buffered saline (PBS) and total cell extracts were prepared by boiling the cell pellets in sodium dodecyl sulfate (SDS) loading buffer. Equal amounts of the clarified cell extracts were separated by SDS-PAGE, and then transferred onto a nitrocellulose membrane (GE Healthcare Life Sciences) or PVDF membrane (Millipore). The membrane was blocked for 1 h in PBST (PBS plus 0.1% Tween 20 [Sigma]) containing 5% skim milk and then washed with PBST. After incubation with the appropriate antibodies, the proteins were visualized by standard procedures using an enhanced chemiluminescence system (Roche) and x-ray film (Kodak).
Cell extracts were prepared and loaded onto a Superose6 10/300 GL column (GE Healthcare) pre-equilibrated with CoIP buffer. The proteins were eluted at 0.5 ml/min. Each fraction (15 μl) was analyzed by immunoblotting with antibodies for RIP1, HA-UL45, and UL48. Apparent molecular mass was evaluated after column calibration with standard proteins [thyroglobulin (669 kDa), ferritin (440 kDa), aldolase (158 kDa), conalbumin (75 kDa), and ovalumin (44 kDa)] in the Gel Filtration Calibration Kit (GE Healthcare).
Cells were fixed in cold methanol for 5 min and rehydrated in cold PBS. Cells were first incubated for 30 min with human γ-globulin (Sigma) (2 mg per ml) at 37°C to block non-specific binding of antibodies to HCMV-induced Fc receptors [49]. They were then incubated for 1 h with primary antibodies in PBS at 37°C and then for 1 h with appropriate secondary antibodies labeled with FITC, Rhodamine Red-X, or Cy5. Antibodies were incubated together for triple labeling. The slides were examined and photographed with a Carl Zeiss LSM710Meta confocal microscope system.
Diluted samples were used to inoculate a monolayer of HF cells (1 x 105) in a 24-well plate. At 24 h post-infection, cells were fixed with 500 μl of cold methanol for 10 min. Cells were then washed three times in phosphate-buffered saline (PBS), incubated with anti-IE1 rabbit PAb in PBS at 37°C for 1 h, followed by incubation with phosphatase-conjugated anti-rabbit IgG antibody in PBS at 37°C for 1 h. Finally, the cells were gently washed in PBS and treated with 200 μl of developing solution (nitroblue tetrazolium/5-bromo-4-chloro-3-indolylphosphate) at room temperature for 1 h according to the manufacturer's instructions. The IE1-positive cells were counted in at least three to five separate fields per well under a light microscope (200X magnification).
Cells were collected and lysed by three freeze-thaw steps in 200 μl of 0.25 M Tris-HCl (pH 7.9) plus 1 mM dithiothreitol. Cell extracts were clarified in a microcentrifuge, and 20 μl of each extract was incubated with 350 μl of reaction buffer A (25 mM glycyl-glycine [pH 7.8], 5 mM ATP [pH 7.5], 4 mM EGTA [pH 8.0], 15 mM MgSO4) and then mixed with 100 μl of 0.25 mM luciferin (Sigma) in reaction buffer A. A TD-20/20 luminometer (Turner Designs) was used for a 10-s assay of the photons produced (measured in relative light units).
Cells were collected and lysed by three freeze-thaw steps in 200 μl of 0.25 M Tris-HCl (pH 7.9) plus 1 mM dithiothreitol. Cell extracts were clarified in a microcentrifuge. Each reaction mix was prepared in a 96-well plate including 20 μl of extract plus 100 μl of Z-buffer (0.06 M Na2HPO4, 0.04 M NaH2PO4, 0.01 M KCl, 0.001 M MgSO4, 0.05 M β-mercaptoethanol) plus 20 μl of o-nitrophenyl-β-D-galactopyranoside (ONPG [Sigma]; 4 mg/ml in sterile water). Reactions were incubated at 37°C until the yellow color developed. The reaction was stopped by adding 50 μl of 1 M Na2CO3. The β-galactosidase absorbance of each reaction was read at 420 nm.
293T cells were co-transfected with plasmids expressing target or effector proteins and plasmids expressing HA-ubiquitin. At 24 h after transfection, cell pellets were re-suspended with 2% SDS lysis buffer containing protease inhibitors (Sigma) and boiled for 10 min. Cell lysates were diluted ten-fold with CoIP buffer and sonicated using a Vibra cell microtip probe. The clarified cell lysates were incubated with 30 μl of 50% slurry of anti-Flag M2 affinity gel (Sigma) for 16 h at 4°C. The mixture was pelleted and washed several times with CoIP buffer. The bound proteins were boiled and analyzed by SDS-PAGE and immunoblotting with the appropriate antibody. To detect endogenous RIP1 ubiquitination in virus-infected cells, the mock-infected or virus-infected cells were treated with 5 mM N-ethylmaleimide (NEM) for 30 min before they were harvested. Cells were lysed in PBS containing 1% SDS and 5 mM NEM and then boiled for 10 min. Cell lysates were sonicated and immunoprecipitated with anti-RIP1 antibody (BD Biosciences). For immunoblotting, proteins were transferred to PVDF membranes and denatured using 6 M guanidine-HCl containing 20 mM Tris-HCl (pH 7.5), 1 mM PMSF, and 0.5 mM β-mercaptoethanol for 30 min at 4°C. Ubiquitinated RIP1 was identified by HRP-conjugated anti-ubiquitin antibody (Biomol).
Samples were compared using Student’s t-test, and p-values <0.05 (*) <0.01 (**) and <0.001 (***) are indicated in the figures.
|
10.1371/journal.ppat.1003414 | IRG and GBP Host Resistance Factors Target Aberrant, “Non-self” Vacuoles Characterized by the Missing of “Self” IRGM Proteins | Interferon-inducible GTPases of the Immunity Related GTPase (IRG) and Guanylate Binding Protein (GBP) families provide resistance to intracellular pathogenic microbes. IRGs and GBPs stably associate with pathogen-containing vacuoles (PVs) and elicit immune pathways directed at the targeted vacuoles. Targeting of Interferon-inducible GTPases to PVs requires the formation of higher-order protein oligomers, a process negatively regulated by a subclass of IRG proteins called IRGMs. We found that the paralogous IRGM proteins Irgm1 and Irgm3 fail to robustly associate with “non-self” PVs containing either the bacterial pathogen Chlamydia trachomatis or the protozoan pathogen Toxoplasma gondii. Instead, Irgm1 and Irgm3 reside on “self” organelles including lipid droplets (LDs). Whereas IRGM-positive LDs are guarded against the stable association with other IRGs and GBPs, we demonstrate that IRGM-stripped LDs become high affinity binding substrates for IRG and GBP proteins. These data reveal that intracellular immune recognition of organelle-like structures by IRG and GBP proteins is partly dictated by the missing of “self” IRGM proteins from these structures.
| Cell-autonomous host defense pathways directed against vacuolar pathogens constitute an essential arm of the mammalian innate immune defense system. Underlying most of these defense strategies is the ability of the host cell to recognize foreign or pathogen-modified structures and to deliver antimicrobial molecules specifically to these sites. Specific targeting of molecules to pathogen-containing vacuoles (PVs) requires host cells to recognize PVs as “non-self” structures that are distinct from intact “self” structures like organelles and other endomembrane components. In this work, we develop a new framework for understanding a critical principle that guides the mammalian immune system in the recognition of PVs as “non-self” structures. Our data indicates that so-called IRGM proteins function as markers of “self” compartments. We find that IRGM proteins act as “guards” that prevent a set of antimicrobial GTPases from stable association with “self” membranes. Because IRGM proteins are largely absent from “non-self” PVs, we propose that intracellular immune recognition of PVs can occur via the missing of “self” IRGM proteins.
| Many intracellular pathogens including the bacterium C. trachomatis and the protozoa T. gondii co-opt the host cell endomembrane system to enclose themselves inside membrane-bound vacuoles. Within the confines of these remodeled PVs, microbes acquire nutrients and replicate [1]. To combat these pathogens, the mammalian host has evolved a large repertoire of cell-autonomous defense mechanisms that kill or restrain the replication of microbes residing within vacuoles [2], [3]. While these defense mechanisms are effective at targeting foreign or “non-self” vacuoles, they also have the potential to cause organelle damage and must therefore be tightly regulated. Control over these host defenses is executed at two critical steps: (i) induction of genes encoding host resistance factors in the context of an infection and (ii) targeting of these resistance factors to the appropriate intracellular location, for example to PVs.
These two modes of regulation are exemplified by the induction and execution of cell-autonomous defenses by the cytokine Interferon-γ (IFNγ). The importance of IFNγ in host immunity is demonstrated by the severe immuno-deficiencies observed in genetically engineered mouse strains lacking IFNγ or its receptor and in patients carrying rare mutations in genes critical for IFNγ signal transduction [4], [5]. IFNγ is produced by immune-activated lymphocytes and exerts its antimicrobial effects by dramatically remodeling the transcriptional expression profile of target cells bearing the IFNγ receptor [3]. IFNγ-induced resistance genes include members of two IFN-inducible GTPase families named IRGs and GBPs. Members of both GTPase families have the ability to translocate and to adhere specifically to PVs in order to inhibit intracellular pathogen growth. Although the specificity of this intracellular targeting event is well documented [3], [6], the underlying mechanism is unclear.
Once docked to PVs, GBP proteins recruit antimicrobial protein complexes that include the NADPH oxidase NOX2, the autophagy apparatus and the inflammasome [3]. IRG proteins on the other hand can directly disrupt PV membranes, thereby releasing vacuolar pathogens into the cytosol where they can be removed through autophagy [6], [7]. IRG GTPases are divided into two categories: (i) the predominantly cytosolic GKS proteins constitute the most abundant group and harbor a conserved GX4GKS sequence in the first nucleotide-binding motif (G1), (ii) the predominantly membrane-bound IRGM proteins instead contain a non-canonical P-loop sequence GX4GMS [6]. Both GKS and IRGM proteins are essential for cell-autonomous resistance to infections with C. trachomatis and T. gondii in mice but fulfill distinct functions in this process [6]. Whereas GKS proteins directly target and eliminate C. trachomatis and T. gondii PVs, IRGM proteins appear to orchestrate the targeting of GKS proteins to PVs by an incompletely understood mechanism [6]. In addition to their role as regulators of GKS protein function, IRGM proteins also exert antimicrobial activities independently of GKS proteins. Both mouse and human IRGM proteins promote the formation of autophagosomes upon IFNγ stimulation [8]–[10]. Additionally, murine Irgm1 loads onto early phagosomes containing beads or live bacteria [11]–[13]. Vacuolar Irgm1 interacts with target SNARE protein complexes and through these interactions can facilitate the rapid fusion of Irgm1-coated phagosomes with degradative lysosomes. Accelerated lysosomal maturation was shown to result in the destruction of the attenuated pathogen Mycobacterium bovis BCG contained within Irgm1-positive phagosomes in mouse macrophages [13]. Similar to Irgm1, Irgm3 was implicated as a mediator of direct antimicrobial activities towards T. gondii [14].
To initially establish vacuoles permissive for microbial survival in IFNγ-activated cells, pathogens must have evolved strategies to evade the direct, fast-acting immune responses mediated by membrane-bound Irgm1 and Irgm3 proteins. In agreement with the existence of such evasion mechanisms, we observed that Irgm1 and Irgm3 failed to robustly associate with PVs formed by either C. trachomatis or T. gondii. The absence of substantial amounts of Irgm1/m3 from PVs contrasted with the abundant localization of Irgm1/m3 to “self” structures like LDs. We found that Irgm1/m3-decorated LDs are largely devoid of GKS and GBP proteins, whereas Irgm1/m3-deficient PVs are targets for GKS and GBP proteins. These observations led us to hypothesize that the absence of Irgm1/m3 proteins marked intracellular structures as targets for a “second line of defense” mediated by GKS and GBP proteins. In support of this hypothesis, we demonstrated that stripping LDs of Irgm1/3 resulted in mistargeting of GKS and GBP proteins to LDs independently of an infection. Because IRGM proteins were previously shown to inhibit GKS protein oligomerization [15], we propose a model in which the missing of “self” IRGM proteins from “non-self” PVs results in the formation of GKS (and GBP) protein oligomers with high avidity for membrane binding.
GKS proteins in their GTP-bound state form dimers [6]. Dimerization occurs at the G domain interface and is a prerequisite for the formation of higher order GKS protein oligomers. Mutations that diminish guanine nucleotide binding or disrupt the G domain interface eliminate both protein oligomerization and targeting to T. gondii PVs [16], [17]. To determine if these findings extended to other PVs, we first tested whether guanine nucleotide binding of the GKS protein Irgb10 was essential for their targeting to “inclusions,” the PVs formed by C. trachomatis. We replaced the serine on position 82 of Irgb10 in the conserved P-loop GKS motif with asparagine (Irgb10S82N). This mutation is analogous to the Irga6S83N mutation that abrogates guanine nucleotide binding and T. gondii PV localization [16]. We found that Irgb10-GFP-fusion proteins harboring the S82N mutation or a deletion of the central G-domain (Irgb10ΔG) failed to localize to C. trachomatis inclusions in infected mouse embryonic fibroblasts (MEFs) (Figure 1A).
Combined with previous results in T. gondii [16], our data suggested that protein oligomerization of GTP-bound Irgb10 is essential for tethering this GKS protein to C. trachomatis inclusion membranes. To test whether protein oligomerization of the N- and C-terminal domains of Irgb10 was sufficient to target inclusion membranes, we replaced the G domain of Irgb10 with alternative protein oligomerization domains and monitored the subcellular localization of these protein chimeras. We first substituted the G domain of Irgb10 with the tetramer-forming protein dsRED [18], which emits red fluorescence exclusively in the oligomerized form [19]. Insertion of dsRED between the N-terminal domain (NTD) and C-terminal domain (CTD) of Irgb10ΔG (Irgb10NTD-dsRED-CTD) restored the association of a GFP-tagged fusion protein with C. trachomatis inclusions (Figure 1A). Inserting dsRED in between the N-terminal myristoylation motif (Myr) and the CTD of Irgb10 (Irgb10Myr-dsRED-CTD) similarly redirected the mutant variant Irgb10Myr-CTD to inclusions (Figure 1A). Tetramerized Irgb10 localized to IncG-positive inclusion membranes (Figure 1B). Inclusion targeting required the presence of both the Irgb10 myristoylation motif and the C-terminal amphipathic helix αK (Figure 1A, Figure 2A and data not shown). As an alternative mediator of protein oligomerization, we used the highly oligomeric cytoplasmic yeast protein TyA [20]. Insertion of TyA in between the myristoylation domain and the C-terminus of Irgb10 (Irgb10Myr-TyA-CTD) similarly re-localized these fusion proteins towards inclusions. Myristoylated-TyA (Myr-TyA) localized to microvesicles, as described [21], but failed to associate with inclusions (Figure 1C), demonstrating that the C-terminus of Irgb10 containing the αK amphipathic helix is essential for inclusion targeting. In summary, these data show that the oligomerization of the N- and C-terminal lipid binding domains of Irgb10 was sufficient to drive localization to inclusions.
The targeting of GKS proteins to T. gondii is substantially diminished in the absence of the IRGM proteins Irgm1 and Irgm3 [16], [22]. We found that the association of Irgb10 and other GKS proteins with C. trachomatis inclusions was similarly reduced in infected MEFs derived from Irgm1−/−, Irgm3−/− and Irgm1/m3−/− mice (Figure 3). These data indicate that Irgm1 and Irgm3 either directly or indirectly promote the delivery of GKS proteins to inclusions. Because IRGM proteins physically interact with GKS proteins at the G domain interface [16], we hypothesized that IRGM proteins facilitate the delivery of GKS proteins to inclusions through their interactions with the G domain of GKS proteins. In such a scenario, artificially oligomerized Irgb10 lacking a G domain should target inclusions independently of IRGM proteins. To test the hypothesis, we expressed two tetramerized, chimeric Irgb10ΔG proteins, Irgb10NTD-dsRED-CTD and Irgb10NTD-dsRED-αK, in wildtype and Irgm3−/− MEFs and scored the frequency of dsRED signal on inclusions. We chose Irgm3−/− MEFs for these experiments, because they displayed the most pronounced defect in targeting endogenous Irgb10 to inclusions (Figure 3). In contrast to endogenous Irgb10 (Figure 2B and Figure 3), tetramerized Irgb10 lacking a G domain targeted inclusions with the same efficiency in wildtype and Irgm3-deficient cells (Figure 2A). These data suggest that Irgm3 regulates the targeting of Irgb10 to inclusions through its interaction with the G domain of Irgb10.
It is currently unknown where inside a cell IRGM proteins interact with GKS proteins to regulate their function. To determine whether IRGM proteins regulate GKS proteins directly at PV membranes, we first monitored the subcellular localization of IRGM proteins in cells infected with either T. gondii or C. trachomatis. As reported previously [14], [23], we found that endogenous Irgm3 but not Irgm1 associated with T. gondii PVs, albeit only weakly relative to its association with endogenous, puncta-like structures (Figure 4A and data not shown). These results are also in agreement with a previous report demonstrating that Irgm3 associates with T. gondii PVs at a lower frequency than GKS proteins do [24]. Next we examined the subcellular localization of endogenous Irgm1 and Irgm3 in C. trachomatis-infected cells. In agreement with a previous report [25], we detected association of Irgm3 with C. trachomatis inclusions at 2 hpi. However, Irgm3 associated only weakly with inclusions relative to its interactions with endogenous structures (Figure 4B). Similar to the staining pattern of T. gondii PVs, we failed to detect the presence of Irgm1 on inclusions (data not shown). To determine whether Irgm1 or Irgm3 could target established inclusions, we infected MEFs with C. trachomatis and subsequently treated cells with IFNγ at 3hpi. Under these experimental conditions endogenous as well as ectopically expressed Irgm1 and Irgm3 were not present at inclusion membranes in detectable amounts at 20 hpi (Figure 4C, Figure 5B and data not shown). Collectively, these data show that PVs formed by either C. trachomatis or T. gondii are devoid of substantial amounts of Irgm1 and Irgm3 proteins.
Because established PVs lack sizeable amounts of Irgm1/m3, we considered the hypothesis that IRGM proteins regulate Irgb10 and other GKS proteins at sites distinct from PVs. It is known that IRGM proteins localize to various endomembranes, including LDs, a neutral lipid storage organelle [6]. Specifically, Irgm3 was shown to localize to LDs in IFNγ-treated dendritic cells [26]. To determine whether or not Irgm3 also localizes to LDs in IFNγ-treated MEFs, we induced the formation of LDs by supplementing the growth media with oleic acid (OA) and subsequently stained these cells with the neutral lipid dye BODIPY493/503 and with anti-Irgm3 antibody. We found that Irgm3 co-localized with the BODIPY dye in IFNγ-treated MEFs (Figure 5A). To determine whether additional IRGM proteins localize to LDs, we monitored the localization of C-terminally V5-tagged Irgm1, Irgm2 and Irgm3 inside OA-treated MEFs. In addition to Irgm3-V5, Irgm1-V5 and Irgm2-V5 co-localized with a subset of LDs but not with inclusions (Figure 5B and Figure S1). Staining for endogenous protein confirmed the presence of Irgm1 but not Irgm2 on a subset of LDs (Figure S2 and data not shown). We next asked if GKS proteins were also found on LDs by immunostaining IFNγ-activated MEFs with antibodies directed against three representative GKS proteins Irga6, Irgb6 and Irgb10. We were unable to detect co-localization of these proteins with BODIPY-labeled LDs in wildtype MEFs by immunofluorescence (Figure 5C and D).
We independently confirmed these observations by assessing the levels of IRG proteins on purified LDs. LDs purified from IFNγ-treated, wildtype MEFs by sucrose gradient centrifugation displayed significant levels of Irgm1 and Irgm3 (Figure 5E), and relatively small amounts of Irgb10 (Figure 5E), suggesting possible transient interactions between IRGM proteins and Irgb10 on the surface of LDs. In summary, these data indicate that LDs of wildtype cells are decorated with Irgm1 and Irgm3 but only weakly associate with GKS proteins.
Although LDs could play an essential role in guiding GKS proteins to inclusions, we thought this was unlikely, because LD-deficient cells lines still target Irgb10 to inclusions (H.A.S. and R.H.V., unpublished data). Because IRGM proteins inhibit GTP acquisition by GKS proteins and are believed to thereby block the ability of GKS proteins to bind lipids [15], [16], we formed an alternative hypothesis in which LD-resident IRGM proteins would prevent GKS proteins from binding to LDs. To test our hypothesis, we examined the localization of Irgb10 in Irgm1/m3−/− MEFs that contain IRGM-deficient LDs. We found that the LDs of Irgm1/m3−/− MEFs were heavily decorated with Irgb10 (Figure 5D). Targeting of Irgb10 to LD in Irgm1/m3−/− MEFs was primarily due to the absence of Irgm3, because Irgm3−/− MEFs but not Irgm1−/− MEFs displayed a substantial increase in the number of Irgb10-positive LDs (Figure 5D). The simultaneous deletion of both Irgm3 and Irgm1, however, exacerbated the association of Irgb10 with LDs (Figure 5D) suggesting that these proteins fulfill partially redundant functions in protecting LDs against Irgb10 targeting. The role of Irgm1 and Irgm3 in guarding LDs was not limited to Irgb10 but extended to other GKS proteins including Irga6 and Irgb6 (Figure 5C). Again, Irgm3 was predominantly responsible for guarding LDs, because ectopic expression of Irgm3 in either Irgm3−/− or Irgm1/m3−/− MEFs prevented deposition of Irga6 on LDs (Figure S3). Irgm1 and Irgm3 were also required to prevent Irgb10 accumulation on LD in primary macrophages, indicating that the observed phenomenon is not cell type specific (Figure S4). Furthermore, endogenous LDs found infrequently in MEFs not treated with OA also acquired Irgb10 in the absence of Irgm1 and Irgm3 (Figure S5A), demonstrating that the aberrant localization of Irgb10 was not induced by OA treatment. Lastly, consistent with our immunofluorescence observations, we detected a robust increase in the amount of Irgb10 protein present in the LD fraction derived from Irgm1/m3−/− MEFs compared to wildtype MEFs (Figure 5E). These data combined demonstrate that GKS proteins target LDs in the absence of Irgm1/m3.
Because IRGM proteins can act as positive regulators of autophagy [8], [10], [27], we also considered the possibility that the mislocalization of GKS proteins to LDs in Irgm1/m3−/− cells was a consequence of disrupted autophagy. To test this hypothesis, we examined the subcellular localization of Irgb10 in autophagy-deficient Atg5−/− MEFs. We did not observe an increase in the association of Irgb10 protein with LDs in Atg5−/− MEFs (Figure 5E and Figure S6), indicating that a defect in autophagy is not the underlying cause for the mislocalization of GKS proteins to LDs in Irgm1/m3−/− MEFs.
Next, we asked whether IRGM proteins exclusively guard LDs. We observed that Irgb10 formed “aggregate-like structures” in Irgm1/m3−/− cells that did not identify as LDs (Figure S5A), suggesting that GKS protein could target additional “self” structures in the absence of Irgm1/m3. In support of this hypothesis, we found that GKS proteins also targeted mitochondria (Figure S5B) and peroxisomes (Figure S5C) in Irgm1/m3−/− cells. In wildtype cells mitochondria are decorated with Irgm1 (G.A.T., manuscript in preparation) and subsets of peroxisomes stain positive for Irgm3 (Figure S5D). In summary, our data suggest that IRGM proteins guard LDs and other organelles against the stable association with GKS proteins.
We demonstrated that endogenous GKS proteins like Irgb10 stably associate with LDs in the absence of IRGM proteins (Figure 5). Similarly, ectopically expressed Irgb10 frequently targets LDs in Irgm1/m3−/− MEFs (Figure S7A) but not in wildtype MEFs (Figure 6A). Two distinct models could explain the differential targeting of Irgb10 and other GKS proteins to IRGM-deficient but not IRGM-positive LDs: in the first model, the presence of IRGM proteins alters the molecular properties of LDs such that LDs do not serve as binding substrates for GKS proteins; in the second model, IRGM proteins directly interact with GKS proteins on the surface of LDs and block lipid binding. Previous studies have shown that IRGM proteins can transiently interact with GKS proteins and thereby retain GKS proteins in the GDP-bound, inactive state [15], [16], thus supporting the second model. We therefore predicted that an Irgb10 variant locked in the active, GTP-bound state should be able to overcome IRGM protein mediated restrictions on lipid binding and be able to target IRGM-positive LDs. To generate a GTP-locked Irgb10 mutant, we replaced the lysine residue of the conserved GKS motif with alanine (Irgb10K81A), as homologous mutations in Irga6 or Irgb6 interfere with GTP hydrolysis and force these GTPases into a GTP-locked state [16]. Similar to previous observations demonstrating the targeting of GTP-locked Irga6 to T. gondii vacuoles [16], we found that Irgb10K81A co-localized with C. trachomatis inclusions (Figure 6). However, in contrast to Irgb10WT, Irgb10K81A co-localized with Tip47-positive LDs (Figure 6A) and Irgm3 (Figure 6B) in wildtype MEFs. In contrast to Irgb10WT and Irgb10K81A, a mutant with low affinity binding for GTP (Irgb10S82N) failed to associate with either inclusions or LDs (Figure S7). Overall, these data are consistent with a model, in which IRGM proteins on LDs block the activation of endogenous Irgb10 on the surface of LDs and prevent their stable association of Irgb10 with LDs.
GBP proteins constitute a second large family of IFNγ-inducible GTPases known to target PVs and to provide resistance to infections with vacuolar pathogens [3]. Because we previously observed that the subcellular location of the GBP protein Gbp2 is altered in the absence of IRGM proteins [27], we hypothesized that IRGM proteins could guard self-membranes against the improper deposition of not only GKS but also GBP proteins. Consistent with this, we found that Gbp2 co-localized with LDs in Irgm1/m3−/− but not in wildtype MEFs (Figure 7A) and was enriched in LD fractions obtained from Irgm1/m3−/− cells (Figure 7B). Irgm1 and Irgm3 appeared to fulfill partially redundant functions in guarding LDs against Gbp2 targeting, because Gbp2 localization to LDs was more pronounced in Irgm1/m3−/− MEFs than in Irgm1−/− or Irgm3−/− single gene deletion cells (Figure 7C and Figure S8). Because both Irgm1 and Irgm3 regulate the formation and/or maturation of autophagosomes [8], [27], it was formally possible that the mislocalization of Gbp2 to LDs in Irgm1/m3−/− cells resulted from a defect of these cells in autophagy. However, it is unlikely that defective autophagy is the primary cause for mislocalization of Gbp2 to LDs, because LDs inside autophagy-deficient Atg5−/− MEFs remained exempt from Gbp2 targeting (Figure 7A and B).
Similar to endogenous Gbp2, we found that ectopically expressed, N-terminally tagged FLAG-Gbp1 protein was redirected to LDs in the absence of Irgm1 and Irgm3 proteins (Figure 7D). To determine whether activation of Gbp1 was critical for targeting IRGM-deficient LDs, we expressed a FLAG-Gbp1K51A mutant form that has previously been shown to be defective for nucleotide binding and protein oligomerization [28]. In contrast to wildtype FLAG-Gbp1, we found that FLAG-Gbp1K51A failed to associate with LDs in Irgm1/m3−/− cells (Figure 7D). These data suggest that the active form of Gbp1 associates with IRGM-deficient LDs.
Our data demonstrated that Gbp1 and Gbp2 localized to IRGM-deficient LDs. One of the known effector molecules of Gbp1 is the autophagic adaptor protein p62/sequestosome-1 [29]. We therefore hypothesized that Gbp1 proteins residing on IRGM-deficient LDs would be able to recruit p62 to LDs. In support of our hypothesis we found that 4–5% of LDs in IFNγ-treated Irgm1/m3−/− cells stained positive for p62 (Figure 8A). In contrast to Irgm1/m3−/− cells, we were unable to detect p62 on LDs of IFNγ-treated wildtype cells using immunofluorescence microscopy (Figure 8A).
A critical function of p62 is to bind to macromolecular cargo that is destined for autophagic destruction [30]. To deliver its cargo to autophagosomes, p62 also binds directly to the ubiquitin-like protein LC3, a maker of autophagosomes. To determine whether IRGM-deficient LDs are delivered to autophagosomes upon IFNγ activation, we incubated both wildtype and Irgm1/m3−/− MEFs with OA and IFNγ and subsequently stained cells with anti-LC3 and BODIPY. In these experiments, we frequently observed LDs that were engulfed within ring-like LC3-positive structures in Irgm1/m3−/− but not in wildtype MEFs (Figure 8B). Similarly, LDs purified from Irgm1/m3−/− cells were enriched for LC3-II, the lipidated form of LC3 that is associated with autophagosomes (Figure 8C). Collectively, these data strongly suggested that IRGM-deficient LDs were captured inside autophagosomes upon IFNγ activation.
To test this model further, we treated cells with the lysosomotropic H+-ATPase inhibitor bafilomcyicn (BAF), a known inhibitor of autophagic flux [31]. We observed a substantial increase in the number of p62-positive LDs in Irgm1/m3−/− MEFs upon combined treatment with IFNγ and BAF (Figure 8A). BAF treatment also resulted in the appearance of p62-positive LDs in IFNγ-treated wildtype MEFs, however, at a frequency significantly lower than what we observed in BAF-treated Irgm1/m3−/− MEFs (Figure 8A). These data indicated that the targeting of p62 to IRGM-deficient LDs resulted in the degradation of LD-bound p62. Furthermore, our observations excluded an alternative model in which the increase in the number of p62-positve LDs in Irgm1/m3−/− MEFs was due to a defect in autophagosome maturation in these cells.
We then asked whether the increased association of p62 and LC3 with IRGM-deficient LDs would affect the total mass of LDs. To quantify LD mass, we used a flow cytometry approach using BODIPY staining, as previously described [26]. We found that IFNγ treatment resulted in an increase in the BODIPY signal in wildtype MEFs, similar to the observations previously made in dendritic cells [26]. In contrast to the increase in the BODIPY signal observed in IFNγ-treated wildtype cells, the BODIPY signal decreased in IFNγ-treated Irgm1/m3−/− MEFs (Figure 9 and Figure S9), suggesting increased rates of LD degradation in IRGM-deficient cells. To determine whether the decrease in LD mass in Irgm1/m3−/− MEFs was due to autophagy ( = lipophagy), we treated cells with BAF. BAF treatment blocked the IFNγ-induced decrease in LD mass in Irgm1/m3−/− MEFs (Figure 9). In sum, these data strongly support a model in which p62 targets IRGM-deficient but not IRGM-guarded LDs and delivers IRGM-deficient LDs to autophagosomes for degradation.
It has previously been reported that targeting of GBP proteins to T. gondii PVs is facilitated by unknown IFNγ-inducible factors [32], [33]. Because our data had already established functional interactions between GBP and IRGM proteins, we asked whether IRGM proteins could act as IFNγ-inducible co-factors promoting the recruitment of Gbp2 to PVs. We found that ectopically expressed Gbp2-GFP fusion proteins failed to localize to C. trachomatis inclusions in the absence of IFNγ treatment or in IFNγ-activated Irgm1/m3−/− cells (Figure 10A). Similarly, we observed that both Irgm1 and Irgm3 played critical roles in facilitating targeting of endogenous Gbp2 protein to inclusions (Figure 10B and C). Similar to Gbp2, recruitment of FLAG-Gbp1 to inclusions was also dependent on IRGM proteins (Figure 10D). To determine whether the regulatory role of IRGM proteins extends to the recruitment of GBP proteins to PVs formed by pathogens other than C. trachomatis, we monitored co-localization of both Gbp2 and, as a control, Irgb10 with T. gondii vacuoles in wildtype, Irgm1−/−, Irgm3−/− and Irgm1/m3−/− MEFs. We observed that the deletion of both Irgm1 and Irgm3 caused a near complete defect in the recruitment of Irgb10 (Figure 10E) and Gbp2 to T. gondii PVs at 0.5 hpi (Figure 10F). These observations demonstrate that the expression of IRGM proteins is critical for the efficient delivery of GBP proteins to vacuoles formed by distinct intracellular pathogens.
The data presented in this study support a model in which Irgm1 and Irgm3 proteins act as “guard molecules” that block GKS and GBP proteins from stably associating with “self” structures (Figure 11). On PVs, however, guarding Irgm1 and Irgm3 proteins are present at such low levels that GKS and GBP proteins can firmly attach to these unprotected membranes. In support of our model we found that Irgm1 and Irgm3, but not GKS and GBP proteins, are present in LDs of wild type cells (Figure 5 and Figure 7). In the absence of Irgm1 and Irgm3, however, normally GKS-/GBP-deficient LDs become decorated with various GKS and GBP proteins. We provide evidence that GKS and GBP proteins assemble on IRGM-deficient LDs in their GTP-bound, i.e. “active” state (Figure 7 and Figure S7). According to our model GKS-/GBP-decorated LDs should resemble GKS-/GBP-decorated PVs and would therefore be expected to become targets of GKS-/GBP-solicited immune responses. Consistent with such a scenario, we demonstrate that the Gbp1 effector protein p62 is recruited to IRGM-deficient LDs. The targeting of p62 to LDs in Irgm1/m3−/− MEFs likely accounts for the enhanced association of IRGM-deficient LDs with the autophagic marker LC3-II and the decrease in LD mass upon IFNγ activation that we observed in Irgm1/m3−/− MEFs. Our observations are consistent with a previous report that showed that the number of LDs is significantly reduced in IFNγ-activated Irgm3−/− dendritic cells compared to IFNγ-activated wildtype dendritic cells [26]. Whereas the authors of this previous study speculated that Irgm3 could play a role in the neoformation of LDs triggered upon IFNγ receptor signaling, our data strongly suggest that the decrease in LDs observed in IRGM-deficient cells primarily results from GBP-mediated autophagy of LDs. However, because our experiments were conducted in MEFs, additional studies are needed to determine whether Irgm3 may also play a role in LD neoformation in dendritic or other cell types.
While we propose that IRGM proteins guard self-organelles against misdirected attacks by GKS and GBP proteins, our studies do not exclude additional roles for IRGM proteins in organelle homeostasis. For example, human IRGM protein translocates to mitochondria and induces mitochondrial fission [10]. Because mitochondrial fission not only results in the production of radical oxygen species and the induction of antimicrobial autophagy, but also contributes to the isolation and removal of damaged segments of mitochondria, IRGM proteins may indeed regulate the homeostasis of specific organelles like mitochondria.
The question now arises as to why the “guarding” Irgm1 and Irgm3 proteins are present on “self” membranes but largely absent from “non-self” PVs. The answer to this question may be quite obvious, if one considers that IRGM proteins can exert antimicrobial activities directly, once localized to PVs [3]. To escape from IRGM-mediated antimicrobial activities like lysosomal targeting, we propose that vacuolar pathogens have evolved strategies to actively avoid co-localization with IRGM proteins. The absence of Irgm1/m3 from PVs would initially allow pathogens to establish a vacuolar niche permissive for microbial replication. However, the evasion of IRGM proteins would simultaneously mark PVs for immune targeting by GKS and GBP proteins. The principle underlying this type of intracellular immune recognition is similar to the extracellular immune recognition process by which NK cells detect transformed and/or virus-infected cells [34]. NK cells express inhibitory receptors on their cell surface. The ligands for one set of inhibitory NK receptors are MHC class I molecules displayed on the surface of host cells. The primary function of the MHC class I molecules is to display viral or tumor antigens to cytotoxic T cells. To avoid immune recognition by cytotoxic T cells, many tumor cells and viruses have evolved mechanisms to downmodulate MHC class I surface expression. However, the failure of MHC class I-deficient cells to provide an inhibitory signal to NK cells, allows NK cells to recognize the missing of “self” MHC class I in transformed or infected cells. In this analogy IRGM proteins resemble MHC class I molecules: just like MHC class I molecules, IRGM proteins fulfill dual functions in that they can promote antimicrobial activities directly and simultaneously act as inhibitory molecules that block the activation of an alternative defense system.
How do Irgm1 and Irgm3 proteins guard membranes against GKS and GBP proteins? Studies performed by Howard and colleagues indicate that IRGM proteins act as Guanine nucleotide Dissociation Inhibitors (GDI) for GKS proteins [15], [16]. Based on these findings, Howard and colleagues proposed that by maintaining GKS GTPases in the GDP-bound, monomeric state, IRGM proteins reduce the lipid binding capacity of GKS proteins and block their stable association with IRGM-coated membranes. Here, we provide direct evidence in support of this model. As originally proposed by Hunn et al., our data indicate that the absence of IRGM proteins from PVs promotes the transition of GKS proteins into the GTP-bound, active state and their stable association with IRGM-deficient PVs. We show here that IRGM-deficient membranes are also targets for GBP proteins (Figure 7). Whether IRGM proteins act as GDIs for GBP proteins or block the ability of GBP proteins to associate with LDs by an alternative mechanism will need to be elucidated in future studies.
The lipid binding substrates for GKS and GBP proteins are currently unknown. However, lipid components that are present in LDs as well as in T. gondii parasitophorous and Chlamydia inclusion membranes are obvious candidates to act as GKS- and GBP-interacting molecules. It is tempting to speculate that GKS and GBP proteins might have evolved to preferentially bind to lipids that are frequently found in PVs but infrequently found on the cytosolic face of most endomembranes [35]. According to this model, most “self” structures would be protected against the erroneous attack by GKS and GBP proteins for two reasons: 1) the presence of guarding IRGM proteins and 2) the relative sparsity of lipid binding substrates on the cytosolic leaflet of “self”-membranes. This model would suggest that GKS and GBP proteins tether specifically to PVs due to the missing of “self” IRGM proteins from PVs and the presence of an unknown “second signal” on PVs. As suggested above, a unique pattern of lipids may provide such a second signal, although other molecules may also be involved. Albeit speculative at this point, we propose that LDs feature such a second signal and therefore become primary targets for GKS and GBP proteins in the absence of IRGM guard molecules.
The requirement of a second signal for GKS/GBP membrane targeting as proposed in the model outlined above could in part explain why Irgm1/m3−/− mice and cells are viable in spite of lacking two critical “guard” proteins. Alternatively, expression of Irgm2 in Irgm1/m3−/− cells may provide sufficient protection to assure survival of Irgm1/m3−/− cells upon immune activation. This second model would necessitate that Irgm2 like Irgm1/m3 guards “self” membranes against GKS and GBP proteins.
In addition to guarding self-structures, expression of IRGM proteins is required for the efficient targeting of endogenous GKS proteins to C. trachomatis inclusions (Figure 3) and T. gondii PVs [16]. However, tetramerized Irgb10-dsRED, when overexpressed, targets PVs efficiently in IRGM-deficient cells (Figure 2A). These data argue against a direct role for IRGM proteins in delivering GKS proteins to PVs. We therefore propose a model in which IRGM proteins fulfill an indirect role in targeting endogenous GKS proteins to PVs: in this model GKS proteins can bind to an excess of unguarded “self” membranes in IRGM-deficient cells. Consequently, the cellular pool of available GKS proteins is diminished in IRGM-deficient cells and the efficiency of PV targeting is reduced.
In addition to GKS proteins, GBP proteins also bind to PVs. The delivery of GBP proteins to PVs requires the presence of a previously unknown IFN-inducible cofactor(s) [32], [33]. Here, we identify IRGM proteins as one such co-factor. We propose that IRGM proteins promote the recruitment of GBP proteins to PVs by a mechanism similar to that which regulates the subcellular localization of GKS proteins. GBP proteins bind to lipids as activated oligomers [36], [37] and GBP mutants deficient in GTP binding fail to localize to PVs [29], [33]. In this study, we demonstrate that IRGM proteins on LDs prevent GBP recruitment, suggesting that IRGM proteins interfere with the ability of GBP proteins to transition into the GTP-bound, oligomeric state. In support of this hypothesis, we found that Gbp2 forms high molecular weight aggregates in the absence of Irgm1/m3 (A.S.P. and J.C., unpublished results). Therefore, IRGM proteins may promote GBP recruitment to PVs by maintaining a pool of GDP-bound, monomeric GBP proteins that are able to diffuse to their target sites.
Additional evidence for functional interactions between the GBP and IRG protein families comes from the observation that one or more members of GBP protein family associate with Irgb6 in complexes [38]. Deletion of the chromosomal region containing the genes Gbp1-3, Gbp5 and Gbp7 causes a partial defect in the recruitment of Irgb6 and Irgb10 but not Irga6 to T. gondii PVs [38], suggesting that physical interactions between specific GBP proteins and Irgb6/b10 promote targeting of Irgb6/b10 to PVs. In contrast to the partial GKS targeting defects of Gbp-deficient cells, Irgm1/m3−/− cells display a nearly complete deficiency in recruiting either Irgb10 (Figure 10E) or Gbp2 protein to T. gondii PVs (Figure 10F). The combined results from both studies suggest that Irgm1 and Irgm3 regulate the recruitment of both GKS and GBP proteins to PVs, while one or more PV-targeted GBP proteins augment the recruitment of a subset of GKS proteins through direct physical interactions. In summary our data demonstrate that IRGM proteins orchestrate the proper targeting of antimicrobial GBP and GKS proteins away from “self” membranes and towards “non-self” PVs.
MEFs derived from wildtype, Irgm1−/−, Irgm3−/− and Irgm1/m3−/− mice were previously described [39], [40]. MEFs and African green monkey kidney Vero cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% heat-inactivated fetal bovine serum (FBS) (Denville and Life Technologies). Primary murine bone marrow macrophages were isolated from the tibia and femurs of 2- to 4-months-old mice as described before [41]. C. trachomatis LGV-L2 was propagated as described [39]. A previously described GFP-expression vector [42] was introduced into LGV-L2 for visualizing C. trachomatis at 2 hpi. GFP-expressing Toxoplasma gondii tachyzoites of the type II strain Prugniaud A7 were a generous gift from Dr. John Boothroyd (Stanford University, Stanford, CA) [43]. Infections with C. trachomatis were performed at a nominal multiplicity of infection of 1–5 as described [39]. For T. gondii infections cells were incubated overnight with or without 200 U/ml of IFNγ and asynchronously infected with tachyzoites at a nominal multiplicity of infection of 5–10 for thirty minutes.
For lipid loading experiments, OA (Sigma) was precomplexed with fatty acid-free BSA (Sigma) in PBS and emulsified by sonication. OA was added to growth media at final concentration of 100 µM for immunofluorescence experiments. LDs were isolated from MEFs as described before [44] with minor modifications as outlined here. Cells were grown in 150 mm dishes in DMEM +10% FBS and incubated with OA at 300 µM in the presence or absence of 100 U/mL of IFNγ for 14 h before harvesting LDs. Cells were washed with PBS and collected in 5 ml TNE buffer [20 mMTris-HCl (pH 7.5), 0.15 M NaCl, and 1 mM EDTA] containing protease inhibitors (Roche Diagnostics). Cells were lysed on ice with ∼30 strokes/150 mm dish in a Dounce homogenizer and 80 µl of total lysates were collected from each sample and stored at −20°C for Western blotting. Cell lysates were then adjusted to 0.45 M sucrose, overlaid with 2 ml each of 0.25 M, 0.15 M, and 0 M Sucrose/TNE and centrifuged at 30,000 rpm for 90 min at 4°C in an SW41 rotor (Beckman Coulter). The floating LD-enriched fat layer was collected, diluted in TNE, and refloated at 47,000 rpm for 45 min in a TLA55 rotor (Beckman Coulter). LDs were collected, and lipids were extracted with 4 volumes of diethyl ether. Delipidated proteins were precipitated with ice-cold acetone for 1 h, solubilized in 0.1%SDS and 0.1 N NaOH, and normalized for total protein content by Bradford assay before SDS-PAGE and immunoblot analysis. Following protein transfer to nitrocellulose membranes, membranes were incubated with antibodies as listed below. Densitometric analyses for protein quantification in Western blots were carried out using Image J 1.45 s software.
Immunocytochemistry was performed essentially as described previously [39], [44]. Cells were washed thrice with PBS, pH 7.4 prior to fixation. Cells were fixed with 3% formaldehyde and 0.025% glutaraldehyde for 20 min at room temperature (RT) in all experiments that visualized LDs. T. gondii-infected cells were fixed with 4% paraformaldehyde in PBS, pH 7.4, cells for 20 min at RT. In all experiments involving LD staining, fixed cells were permeabilized/blocked with 0.05% (v/v) saponin and 2% BSA/PBS (SBP) for 30 min at RT. When preserving LD structures was not required, fixed cells were permeabilized in 0.1% (v/v) Triton X-100 in PBS for ten minutes, blocked for 1 h with 2% (w/v) BSA (Equitech-Bio Inc.) in PBS, and then stained with various primary antibodies, followed by Alexa Fluor-conjugated secondary antibodies (Molecular Probes/Invitrogen). Working solutions of antibodies and BODIPY 493/503 (10 µg/ml) (Invitrogen) for immunofluorescence were prepared in SBP (for LD visualization) or in 2% (w/v) BSA/PBS (for all other experiments). Nucleic and bacterial DNA were stained with Hoechst 33258 according to the manufacturer's protocol. Mitochondria were visualized using MitoTracker Red CMXRos (Invitrogen) according to the manufacturer's instructions. Stained cells were washed with PBS, mounted on microscope slides with FluorSave (Calbiochem) or ProLong Gold (Invitrogen), and allowed to cure overnight. Cells were imaged using either a Zeiss LSM 510 inverted confocal microscope or a Zeiss Axioskop 2 upright epifluorescence microscope. Co-localization of proteins with PVs was quantified in at least 3 independent experiments. In each experiment at least ten randomly selected fields were imaged for each condition for each cell type. Differential interference contrast images were used to identify extracellular T. gondii tachyzoites because the vacuoles typically contained only one parasite under the experimental conditions used. The fraction of Gbp2- or Irgb10-positive vacuoles was determined for each field by dividing the number of Gbp2- or Irgb10-labeled vacuoles by the total number of vacuoles. Co-localization with C. trachomatis inclusions was quantified using the identical approach. Co-localization of Irgb10 and Gbp2 with LDs was quantified using MBF ImageJ software (developed by Wayne Rasband, National Institutes of Health, Bethesda, MD; available at http://rsb.info.nih.gov/ij/index.html). Images were pre-processed to correct uneven illumination and to minimize noise and background. The co-localization rates were measured based on Manders' coefficient, which varies from 0 to 1. A coefficient value of zero corresponds to non-overlapping images while a value of 1 reflects 100% co-localization between the images being analyzed. To perform line tracings, i.e. analyze the fluorescence signal intensity profiles of pixels along a selection from images, we used ImageJ software.
Cells were treated with or without IFNγ (200 U/ml) in the absence or presence of OA for 20 to 24 hours. Where indicated, BAF was supplemented at a final concentration of 100 nM at 12 hours post IFNγ activation. LD mass was determined by Flow Cytometry as described elsewhere [26]. Briefly, after fixing the cells with 2% PFA, cells were stained with BODIPY 493/503 at 5 µg/ml in FACS buffer (PBS, 1% BSA and 0.1% NaN3) for 30 minutes and washed with FACS buffer prior to analysis.
The primary antibodies used included anti-Irgm1 mouse monoclonal antibody 1B2 [11] at 1∶10; anti-Irga6 mouse monoclonal antibody 10D7 [12] at 1∶10; anti-Irgb10 rabbit polyclonal antiserum [39] at 1∶1000; anti-Irgb6 rabbit polyclonal antisera [27] at 1∶1000; anti-Irgm3 rabbit polyclonal antisera [45] at 1∶1000; mouse monoclonal anti-Irgm3 antibody (BD-Transduction Labs) at 1∶300; FITC-labeled mouse monoclonal anti-C. trachomatis MOMP [39] at 1∶200; rabbit anti-IncG [46] at 1∶50; anti-V5 mouse monoclonal antibody (Invitrogen) at 1∶1000; anti-FLAG mouse monoclonal antibody F1804 (Sigma) at 1∶500, rabbit anti-Pmp70 (abcam) at 1∶500; anti-TIP47 polyclonal antisera (Proteintech) at 1∶1000; anti-p62/SQSTM1 rabbit polyclonal antibody (MBL International) at 1∶500; and anti-LC3 rabbit polyclonal antibody (MBL International) at 1∶1000. An affinity-purified polyclonal rabbit anti-Gbp2 antibody was generated against the peptide EVNGKPVTSDEYLEHS of Gbp2 and used at 1∶1000.
An Irgb10-GFP expression construct has been previously described [39]. Site-directed mutagenesis was performed using the QuikChange II Site-Directed Mutagenesis Kit (Agilent Technologies Inc.) to introduce the listed point mutations and deletions into the same construct. Standard cloning techniques were used to generate to insert DNA encoding dsRED and the yeast protein TyA into the listed GFP expression constructs. DNA oligonucleotides used for cloning are listed in Table 1. A previously described TyA expression construct [21], a kind gift from Dr. Stephen Gould, was used as a template for DNA amplification. The C57BL/6J-derived cDNAs of Irgm1, Irgm2 and Irgm3 were cloned into pcDNA3.1/V5-His-TOPO (Invitrogen) following the manufacturer's instructions. FLAG-tagged and GFP-tagged expression constructs of Gbp1 and Gbp2 and Gbp1 mutant variants have been previously described [33]. MEFs were transduced using the MSCV-based delivery system (Clontech) or transfected using Attractene (Qiagen) following the manufacturers' instructions.
Results are represented as means ± SD. All comparisons were evaluated for statistical significance through the use of unpaired two-tailed t tests. When necessary, significant differences between data points were highlighted and the level of significance was depicted as: *, p<0.05; **, p<0.01; and ***, p<0.005.
|
10.1371/journal.ppat.1002861 | Exon Level Transcriptomic Profiling of HIV-1-Infected CD4+ T Cells Reveals Virus-Induced Genes and Host Environment Favorable for Viral Replication | HIV-1 is extremely specialized since, even amongst CD4+ T lymphocytes (its major natural reservoir in peripheral blood), the virus productively infects only a small proportion of cells under an activated state. As the percentage of HIV-1-infected cells is very low, most studies have so far failed to capture the precise transcriptomic profile at the whole-genome scale of cells highly susceptible to virus infection. Using Affymetrix Exon array technology and a reporter virus allowing the magnetic isolation of HIV-1-infected cells, we describe the host cell factors most favorable for virus establishment and replication along with an overview of virus-induced changes in host gene expression occurring exclusively in target cells productively infected with HIV-1. We also establish that within a population of activated CD4+ T cells, HIV-1 has no detectable effect on the transcriptome of uninfected bystander cells at early time points following infection. The data gathered in this study provides unique insights into the biology of HIV-1-infected CD4+ T cells and identifies genes thought to play a determinant role in the interplay between the virus and its host. Furthermore, it provides the first catalogue of alternative splicing events found in primary human CD4+ T cells productively infected with HIV-1.
| Some previous studies have monitored HIV-1-induced gene expression in various host cell targets and tissues but the discrimination between productively infected cells and uninfected bystander cells represents a technical challenge yet to be solved. Consequently, data interpretation has always been biased towards the transcriptional response of a majority of uninfected bystander cells that were exposed to soluble factors released by virus-infected cells. Following the design of a unique and innovative molecular tool to identify cells productively infected with HIV-1 and the description of an efficient magnetic beads-based technique to separate them from uninfected bystander cells, we undertake this challenge and perform the first comparative whole-genome transcriptomic and large-scale proteomic profiling of both HIV-1-infected and uninfected bystander CD4+ T cells. We demonstrate herein that HIV-1- infected and uninfected bystander cells display distinctive transcriptomic signatures which might permit to identify new susceptibility and resistance factors.
| CD4+ T cells – the primary cellular target of HIV-1 – are progressively depleted over the course of infection. This long-term process culminates in the onset of AIDS, a condition in which the immune system is too weak to efficiently mount an effective defence against opportunistic pathogens. Yet, HIV-1 uses only 15 proteins to disable the natural immune defences and harness the host cell machinery to complete its replicative cycle. To do so, viral proteins interact with multiple cellular proteins, perturbing the normal flow of cellular processes. Moreover, the virus influence extends beyond the cells it infects. Indeed, the apoptosis rate of uninfected bystander CD4+ T cells is elevated in individuals carrying HIV-1 [1]. The dichotomy between uninfected bystander and HIV-1-infected CD4+ T cells is an important topic to study, as a deeper understanding of HIV-1 pathogenesis mechanisms might lead to new therapeutic approaches.
Powerful technologies developed in recent years have provided high-throughput tools to study cellular dynamics. Among them, microarrays allow for the quantification of expression levels of thousands of genes at once. Since the inception of this technology, few studies have used microarrays to characterize the effect of HIV-1 on various cell types that compose the immune system. However, productive infection rates in primary human cells such as CD4+ T lymphocytes are very low. As microarray technology captures the average transcriptomic profile of a cell population, achieving a high level of purification of subpopulations of interest is crucial to accurately quantify any possible virus-mediated changes in the host transcriptome [2]. We recently developed a new reporter virus system that allows the efficient separation of HIV-1-infected cells from their uninfected bystander counterpart in vitro [3]. In the current work, we used this unique and versatile tool to define the HIV-1-induced modulation of host gene expression in one of its most worrisome cellular reservoir. We hereby provide the first time-course comparative and comprehensive study of the effect of HIV-1 in productively infected versus uninfected bystander primary human CD4+ T cells using Affymetrix Human Exon arrays. We pinpoint understudied genes which seemingly play important roles in the subset of CD4+ T cells preferentially infected by HIV-1 and provide an overview of splicing events found in this subpopulation.
While an impressive amount of data has been gathered about HIV-1 and its putative relationships with various components of the immune system, the most favorable cellular microenvironment for HIV-1 establishment and replication within CD4+ T lymphocytes have never been fully elucidated. It is known that this retrovirus preferentially infects activated CD4+ T cells [4], but infection rates are very low even in this susceptible population. We confirmed this phenomenon using a replication competent reporter virus system containing all viral genes along with the small murine heat stable antigen (HSA) protein [3] (Figure 1A), which is rapidly expressed on the surface of virus-infected cells. PHA/IL-2-treated primary human CD4+ T cells exposed to this recombinant reporter virus reached 1.1, 5.8 and 6.6% of HSA-positive cells after 24, 48 and 72 h post-infection, respectively (Figure 1B, left panels). Although it is difficult to conclude that the enhanced percentage of HIV-1-infected CD4+ T cells during the first three days is strictly due to additional infection events and not due, at least partially, to cell death in cells not productively infected with HIV-1, it can be postulated that factors other than cellular activation are essential for the successful establishment of productive HIV-1 infection.
Primary human activated CD4+ T cells can be classified in distinct subsets which include naïve (Tnaïve/CD45RO−CD27+), intermediate memory (TIM/CD45RO+CD27+) and effector memory (TEM/CD45RO+CD27−) [5], [6]. We characterized the profile of CD4+ T cells most permissive to productive HIV-1 infection by staining for CD45RO, CD27 and HSA. Results show that HIV-1 displays a preference for TEM cells that is increasing over time (Figure 1B, compare columns showing mock, HSA− and HSA+) with enrichment ratios of 1.5, 1.9 and 2.3 in HIV-1-infected versus uninfected bystander cells at 24, 48 and 72 h post-infection. However, the phenotype does not fully account for HIV-1 selectivity as we find a significant amount of virus-infected Tnaïve and TIM cells, even though these subsets are less susceptible to productive HIV-1 infection compared to TEM cells (enrichment ratios of 0.7, 0.4 and 0.4 for Tnaïve and 1.1, 0.8 and 0.8 for TIM).
We next performed a comparative microarray analysis of mock-infected, uninfected bystander (HSA−) and HIV-1-infected primary human CD4+ T cells (HSA+) in an attempt to identify both virus-induced genes and intracellular environment most permissive to productive infection. To this end we magnetically separated virus-infected cells from three different donors on the basis of HSA expression from the whole cell population exposed to the R5-tropic reporter virus for 24, 48 or 72 h (Figure 1C). We next confirmed productive HIV-1 infection in isolated cell fractions using a spliced Tat-specific qRT-PCR based on the idea that such a sensitive technique allows the quantification of early expressed viral transcripts without detection of input viral RNA. Data shows a strong signal in the fractions containing HSA-expressing cells (called HIV Pos), while uninfected bystander cell fractions (called HIV Neg) and mock-infected (called Mock) displayed an almost complete absence of spliced Tat, thus indicating a very high degree of cell purification (Figure 1D).
Thereafter, we extracted total RNA from the studied cell fractions and performed transcriptome profiling using Affymetrix Human Exon arrays, which interrogate over 5 million probes spanning all exons in the human genome. A careful analysis of differentially expressed genes (DEGs) using a false discovery rate of 1% and a cut-off of 1.7 fold showed no effect of HIV-1 on the transcriptome of uninfected bystander cell population compared with mock-infected cells at any studied time point or in aggregate (Figure 2A, left panels). Comparison between mock-infected and HIV-1-infected cells revealed 287, 236 and 176 DEGs at 24, 48 and 72 h post-infection, respectively (Figure 2A, middle panels) while the aggregate comparison of all time points yielded 289 genes. Given that no DEG was identified in the uninfected bystander cell population, we then compared cells productively infected with HIV-1 to uninfected bystander cells to improve our statistical power, the rationale being that the separation procedure allows to isolate the small percentage of virus-infected cells by removing them from the uninfected bystander cell fraction (constituting the majority of cells), creating a better signal differential than the one obtained comparing with the mock-infected fraction. By doing so, we obtained 502, 366 and 323 DEGs (at 24, 48 and 72 h) in HIV-1-infected cells while the aggregate comparison yielded 464 DEGs (Figure 2A, right panels). Proportional Venn diagrams depict the relative distribution of DEGs and their evolution through time for these two comparisons (Figure 2B). An overview of all modulated genes is presented in Figure 2C as a hierarchical cluster. It can be concluded that most DEGs identified in HIV-1-infected target cells (Figure 2C) are stable at least during the studied period, whereas expression of a small cluster of genes increases significantly over time. Complete lists of DEGs that were found to be modulated are depicted in Dataset S1.
We broadly defined the characteristics of the dataset using the metadata clustering engine DAVID, which identifies enriched biological themes within a list of genes using various biological annotation sources [7]. Using the combined list of 835 DEGs identified in all comparisons performed, we found the following statistically over-represented categories: immune system process, cytokine-receptor interaction, regulation of leukocyte activation, Map kinase phosphatases, FOS/JUN related genes, positive/negative regulation of apoptosis and p53 pathways (Figure 3).
The first three identified categories overlap significantly and contain markers for Th1 (i.e. IFN-γ, TNF-α, TNF-β, IL-1A, IL-3, and TBX21/T-bet), Th2 (i.e. IL-4, IL-5, IL-10, and IL-13) and Th17 profiles (i.e. IL-17A, IL-17F, IL-21, IL-22, IL-23R, and IRF4), all of which are over-expressed in primary human CD4+ T cells productively infected with HIV-1 (Figure 3, top left). The expression patterns associated with these different profiles suggest that Th1 and Th17 cells are slightly more susceptible to infection with the studied R5-tropic reporter virus, as virus-infected cells express more IFN-γ and IL-17A than other cytokines (Figure S1). Other molecules associated with the effector phenotype and activated T cells in general are also differentially expressed (i.e. IL4R, IL18R, MCSF, GMCSF, ICAM-1, OX40, OX40L, CD27, and its counterpart CD70, CD80, CTLA4, CD69, CD40LG, IL2RA, FASLG, IL12RB2, IL18R1, IL18RAP, and IL-9). Their expression pattern is consistent with highly activated TEM cells being preferentially infected by HIV-1. Several Map kinase phosphatases involved in T-cell activation are also modulated in virus-infected cells (Figure 3, bottom right).
Fos and Jun bind together to form the AP-1 transcription factor [8], which is essential for the differentiation and proliferation of lymphocytes. Multiple AP-1 binding sites are found in the HIV-1 LTR [9] and it has been demonstrated that this transcription factor promotes viral transcription [10]. The DAVID analysis pinpointed a cluster of over-expressed Fos-related genes (i.e. Fos, Jun and related genes BATF3, ATF3, FOSB, FOSL1, and FOSL2) in virus-infected cells (Figure 3, bottom left). Interestingly, a promoter analysis of the genes modulated in HIV-1-infected cells reveals that 57% of DEGs identified in this study and 70% of core genes (differentially expressed at 3 time points) contain at least one binding site for AP-1 (Figure S2), confirming that this transcription factor is one of the core determinants of the cellular environment that is most favorable to productive HIV-1 infection in primary human CD4+ T cells.
Among DEGs found precisely in HIV-1-expressing cells are genes involved in both positive and negative regulation of apoptosis (Figure 3, top right). Most of these genes appear to be an integral part of the genetic programme engaged in virus-infected cells and are stable through time. However, expression a subset of genes increases rapidly at 48 and 72 h post-infection, thus suggesting that such DEGs are a consequence rather than the determinant of viral gene expression. Further analysis shows that these genes are implicated in p53-dependent apoptosis – specifically CDKN1A/p21, MDM2, GADD45A, GADD45G, TNFRSF10B (TRAIL), ATM, ZMAT3, PMAIP1 (NOXA), BCL2L11, PERP, TP53I3, RRM2B, and SESN3 (Figure 3, bottom right). Additionally, the p53 pathway is identified by the DAVID analysis as the most significantly over-represented ontology category among the 180 genes exclusive to the 48 h and 72 h time points (p<0.0001; data not shown). The expression pattern of these genes over time is in agreement with the numerous reports of p53-related apoptosis in HIV-1-infected cells following DNA damage caused by the virus integration process [11], [12], [13], [14], [15]. However, patterns of differential expression of p53-regulating genes emerge as early as 24 h, before HIV-1-induced apoptosis occurs. For example, MDM2, a factor responsible for the degradation of p53 [16], is over-expressed in virus-infected cells. On the other hand, ATM, a gene involved in phosphorylation and inactivation of MDM2 [17], is under-expressed during this same period. This suggests that cells with low potential for p53 activation are more susceptible to productive HIV-1 infection, perhaps due to their slower reaction time for triggering p53-dependent apoptosis following viral integration, giving the virus more time to replicate actively.
The DAVID analysis is useful for broad categorization but is ultimately insufficient to fully extract biological significance from the dataset. A bibliography analysis was thus performed to visualize known relationships between genes present in the dataset. Bibliosphere (Genomatix) extracts relationships from co-citation of gene names (including synonyms) in abstracts from Medline. The resulting network illustrated in Figure 4A is consistent with the previous analysis, as the already identified over-represented categories cluster together. Moreover, the analysis uncovered numerous DEGs closely related to these clusters, but missed by the DAVID analysis. For example, it pinpointed CDC25A, CDC25C, and CDC14A near the p53 genes cluster. The pattern of expression of these genes is consistent with cells being most permissive to HIV-1 infection when entering in the M-G2 phase of the cell cycle [18]. The graph can also help identify small clusters of genes that would otherwise have gone unnoticed. Notably, furin and PCSK5 proteases, which are both known to participate in the cleavage of viral glycoprotein gp160 [19], [20], [21], [22], show opposite expression patterns. Indeed, furin is over-expressed in virus-infected cells, while PCSK5 is under-expressed at the 24 h time point. This implies that, although both proteases can cleave the viral envelope in vitro, furin is favored in acute HIV-1 infection studies. The graph file is available as Dataset S2 and can be explored dynamically by using graph visualization software Gephi (http://gephi.org/).
It was recently suggested that poorly characterized genes deserve a more careful analysis [23]. We performed a careful literature analysis and found 178 unconnected genes for which no known relationship with other genes in the dataset is currently described and/or little information is available (Figure 4A). We repeated the DAVID analysis on this subset of genes and found two over-represented protein domains, i.e. Krüppel associated box (KRAB) and GTPase regulator activity (Figure 4B).
The KRAB domain is mainly found in KRAB-ZNFs, a large family of mammalian transcription factors responsible for negative regulation of transcription through chromatin remodelling via their association with KAP1 [24]. They have been recently implicated in control of endogeneous retroelements [25]. Most of the modulated KRAB-ZFPs are under-expressed in virus-infected cells, which might indicate that their absence creates a permissive environment for HIV-1, probably by playing a role in T-cell regulation. Two of the identified KRAB-ZFPs can modulate crucial components of HIV-1 and lymphocyte biology. Indeed, overexpression of ZNF383 inhibits the transcriptional activities of AP-1 [26], while ZNF675 can suppress TRAF6-induced activation of NF-κB and c-Jun N-terminal kinase [27]. Interestingly, the two members that are over-expressed in virus-infected cells show mutations in the MLE motif of their KRAB domain (i.e. ZNF79 and ZNF282) (Figure S3), a pattern associated with loss of repression potential for KRAB-ZFPs [28]. It should be noted that ZNF79 is also a known p53 target [29].
The second over-represented category contains GTPase regulators, mainly of the RHO/RAC family. These influence a variety of cellular processes such as endocytic trafficking, actin dynamics and cell growth by affecting the rate of GTP hydrolysis by GTPases [30]. Many of the identified members either participate in T-cell signaling responses or directly influence proteins known to be important for HIV-1 entry in human primary CD4+ T cells, such as HERC1 (under-expressed in virus-infected cells) that is known to regulate ARF6 [31], [32].
Amongst the list of DEGs in HIV-1-infected cells, most genes showing high levels of differential expression are poorly studied. Since we isolated cells productively infected with HIV-1 from the total population (primarily composed of uninfected bystander cells), under-expressed genes potentially impede HIV-1 at some step of its replicative cycle, while the reverse holds true for over-expressed genes. For example, TRIM22, known for its role in inhibition of HIV-1 transcription [33], is under-expressed in virus-infected cells. Of note TRIM22 RNA levels progressively return to normal, suggesting that HIV-1 can overcome its effects over time (Table 1). Transcription factors known to repress HIV-1 transcription such as YY1 [34] and BCL11B/CTIP2 [35] are also found to be under-expressed in virus-infected cells. MARCH8, a regulator of vesicular transport of proteins between cellular compartments, was recently identified in a large scale siRNA screen as a top candidate for the inhibition of HIV-1 infection [36] and is found under-expressed in virus-infected cells in the present dataset.
We hereby present a list of the most promising understudied genes in regard to their importance for HIV-1 pathobiology and T lymphocyte biology. For example the transcription factor ZBED2 is highly over-expressed in HIV-1-infected cells. Although little is known about its function, this gene has been reported to be over-expressed in differentiated T cells [37], [38]. Its precise direct and indirect roles in the regulation of HIV-1 expression or lymphocyte differentiation remain to be more clearly defined. GJB2 and GJB6 are connexins that are members of the gap junction protein family involved in the formation of cell-cell channels [39]. It is possible that they play a role in viral entry or cell-cell transmission. GSDMB and DFNA5 are two members of the gasdermin protein family – DFNA5 has been recently shown to participate in p53-dependent apoptosis [40]. MYOF is a membrane-associated protein involved in both caveolin and clathrin-mediated endocytosis pathways along with membrane resealing after damage [41]. This gene is highly over-expressed in HIV-1-infected cells and could participate in the repair of the plasma membrane following the budding of a multitude of virions, allowing the cell to live longer and to produce more viruses. The expression pattern of CLC (also called galectin-10) differs from that of other modulated genes. CLC mRNA levels are significantly diminished in virus-infected cells whereas they steadily increase over time in uninfected bystander cells (although not reaching statistical significance). CLC has been shown to be a crucial determinant of Treg suppressor function [42]. PCR quantification for the aforementioned genes reveals excellent concordance with the microarray data (Table 1). We are currently investigating the precise role of some of these genes in the HIV-1 infection process and T-cell biology.
Affymetrix Exon arrays allow for the detection of alternative splicing events. Using the PECA-SI algorithm [43], we detected 323, 129 and 107 differential splicing events in transcripts of virus-infected versus uninfected bystander CD4+ T cells at 24, 48 and 72 h post-infection, respectively (Figure 5A and Dataset S3; see Methods for filtering parameters). Similarly to gene expression, more events were detected when comparing HIV-1-infected to uninfected bystander cells than to mock-infected ones, and no alternative splicing events were detected in the uninfected bystander population.
We looked for a splicing event known to occur in HIV-1-infected cells. In T lymphocytes, CD45 isoforms are a marker of naive (CD45RA) and memory phenotypes (CD45RO) – the latter being preferentially infected by HIV-1 [44]. This is indeed reflected in the data, as exon 4 of CD45 (corresponding to CD45RA) is under-expressed in virus-infected cells, implying that this fraction contains more CD45RO (Figure 5C, top).
We next focused our attention on events susceptible to be important in the biology of both HIV-1-infected and CD4+ T cells. Multiple events are detected in the C-terminal portion of inositol 1,4,5-triphosphate receptor (ITPR1), yielding a short isoform containing only the calcium-channel domains of the protein (Figure 5C, center). ITPR1 plays a role in lymphocyte activation [45] and this isoform could represent a constitutionally active form of the receptor present in highly activated and/or memory cells. RUNX1 (AML1/EVI1) is a master regulator of hematopoietic development important in T-cell differentiation [46] and is known to have multiple isoforms arising from alternative promoter usage [47] (Figure 5C, bottom). We identify a short isoform of RUNX1 being enriched specifically in virus-infected cells. The isoforms of RUNX1 have been implicated in different stages of haematopoiesis [47]. However their precise role in T cells or potential interaction with HIV-1 are still unclear. Other interesting alternative splicing events include an isoform of LEF1, a transcription factor implicated in the regulation of HIV-1 transcription [48], and EVI5, an oncogene implicated in cell cycle control [49] – both are RUNX1 interaction partners. qRT-PCR confirmation for all the aforementioned events was performed and results are highly consistent with microarray-acquired data (Figure 5B).
In this manuscript, we provide the first comparative analysis of exon-level transcriptomic profiles between HIV-1-infected primary human CD4+ T cells and their uninfected bystander cell counterparts. By doing so, we define the virus-induced genes and microenvironment most favorable to allow productive HIV-1 infection and show that even within a population of activated CD4+ T cells, the permissive environment for HIV-1 infection is very specific. The profile of virus-infected cells is consistent with activated/effector memory CD4+ T cells expressing high levels of cytokines. We found that Th1 and Th17 were to some extent more permissive to virus infection in this specific in vitro experimental setting. The expression patterns of most genes identified in this study are in agreement with the current literature in regard to HIV-1 and lymphocyte activation status, which provide significance to our observations. However, it must be emphasized while we rediscovered many well-studied genes and pathways known to be important for HIV-1, the precise function of a large number of other genes that were identified in this work is still currently poorly described. We believe that efforts should be made to understand their function, as yet unexplored avenues might allow deepening our understanding of the interplay between HIV-1 and lymphocyte biology.
The design of our microarray experiment captures both the transcriptomic portrait of highly permissive cells and the changes induced by the virus itself. While a clear discrimination between theses two events is complicated by confounding factors such as asynchronous infection and a potential cell death of some CD4+ T cells, a longitudinal analysis of the data allows us to achieve a comparison of gene expression patterns between HIV-1-infected and uninfected bystander cells. We concluded that genes for which levels change significantly over time in virus-infected cells are directly modulated by HIV-1. As an example, the p53 apoptosis pathway is clearly induced in virus-infected cells at 48 and 72 h post-infection. However, the vast majority of differentially expressed genes are relatively stable over time in virus-infected CD4+ T cells and thus define the transcriptomic programme of a subpopulation preferentially infected by HIV-1. In this context, over-expressed genes are potential viral permissiveness factors, while underexpressed genes are candidate restriction factors. Functional assays are currently underway to determine which of the newly identified candidate genes play a key role in HIV-1 and/or lymphocyte biology.
One of our primary objectives was to identify changes occurring in uninfected bystander CD4+ T cells exposed to HIV-1, as previous studies have reported dramatic effects in cells exposed to virions or its components. To this end, we purposely left the initial viral inoculum in our purified CD4+ T cell populations to allow for putative virus-mediated signal transduction events. We were surprised to find no significant changes in the transcriptome of the uninfected bystander cell population at least at early time points after HIV-1 infection (i.e. 24, 48 and 72 h). It is possible that the presence of cells other than CD4+ T lymphocytes is required to mediate changes in gene expression in uninfected bystander cells. We nonetheless detected an increase of apoptosis in uninfected bystander cells at late time points (data not shown). This could imply that gene regulation is not necessary for apoptosis induction. For example, direct caspase activation induced by FAS or TRAIL ligation could explain this phenomenon. It should also be noted that we used an R5-tropic variant of HIV-1, as this tropism is more representative of early infection events. However, it cannot be excluded that an X4-tropic virus would have an effect on the transcriptome of uninfected bystander cells, as this variant characteristic of late-stage infection is known to have higher apoptosis-inducing activity [50]. On a similar note, a somewhat different HIV-1-mediated gene expression pattern might be obtained when using a distinct R5-tropic virus. Additional experiments are needed to solve these issues.
As we demonstrated in this work, the isolation of human primary CD4+ T cells productively infected with HIV-1 is a powerful approach which amplifies the power of transcriptome analysis. We believe that further dissection of virus-infected CD4+ T cell subtypes could yield even more information, as the profile we obtained is undoubtedly an average of different types of susceptible subpopulations. While observations made in this manuscript describe the relationship between HIV-1 and CD4+ T cells, it is in the absence of the multitude of other factors influencing dynamics of infection in vivo. Therefore, transcriptomic analysis of virus-infected cells in more complex experimental settings such as total peripheral blood mononuclear cells or humanized mice models would provide additional insight in the intricate relationship between the virus and its host environment. We hope that the data provided here can serve as a roadmap to focus efforts on neglected aspects of T-cell and HIV-1 biology, leading to a better understanding of the complex relationship between the virus and its host.
Human peripheral blood mononuclear cells were obtained from healthy blood donors, in accordance with the guidelines of the Bioethics Committee of the Centre Hospitalier de l'Université Laval Research Center, by density-gradient centrifugation on Ficoll-Hypaque (Wisent, St-Bruno, QC). All blood donors were informed and agreed to a written consent prior to blood donation. Cells were plated in 75-cm2 flasks at 15×106/mL for 2 h. The non-adherent cells from the supernatant were enriched in CD4+ T cells with the human CD4+ T Cell Enrichment Kit (Stemcell Technologies Inc., Vancouver, BC). The purity obtained was routinely higher than 98% of CD3+CD4+ T cells. Cells were cultured at 2×106/ml in RPMI-1640 medium (Invitrogen, Burlington, ON) supplemented with 10% foetal bovine serum (Invitrogen), L-glutamine (2 mM) (Wisent), penicillin G (100 U/ml), streptomycin (100 µg/ml) (Wisent) and primocine (Amaxa Biosystems, Gaithersburg, MD). Cells were rested for 24 h after isolation and treated with phytohemagglutinin-L (PHA) (1 µg/ml) (Sigma-Aldrich, St-Louis, MO) and recombinant human IL-2 (rhIL-2) (30 U/ml) (AIDS Research and Reference Reagent Program, Germantown, MD) for 2 days at 37°C under a 5% CO2 atmosphere prior to HIV-1 infection and rhIL-2 was refreshed when the medium was changed. Human embryonic kidney 293T cells were maintained in DMEM (Invitrogen) supplemented with 10% FBS, L-glutamine, penicillin G and streptomycin.
NL4-3 BAL-IRES-HSA virions were described previously [3] and produced in 293T cells using a commercial calcium phosphate kit (CalPhos Mammalian Transfection kit, Clontech, Palo Alto, CA). Cell-free supernatants were ultracentrifuged to eliminate free p24. Finally, samples were aliquoted before storage at −85°C. A homemade ELISA test was used to normalize the p24 content in all viral preparations [51]. Virus infection was achieved by inoculating primary human CD4+ T cells with a fixed amount of reporter virus standardized in term of p24 (i.e. 10 ng of p24 per 1×105 target cells). All virus preparations underwent a single freeze-thaw cycle before initiation of infection studies. A control consisting of mock-infected cells was obtained by transient transfection of 293T cells with an equimolar amount of pCDNA 3.1 (control empty vector). Next, purified CD4+ T cells were treated with a volume of supernatant from 293T cells transfected with pCDNA 3.1 similar to the one used for HIV-1 infection experiments. A total of 100×106 primary human CD4+ T cells were exposed to NL4-3 BAL-IRES-HSA and 30×106 target cells were used for mock-infected controls for each of the three donors tested. In order to get about 5×105 CD4+ T cells productively infected with HIV-1 (i.e. HSA+), a total of 50, 30 and 20×106 cells were used to isolate an average of 5×105 virus-infected cells at 24, 48 and 72 h post-infection, respectively.
Following infection of primary human CD4+ T cells with the fully competent HSA-encoding virions, HIV-1-infected and uninfected bystander cell populations were isolated using a previously described protocol with slight modifications [3]. In brief, all magnets were pre-cooled for one hour at 4°C. Only the first HSA-negative fraction was used to obtain uninfected bystander cells because subsequent negative fractions have an increased risk of containing HIV-1-infected cells (unpublished data). Mock-infected cells were subjected to the same procedure as the uninfected bystander fraction.
A protocol was designed for RNA isolation, as neither Trizol nor silica-based column methods could yield sufficient amounts of high quality RNA from HIV-1-infected fractions. Indeed, the positive fractions contain magnetic beads which were found to interfere with the Trizol reagent (See Protocol S1).
Flow cytometry analyses were performed with 5×105 cells that were incubated with 100 µl of wash buffer (PBS [pH 7.4], BSA 1%, and EDTA 2 mM) containing a saturating amount of a monoclonal rat anti-mouse HSA antibody (clone M1/69, PE-coupled, BD Biosciences, Mississauga, ON), anti-CD4, anti-CD3, anti-CD27, anti-CD45RO (all from BD Biosciences) or a corresponding isotype-matched control antibody for 30 min at 4°C. Cells were then washed, fixed with 2% paraformaldehyde for 30 min at 4°C and analyzed on a cytofluorometer (FACSCanto, BD Biosciences). Further analyses were performed using FCS Express V3.0 software (De Novo Software, Los Angeles, CA).
A total input of 200 ng of RNA was used to prepare targets for array hybridization using the Ambion WT Expression Kit (Applied Biosystems, Austin, TX). Data was normalized using RMA at the gene and exon level with Affymetrix Power Tools – core-level probe definition were used in both cases for results presented in this manuscript. Bioconductor package limma [52] was used to find modulated genes – an FDR of 1% and a fold change of 1.7 were used to filter the lists. A minimum signal filter of 100 on the average of three replicates was also applied. Time-wise and aggregate comparisons were done between infected, bystander and mock treated cells.
DAVID analysis was done using version 6.7 with standard parameters using the following categories: GOTERM_BP_FAT, GOTERM_MF_FAT, KEGG_PATHWAY, BIOCARTA, SP_PIR_KEYWORDS, UP_SEQ_FEATURE, SMART, INTERPRO, UCSC_TFBS. Bonferroni corrected p-values<0.001 were considered as significant.
Bibliosphere analysis was performed using version 7.24 with the compiled list of 835 modulated genes identified by limma. We settled on the presence of three co-citations or more in the same sentence in a PubMed abstract as a relationship criterion. We exported the data to Gephi (http://gephi.org/), a powerful and interactive network visualization and exploration platform for further analysis using the Gefx library. Nodes were arranged using a directed force algorithm (Force2), colored according to fold change between uninfected bystander and HIV-1-infected cells at 24 h and sized according to the –log10 of the p-value for the same comparison – these can be dynamically changed at will using the original Gephi file containing the graph and all associated data, available as Dataset S2.
Alternative splicing analysis was performed with PECA-SI. The following filters were applied: a threshold of splicing index of at least 1.7 fold, a significant DABG signal (p<0.001) in at least 3 groups (equivalent to 9 chips), probes contained in exonic regions and an FDR of 1%. Data was dynamically overlaid and visualized on both Annmap (http://annmap.picr.man.ac.uk/) via Bioconductor xmapcore and xmapbridge packages and Splice Center [53].
qRT-PCR. qRT-PCR against the Tat-spliced isoform was performed to quantify the level of enrichment of HIV-1-infected CD4+ T cells achieved in the HSA-positive fraction and their absence in the HSA-negative fractions. TaqMan RNA-to-CT 1-Step Kits from Invitrogen was used for quantification. The following probe and primers were used in our study: probe, TATCAAAGCAACCCACCTCC, forward primer, GAAGCATCCAGGAAGTCAGC, reverse primer, CTATTCCTTCGGGCCTGTC. PCR was performed under standard TaqMan cycling conditions using a Rotor-gene 3000 (Corbett Life Science, San Francisco, USA). SYBR Green detection was used for all subsequent PCR targets. qRT-PCR confirmation of genes was performed with the Power SYBR Green RNA-to-CT 1-Step kit from Invitrogen. Expression values were normalized to 18S and quantification was performed using the CT method. Primers for gene expression and alternative splicing confirmations are shown in Table S1.
|
10.1371/journal.pcbi.1000109 | The Generation of Promoter-Mediated Transcriptional Noise in Bacteria | Noise in the expression of a gene produces fluctuations in the concentration of the gene product. These fluctuations can interfere with optimal function or can be exploited to generate beneficial diversity between cells; gene expression noise is therefore expected to be subject to evolutionary pressure. Shifts between modes of high and low rates of transcription initiation at a promoter appear to contribute to this noise both in eukaryotes and prokaryotes. However, models invoked for eukaryotic promoter noise such as stable activation scaffolds or persistent nucleosome alterations seem unlikely to apply to prokaryotic promoters. We consider the relative importance of the steps required for transcription initiation. The 3-step transcription initiation model of McClure is extended into a mathematical model that can be used to predict consequences of additional promoter properties. We show in principle that the transcriptional bursting observed at an E. coli promoter by Golding et al. (2005) can be explained by stimulation of initiation by the negative supercoiling behind a transcribing RNA polymerase (RNAP) or by the formation of moribund or dead-end RNAP-promoter complexes. Both mechanisms are tunable by the alteration of promoter kinetics and therefore allow the optimization of promoter mediated noise.
| Noise in gene expression is important for phenotypic variation among genetically identical cells. The gene expression will be particularly sensitive to noise in transcription initiation. Transcription initiation from a given promoter involves multiple steps, each of which could be rate limiting. In this paper we discuss how transcription initiation could come in bursts, separated by long periods where the promoter is inactive. Our results are compared to recent data of Golding et al. (2005), which suggest that transcriptions from some prokaryotic promoters occur in a highly irregular burst-like fashion. We show that the observed bursting could be caused by one of two alternate mechanisms. One possibility is that changes in supercoiling induced by previous RNA polymerase can help a subsequent RNAP to enter directly into open complex. Another possibility is that an RNAP at the promoter sometimes forms a dead-end complex, and thereby occludes the promoter for a sizeable amount of time.
| Cellular processes involve stochastic reactions between limited numbers of molecules, and therefore are subject to random noise. The existence of noise in the intracellular concentration of various species has been highlighted in a number of natural and engineered genetic circuits [1]–[6], which has been coupled with an increasing focus on the theory of how noise might be controlled or exploited by the cell.
Gene expression is perhaps the most important stochastic process in the cell. Transcription involves the production of small numbers of mRNAs, which are then translated multiple times, creating and amplifying noise in protein concentrations. Therefore, the probability distribution underlying the timing of transcription initiation is important for understanding cellular dynamics. A distribution where initiations are evenly spaced will result in less noise and a more uniform cell population. In contrast, a highly variable rate of initiation will produce large fluctuations that can lead to heterogeneous behavior across populations of genetically identical cells. This variability is important to allow populations of unicellular organisms to cope with variable environments [1],[5]. Another example is the spontaneous induction of ‘non-inducible’ prophages such as P2 [7], where stochastic flipping of a genetic switch allows a low rate of transition from lysogeny into lytic development. Noise in transcriptional initiation also has implications for transcriptional interference between convergent promoters [8].
Bertrand [9] and colleagues have developed a system where an mRNA containing multiple MS2 binding sites can be visualized by the binding of MS2-GFP fusion proteins to the mRNA. Golding and colleagues [10] placed such an mRNA under the control of the Plac/ara promoter in E. coli and could thereby detect production of individual mRNAs. When the promoter was induced, transcription was observed to occur in an unexpectedly irregular fashion, with bursts of transcription separated by long periods of inactivity. This phenomenon was called transcriptional bursting. The bursts of activity (on-periods) lasted an exponentially distributed amount of time, with a mean of 6 minutes at 22°C. During an on period a geometrically distributed number of transcripts are produced in rapid succession, with a mean of 2.2 transcripts per on-period. The long periods without transcription (off-periods) were also exponentially distributed, with a mean of 37 minutes. Golding et al. also report that similar behavior is seen with the PRM promoter of phage lambda.
Golding et al. [10] showed that this behavior was inconsistent with transcription occurring as a Poisson process. Here we consider the McClure model of transcription initiation [11]–[13], a more general model of transcription initiation, and show that it is still unable to reproduce the transcriptional bursting observed by Golding et al. We then consider current hypotheses for the mechanism of transcriptional bursting and find them wanting. Finally we propose two novel hypotheses for the mechanism behind transcriptional bursting, demonstrating that they are able to explain the results of Golding et al.
Golding et al. showed that their results were not consistent with transcription initiation being a single Poisson process. By considering the McClure model of transcription initiation (Figure 1A) we show that initiation as a single Poisson process is a special case where only one step is rate limiting, and that while the more general case is not a single Poisson process it is still unable to fit the results of Golding et al.
In prokaryotes, the initiation of transcription requires the binding of an RNAP to the promoter, the isomerisation of the RNAP through several intermediate forms, rounds of abortive initiation and then finally release from the promoter. Here we consider the McClure model of transcription [11]–[13] (Figure 1A), where transcription initiation requires three steps: RNA polymerase (RNAP) binding to the promoter to form a closed complex, followed by isomerisation of the closed complex to an open complex in which the DNA at the promoter is melted, and the escape of the open complex to form an RNAP complex engaged in elongation of the transcript. The closed complex is assumed to be in rapid equilibrium with free RNAP, while isomerisation and escape are treated as being slower and irreversible. This model is a simplified but useful version of the full kinetics of initiation.
The kinetics of each elementary reaction in initiation determines the final distribution of transcription initiation. Transcription is often treated as a Poisson process, i.e. the probability of initiation at a given moment is a constant, which results in an exponential distribution of times between transcripts. Golding et al. were able to show through several methods that the distribution of transcription initiation was non-Poisson. However, the exponential distribution is a special case where there is only one rate limiting step in the initiation of transcription.
For the analytical analysis of the McClure model, we make the assumption that the rates of binding kb and unbinding ku of the closed complex are relatively fast, and therefore that there are only two kinetically significant steps, isomerisation of the closed complex to an open complex, and promoter escape by the open complex. We assume that each step is elementary, i.e. that it can be approximated as a single chemical reaction. We also ignore the effect of self-occlusion, where an RNAP prevents further initiation at the promoter until it has transcribed far enough to no longer occlude the promoter (50 bp), as the time needed to transcribe this distance (1–4 seconds) is negligible compared to the time between initiations in the Golding et al. experiments. The average time needed to complete the first step, to, is therefore to = (1+K)/O, where K = ku/kb is the equilibrium constant of dissociation for the closed complex and O is the rate of transition from closed to open complex. The inverse of the rate of the open to elongating transition (E) gives the average time needed for the second kinetically important step, tE (Figure 1A). The average time taken for initiation (and therefore the time gap between initiations, 〈Δt〉, with 〈…〉 indicating the average) is the sum of two exponentially distributed random variables, 〈Δt〉 = 〈τO+τE〉. The probability distribution of time gaps between initiations is given by(1)for tO≠tE. For tO = tE = t, we get(2)
In the case where one step is much slower than the other (Class I), there is only one rate-limiting step in initiation and the distribution of Dt approaches a single exponential with mean tL = max(tO,tE) (Equation 1; Figure 1B), i.e. it approaches a single Poisson process. Here, the data points in Figure 1B) have been obtained by simulating the model of the promoter in Figure 1A) using the Gillespie algorithm [14], which stochastically determines the next reaction to occur and the time interval between reactions based on the given rates. The other extreme, where tO = tE (Class II), is shown in Figure 1C. In Class II, the chance of rapid successive firings faster than the average (Dt<<tO+tE) is smaller than for a Class I promoter, as for a Class II promoter a low Dt requires both the isomerisation and the escape to productive transcription to occur in rapid succession, whereas for a Class I promoter a low Dt requires the rapid occurrence of the rate limiting step only. As a consequence the distribution in Class II shows a peak at non-zero Dt. Promoter models that specify more kinetically significant reaction intermediates produce more extreme versions of the Class II distribution, with a larger peak centered around 〈Δt〉, resulting in more regular firing intervals.
The Class I type promoter shows the most fluctuation in Dt, and the effect of adding more kinetically significant intermediate steps is to reduce the amount of variability in Dt. Therefore neither the standard model nor models that take into account more intermediates can reproduce the bunched activity observed by Golding et al. [10], which show greater fluctuations in Dt than a Poisson process. In order to reproduce the bunched activity, it is necessary to consider a model with a branched pathway, where the system can go into either an active state or an inactive state with a switching mechanism between them.
Here we consider several hypotheses for the mechanism of transcriptional bursting and argue that they are unlikely to be correct. The promoter used by Golding et al. [10], Plac/ara, can be repressed about 70-fold by the lac repressor and activated about 30 fold by AraC [15]. Therefore, a simple hypothesis put forward by Golding et al. is that the silent periods are periods where the lac repressor is bound to the promoter, and the bursts are periods of activity when the promoter is free. However, the mean duration of off-periods is 37 min while on periods are only 6 min in duration, despite the fact that the promoter has been fully induced by 1 mM IPTG. It seems impossible for the lac repressor to remain bound to the DNA for 37 minutes under these conditions; especially considering that 1 mM IPTG derepresses the lac promoter in less than 5 sec [16].
A similar idea is that the off-periods represent periods where AraC is not bound to the promoter [10]. To make this feasible the on rate for AraC in an E.coli cell would have to be exceedingly small given the large off periods. This is unreasonable in view of the high association rate for AraC to other operators [17]. Presumably association rate is diffusion limited, meaning that it would take one AraC molecule less than a minute to bind to the operator [18]. In conclusion we find it unlikely that binding AraC is sufficient to produce bunched activity.
Another hypothesis put forward by Golding et al. is that RNAP might be able to re-initiate after termination, aided by the retention of sigma factor during transcription [19]. Presumably the RNAP would have to be positioned to rebind to the same promoter after termination for re-initiation to occur with any reliability, and it is not clear how this would be caused. One possibility is that a transcription factor might remain in contact with both the RNAP and the promoter via a DNA loop. This would render the promoter unavailable during transcription, which has some support from the data in that the lengths of the observed on-periods were approximately equal to the number of initiations multiplied by the time taken to transcribe the reporter mRNA for both Plac/ara and PRM (Golding, private communication), which would be expected if transcription does not occur simultaneously. However, this data is somewhat anecdotal, and stands in contradiction to the simultaneous transcription observed with electron microscopy [20]. Also, this mechanism requires binding of a closed complex to the DNA to be the rate limiting step that causes the 37 minute long off-period, and we consider it unlikely that simple recognition of the promoter by RNAP would take this long, especially given that closed complex formation is often thought to be a rapid equilibrium process.
Multiple RNAP can cooperate to overcome pause sites [21]. It might therefore be possible that the burst is due to multiple RNAP building up at a pause site and overcoming it together. However, this would require the RNAP to pause for a length of time on the same scale as the off-period; such an extreme pause is unlikely given that even the strongest pauses measured in vitro only last for around one minute.
Bursting could also result if there were distinct regions of high and low transcriptional activity within bacteria, akin to the idea of transcription factories in eukaryotes, and the promoter moved in and out of these regions on a slow time scale [22],[23]. Although this is an interesting possibility, not enough is known to evaluate such a mechanism in bacteria in much detail.
Fluctuations in the availability of free RNAP within the cell could contribute to variable initiation rates but it is difficult to see how such severe and long-lasting fluctuations capable of producing extended periods of complete inactivity could occur in cells where ∼3000 RNAPs [24] produce >105 RNAs per generation.
There is both theoretical [25] and experimental evidence [26],[27] that an elongating RNAP can increase the negative supercoiling of the DNA behind it.
Promoters can be very sensitive to supercoiling; for example, in vitro the activity of the LacP promoter increases by more than a factor of 10 when the super-coilings is changed from zero to −0.065 (which is the average supercoiling of DNA in E. coli) [26]. We therefore consider it a possibility that the bursts of transcription might be caused by a transcribing RNAP assisting the recruitment of further RNAP via the wake of supercoiling left behind it. In principle one could argue that perturbed supercoil states could relax quickly in a plasmid [25] like the one used by Golding et al., but it has been demonstrated that a promoter can induce huge changes in supercoiling of a plasmid [28].
Consider a promoter where open complex formation is a rate limiting step that is assisted by negative supercoiling. To model this, we assume that the negative supercoiling assists this step to the extent that it is no longer rate limiting. We parameterize this effect of supercoiling into a single number q, the probability that supercoiling left in the wake of a prior RNAP allows a subsequent RNAP to rapidly form an open complex before the supercoiling is relaxed (Figure 2A). This then creates two possible behaviors at the promoter. If the promoter is in the supercoiled state, open complex formation is enhanced to the point where it is not rate limiting, and transcription events occur at rate E and are exponentially distributed. If the promoter is not in the supercoiled state, then open complex formation is very much slower and now rate limiting; transcriptional events are still exponentially distributed but now with the much lower rate O. This creates the long periods of inactivity associated with off periods (Figure 3A) and holds when O≪E, and gives a distribution(3)(shown in Figure 3B).
The supercoiling need not persist for the full length of the on-period, or for the length of time between two initiations. In the scheme we present here, it is only required that the supercoiling persists long enough to allow an open complex to form rapidly. The final escape step is assumed to be neutral with respect to supercoiling and hence as soon as an open complex has formed at the promoter the supercoiling can be relaxed without interrupting the on-period. This assumption can be varied without changing the general behavior of the model.
If the supercoiling is relaxed before an open complex is formed, the promoter has switched to an off-period where initiation occurs at a much slower rate. The parameter q determines the size of the on-periods, as after each initiation there is a probability q that another open complex will be recruited and the on-period will continue, or a probability 1-q that an off-period will start. Therefore, the probability of getting a burst of 〈Δn〉 initiations is proportional to qDn−1. In this model a promoter is in the on-state when it is in the supercoiled state or when it has an open complex. Table 1 gives equations relating model parameters to the average 〈Δn〉, 〈ton〉, and 〈toff〉 (Derivations are given in Text S1).
This mechanism can reproduce the observations of Golding et al. [10] with the parameters tO = 37 [min], tE = 29 [min] and q = 0.545. We simulated the recruitment model using the Gillespie algorithm [14]. It gives the expected shape for the P(Dt) distribution (Figure 3B) and matches the distribution of Dn measured by Golding et al. (3C) and also the distributions of on and off-periods measured by Golding et al. (3D). In these plots the on-periods are defined as being the time intervals when there is rapid successive initiation (Figure 3A), following the procedure in Golding et al. [10]; the detailed definition is given in the Materials and Methods section.
Another possibility is that the off periods are due to the formation of long-lived non-productive initiation complexes at the promoter [29]–[31]. These non-productive complexes have been observed in vitro and may be arrested backtracked complexes or complexes that cannot exit the abortive initiation state into productive elongation. In both cases initiation can be made more efficient by the GreA/B RNAP-binding factors [29],[30]. The random formation of such ‘dead-end’ complexes could block the promoter for extended periods of time, causing productive transcription to be confined to those times when the promoter is free. For the promoter lPR the lifetime of these complexes was found to be in the order of 10–20 minutes under in-vitro conditions, thus dead-end complexes can last long enough to cause the observed off-periods [31].
For the analytical treatment of this model we call the probability that a promoter bound complex will undergo a productive initiation Q, and the probability that the promoter bound complex enters a moribund state is therefore 1- Q. We assume that removal of the moribund complexes is a Poisson process with a rate d, which gives 〈toff〉 = τdead/Q with tdead = 1/d, which allows for the fact that a single off-period can be caused by multiple subsequent moribund complexes (Table 1). Here we consider a promoter to be in the off-period if it is occupied by dead-end complexes; otherwise it is on. The derivations of on- and off-times are given in Text S1. The dead-end complex mechanism is also capable of causing the behavior observed by Golding et al. The data of Golding et al. are reproduced with Q = 0.545, tdead = 20 [min], and tO+tE = 2.9 [min]. Figure 3E shows the distribution P(Dt) with these parameters obtained by the simulation using the Gillespie algorithm [14]. It has been confirmed that the distributions of Dn, ton, and toff are reproduced as well as the recruitment model (data not shown).
The formation of dead-end complexes is favored by low temperatures at the lac UV5 promoter [32]. If this were also the case for the Plac/ara promoter, it could be part of the explanation for why the Plac/ara promoter is so weak in the conditions used by Golding et al. (22°C) when it is reported to be a strong promoter elsewhere [15]. However, the activity of the promoter observed by Golding et al. at 37°C is still rather low compared the previously reported estimate [15]. This could be associated with the fact that there is almost no activation of the promoter caused by AraC/arabinose under their experimental conditions (see Figure 1E in Golding et al.). Another possibility could be the presence of an unknown terminator, which would imply that the number of complete transcripts represents only a fraction of the transcription initiation events.
One of observations made by Golding et al. that was used as evidence for transcriptional bursting was that the Fano factor for the distribution of number of transcripts N, ν = 〈(N−〈N〉)2〉/〈N〉, was approximately 4 for the Plac/ara promoter at 37°C, rather than 1 predicted for Poisson transcription. The Fano factor is a measure of noise; higher values indicating a more noisy process. When the on-periods are much shorter than the off-periods, the Fano factor n is linked to the burst size Dn as ν≈〈Δn〉. If the on-time is sizable, on the other hand, 〈Δn〉 needs to be much larger to give the same n.
By considering a population of cells where transcripts are degraded with rate g, we can relate n to model parameters. Figure 4 shows how n varies with model parameters for each model while keeping 〈N〉 = 10 obtained by analytical calculations (The detailed calculations are in the Text S1.). In the recruitment case the Fano factor is larger for smaller a and larger q, i.e., when the open complex formation is the rate limiting step and once a firing has occurred further recruitment occurs successively. In the dead-end model the Fano factor is larger for smaller b = (tO+tE)/tdead and larger Q, which occurs when moribund persist for long periods of time, but transcription during the on periods is rapid and occurs many times before another off period occurs. One should note that the Fano factor can be changed depending on parameters for a given 〈N〉; This means that the noise can be tuned for a given promoter strength under either model, which can allow the promoter noise to evolve to reflect a level that provides the best fitness for the cell.
We have analyzed possible mechanisms of transcriptional bursting in terms of a simple recruitment/isomerisation/escape model. A model where supercoiling created by an RNAP engaged in transcription assists in the recruitment of subsequent RNAPs is able to reproduce all the features of the experiments, without resorting to very large timescales for on-off equilibrium rates, or unknown pause sites or localization effects. Alternatively, the data of Golding et al. could also be reproduced if the investigated promoters spent a sizable fraction of their time by being occupied by an RNAP in a non-productive state.
Transcription bursts have been reported in eukaryotic systems [33],[34] and have also been proposed to facilitate cell to cell variability. These eukaryotic model systems both included transcription factors and in addition they may be influenced by chromatin remodeling. The bunched expression of nearby genes is correlated [34], a feature that fits with extended states of chromatin. The dead end complex cannot give such spatial correlations, whereas supercoiling mediated recruitment in principle could correlate expression from two promoters if they are close to each other.
In one mammalian system, the reported pulse duration and silenced periods are similar to the ones modeled in this paper [33]. However, in that system subsequent bursts of transcription are correlated, with one transcription burst priming the system for another one [33], which has not been reported in Golding et al. This is again consistent with the larger scale genomic silencing associated with, for example, chromatin states or the genes repositioning relative to transcriptional factories [22]. The recruitment model cannot account for correlations between subsequent bursts, whereas the dead end model could give such time correlations between busts if the dead end complexes come in different categories, each with their characteristic lifetime.
Overall we stress that our current modeling demonstrates two plausible mechanisms for generating bursts of transcription at an isolated promoter. Additional mechanisms come into play when the promoter is regulated by a transcription factor with a low on-rate, or when large scale reorganization of the chromosome takes place on a slow timescale.
Both the dead-end and the recruitment model can be simulated on-line using the java applet on http://www.cmol.nbi.dk/models/transcription/RNAPInitiation.html.
The recruitment model implies a number of predictions that can be tested. In particular, promoters with bunched transcription initiation will be highly sensitive to negative supercoiling of the DNA. And conversely, promoters that are insensitive to supercoiling will have transcription events which are separated by more regular time intervals.
For promoters that are sensitive to supercoiling, one could selectively shorten the long off periods by introducing a second nearby promoter. One option is to add a divergent promoter that might be able to donate its negative supercoil wake. Such a construct was investigated by Opel et al. [27], who reported that a second promoter could indeed increase the activity of a supercoiling sensitive promoter in the ilvYC operon. This predicts that if a similar experiment was done with the Plac/ara promoter, then reduced off periods would be observed.
Another prediction is that for promoters with bunched activity the isomerisation step is rate limiting. Thus the fraction of time spent in open complex is small compared to the time between transcription initiations. One might be able to show an inverse correlation between the noisiness of a promoter and the occupancy of the promoter by open complexes using potassium permanganate DNA footprinting [35].
The dead-end mechanism implies that the promoter is mostly occluded by an RNAP with an open transcription bubble. This could be identified permanganate footprinting [35].
The availability of GreA/B could affect the rate of removal of the dead-end complex, d [29],[30]. Overexpression of GreA/B could increase d and reduce off-periods, while longer off-periods, due to lower d, could be observed in greA/B mutants.
It is possible that the dead-end complexes could be removed by a collision with an RNAP transcribing from a second promoter in a fashion similar to the removal of an open complex by transcriptional interference [36]. The off-times of a promoter could therefore in principle be shortened by using other RNAP's initiated from another promoter that transcribes across the promoter in question. If a promoter spent a substantial fraction of the time occupied by a dead-end complex, it could be strongly activated by tandem or even convergent promoters, which would be a novel twist on the usually repressive effect of transcriptional interference. If d is reduced in Table 1, the “off-times” could be reduced by a factor set by the ratio of the strength of the two promoters, and the promoter activity could increase. Thus, if Plac/ara activity is affected by dead-end complex formation, then placing a weak divergent promoter upstream should not increase Plac/ara activity but placing this promoter in a convergent orientation may activate Plac/ara.
The sensitivity of a promoter to supercoiling mediated recruitment or dead-end complex formation provides additional avenues for control of overall promoter strength, either by evolution or by regulatory factors.
DNA supercoiling can increase or decrease promoter activity both in vitro [26] and in vivo [37] in a promoter specific manner. Supercoiling can affect RNAP binding to the promoter and open complex formation in vitro and presumably can affect other steps as well. RNAP recruitment induced by the supercoiling created by an elongating transcription complex may contribute significantly to the activity of certain promoters. We expect that, except for very active promoters, rapid dissipation of the supercoil wake would make inhibition of a supercoiling-repressed promoter by this mechanism unlikely. Stimulation by the departing elongating complex should similarly only apply to the early steps in initiation. Thus only promoters whose early steps are rate-limiting and can be enhanced by supercoiling should be stimulated by this mechanism.
The reduction of promoter activity by the formation of dead-end complexes is potentially very strong. The effect increases with the probability of forming such a complex (1-Q) and with the lifetime of the complex (1/d), parameters which could be determined both by the promoter sequence and by the availability of factors such as GreA/B that may remove the complex [29],[30]. This mechanism would seem to be an inefficient way to set the strength of a promoter, as it would sequester an RNAP. However, it would allow regulation by transcription factors that change the fraction of RNAPs that enter into dead-end complexes or that stabilized the dead-end complex. As a consequence, genes which are silenced through this mechanism will have relatively high fluctuations in expression level, and thereby some cells can explore advantages afforded by relatively high expressions, even when most cells are kept at near zero expression. Bunched activity for a near silenced promoter could, for example, be important in the pathway for the spontaneous induction of lysogeny for some temperate phages, like P2.
High noise in protein levels can also be obtained at the translation level. If a single mRNA molecule is rapidly translated many times the result is a burst of protein production. Therefore transcriptional bursting is not strictly required for protein production to occur in bursts. However, transcriptional bursting might allow for additional modes of regulation by transcription factors or other proteins that influence the state of the DNA around the promoter site. It may also complement bursts of protein production produced by rapid translation by removing constraints placed on burst size by the upper limits of mRNA translation rate.
Dynamics and the interplay between timescales presents an open, and until recently, quite unexplored part of molecular biology. The present analysis suggests a new mechanism for in vivo regulation, where long silent timescales emerge as the result of some particularly large rate limiting step in the promoter. These steps are open for new levels of regulation by transcription factors, which naturally will be most effective when they influence the rate limiting step of transcription initiation [38].
To calculate the activity of a promoter we first calculate the probability that the promoter will be occupied by closed (h) and open (q) complexes using steady state conditions. The total activity of the promoter is given by F = Eq for the standard model and the recruitment model, and F = QEq for the dead-end model. Details of the calculation are found in the Text S1.
The time between subsequent initiations is calculated by considering the time needed for each step as described in the Text S1. For class I there is only one step and the distribution is a simple exponential. For class II there is two steps. If these steps take an average time of to and tE , the total waiting time between events is distributed with(4)giving eq. (1) in the main text for tO≠tE. For one t much greater than the other, this distribution degenerates into a simple exponential. For tO = tE, eq. (4) gives eq. (2) in the main text.
For the recruitment model, the intervals between initiations are partitioned between the supercoiling assisted or unassisted outcomes, with a partitioning ratio given by q. Details are in the Text S1. For the dead-end model the distribution is similarly partitioned between the two distributions with a partition ratio given by Q. Details are in the Text S1. In the Text S1 we also show how to calculate the distribution of “on” and “off” times from q or Q. Finally, we calculate the Fano factor ν = 〈(N−〈N〉)2〉/〈N〉 by using generating functions as described in the Text S1.
We distinguish “on-periods” and “off-periods” in the simulation data following the procedure used by Golding et al. [10]. They analyzed the experimentally obtained time series of fluorescent signal manually. The system is considered to be in “off-period” when the signal does not change for a while, and otherwise it is in “on-period”. The specific time resolution to detect an “off-period” was not given, but the shortest off-time measured was around 6 [min] (Golding, private communication); in other words, transcription events separated by less than 6 [min] were considered to be in the same “on-period”.
During an on-period, the number of messages transcribed, Dn≥1, and the duration ton were recorded; the time to transcribe one message D was 2.5 [min] [10], which corresponds to the on-time for Dn = 1 case.
Considering this protocol used by Golding et al. [10], we defined Δn, ton, and the duration of the off-time toff out of the time series of firings from our model (Figure 3A) as follows: (i) When firings are separated by more than τc = 6 [min]+Δ = 8.5[min], the promoter is in an off period. (iii) Otherwise, if successive firings are separated by an interval less than τc, the gene is considered to be on until we observe an interval greater than τc. This defines the on-time ton, and we count the number of transcripts per on-time Δn. |
10.1371/journal.pntd.0005268 | Soluble Egg Antigens of Schistosoma japonicum Induce Senescence of Activated Hepatic Stellate Cells by Activation of the FoxO3a/SKP2/P27 Pathway | Liver fibrosis was viewed as a reversible process. The activation of hepatic stellate cells (HSCs) is a key event in the process of liver fibrosis. The induction of senescence of HSCs would accelerate the clearance of the activated HSCs. Previously, we demonstrated that soluble egg antigens (SEA) of Schistosoma japonicum promoted the senescence of HSCs via STAT3/P53/P21 pathway. In this paper, our study was aimed to explore whether there are other signaling pathways in the process of SEA-induced HSCs aging and the underlying effect of SKP2/P27 signal on senescent HSCs.
Human hepatic stellate cell line, LX-2 cells, were cultured and stimulated with SEA. Western blot and cellular immunofluorescence analysis were performed to determine the expression of senescence-associated protein, such as P27, SKP2 and FoxO3a. Besides, RNA interfering was applied to knockdown the expression of related protein. The senescence of HSCs was determined by senescence-associated β-gal staining. We found that SEA increased the expression of P27 protein, whereas it inhibited the expression of SKP2 and FoxO3a. Knockdown of P27 as well as overexpression of SKP2 both suppressed the SEA-induced senescence of HSCs. In addition, the nuclear translocation of FoxO3a from the nucleus to the cytoplasm was induced by SEA stimulation.
The present study demonstrates that SEA promotes HSCs senescence through the FoxO3a/SKP2/P27 pathway.
| Activation of hepatic stellate cells (HSCs) is a key event of liver fibrosis. Induction of activated HSCs apoptosis and inhibition of activated HSCs proliferation are the common anti-fibrotic strategies to block liver fibrosis. The induction of senescence of HSCs is responsible for the clearance of the activated HSCs as well. Senescence of HSCs is mediated by exposure to soluble egg antigens (SEA) of Schistosoma japonicum via STAT3/P53/P21 pathway. In this study, we found that SEA induced the senescence of HSCs, accompanied with the increased the expression of P27 protein and the decreased expression of SKP2 and FoxO3a. Either knockdown of P27 or overexpression of SKP2 alleviates the SEA-induced senescence of HSCs. Moreover, SEA droved the nuclear translocation of FoxO3a from the nucleus to the cytoplasm. Hence, the present study demonstrates that SEA promotes HSCs senescence through the FoxO3a/SKP2/P27 pathway.
| Liver fibrosis, a major health problem worldwide [1], results from different etiologies of chronic liver injury, and eventually progresses into cirrhosis or hepatocellular carcinoma. Recently, liver fibrosis was viewed as a reversible process [2]. After years of prevention and treatment of schistosomiasis in China, the new cases of Schistosoma infection have declined significantly, but there are still thousands of patients suffering from schistosomiasis [3]. The main pathological change of schistosomiasis is the formation of granuloma around the eggs of Schistosoma japonicum (S. japonicum) in the liver, leading to liver fibrosis.
Studies indicate that the activation of hepatic stellate cells (HSCs) is a key event in the process of liver fibrosis. HSCs are activated and then transform to myofibroblasts, once the liver is subjected to stimulations. Activated HSCs synthesize large amounts of extracellular matrix proteins (ECM) such as type I or type III collagen, laminin and fibronectin [4]. In the process of liver fibrosis induced by S. japonicum infection, HSCs gather around S. japonicum egg granuloma [5]. Activated HSCs can express a variety of inhibitors of metalloproteinases (TIMPs) to prevent the degradation of matrix proteins, resulting in the replacement of normal liver tissue by collagen matrix and the formation of fibrous scar. Therefore, inhibition of the HSCs activation, proliferation and accelerating the clearance of the activated HSCs are key strategies for the prevention and treatment of liver fibrosis [6].
Substantial evidences support the possibility of the reversibility of liver fibrosis [2]. Recently, studies revealed that with the development of pathologic process, the size of egg granulomas at the chronic phase (12 weeks) and the advanced phase (24 weeks) was smaller than that at the acute phase of S. japonicum egg-induced liver fibrosis [7]. Researches indicate that the reversion of liver fibrosis is closely related to the increase of the apoptosis of HSCs. Expression of the tissue inhibitor of metalloproteinase-1 (TIMP-1) decreased, and the synthesis of metalloproteinases (MMPs) such as MMP-1 and MMP-13 increased, thereby inhibiting HSCs activation and proliferation, increasing the clearance of activated HSCs as well as the degradation of collagen fiber, and eventually alleviating liver fibrosis [8,9].
Studies showed that the induction of senescence of HSCs would accelerate the clearance of the activated HSCs as well [10]. The senescent cells usually display a cell cycle arrest in the G0 or G1 phase but maintain the metabolic activity [11]. Once senescent, senescence-associated β-gal (SA-β-Gal), the specific maker of senescence, is detected in these cells. In our previous study, we demonstrated that SEA induced the HSCs senescence through the STAT3/P53/P21 pathway [12]. Besides, it has been well established that FoxO3a signaling cascade is implicated in the senescent process of multiple cells [13–15]. It has been revealed that FoxO3a inhibited the senescence of hepatocytes [16]. Additionally, S phase kinase associated protein 2 (SKP2) was reported to suppress cellular senescence induced by oncogenic stimuli independent of ARF/p53 signaling. And cell cycle inhibitor P27, the SKP2 substrate, is targeted by SKP2 for ubiquitination and degradation [17,18]. In this study, we investigate whether the FoxO3a/SKP2/P27 signaling participates in the SEA-induced HSCs senescence.
SEA of S. japonicum were obtained from Jiangsu Institute of Parasitic Diseases (China). SEA was sterile-filtered and endotoxin was removed with Polymyxin B agarose beads (Sigma, USA). Limulus amebocyte lysate assay kit (Lonza, Switzerland) was used to confirm the removal of endotoxins from the SEA as previously described [19]. Primary antibodies for FoxO3a, SKP2, P27 and AKT were purchased from Santa Cruz Biotechnology (USA, antibody dilution for Western blot of all antibodies from this company is 1:200). Primary antibody for phospho-AKT was purchased from Cell Signaling Technology (USA, antibody dilution for Western blot is 1:1000). All of the secondary antibodies were obtained from Santa Cruz Biotechnology (USA, antibody dilution is 1:2000). The staining kit for SA-β-Gal was purchased from GenMed Scientifics Inc (USA).
LX-2 cells, the ‘immortalised’ human HSCs, were provided by Xiangya Central Experiment Laboratory (Hunan, China) and maintained in DMEM with 10% Fetal Bovine Serum in a humidified incubator with 5% CO2. Culture medium was replaced every day and cells were subcultured with trypsin when they were at 80% confluence.
Cells were lysed in RIPA cell lysis buffer including protease inhibitor (1mM) and phosphatase inhibitors (1mM). Equal amounts of protein extract were separated by SDS-PAGE and then transferred onto polyvinylidene difluoride (PVDF) membranes. The membranes were blocked in 5% nonfat milk for 2 hours, incubated with the indicated primary antibodies at 4°C overnight, and then incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies for 1 hour at room temperature.
SA-β-Gal staining was performed according to the instruction of SA-β-Gal staining kit, in which cleaning solution, fixation fluid, acidic solution and staining fluid were provided as the main kit contents. Briefly, LX-2 cells were washed with cleaning solution and fixed by fixation fluid for 5 minutes at room temperature. Afterwards, cells were washed by acidic solution twice and stained with staining fluid for 16 hours at 37°C. Finally, SA-β-Gal staining positive cells were assayed using a bright field microscope.
For the Immunofluorescence staining, cells were seeded in 6-well culture plates and fixed with 4% paraformaldehyde. Afterwards, cells were permeabilized with 0.1% Triton X-100 and then blocked in 5% BSA. After that, cells were incubated with FoxO3a antibody (dilution is 1:50) and visualized with Alexa Fluor 568 conjugated secondary antibody (Invitrogen, USA, antibody dilution is 1:200) under a fluorescent microscopy.
pcDNA3.1 plasmid was digested with EcoRI and BamHI (TaKaRa, China), and CDS region of SKP2 (GenBank: NM_005983) was subcloned into pcDNA3.1 vector to generate the recombinant vector pcDNA3.1-SKP2. The recombinant plasmids were verified by restriction analysis and sequencing.
LX-2 cells were transfected with P27 siRNA (GenePharma, China) or pcDNA3.1-SKP2 overexpression plasmid by Lipofectamine 2000 reagents (Invitrogen, USA) according to the manufacturer’s instructions. After 24 hours, cells were subjected to various stimulations for indicated time.
Data is expressed as mean ± SEM (standard error of mean) of three independent experiments. All p values were calculated using a two tailed paired Student’s t test or a one way ANOVA. p < 0.05 was considered as statistically significant.
Previously, we found that SEA-induced LX-2 cells senescence via the STAT3/P53/P21 pathway [11]. Since P27, the cell cycle inhibitor, plays an important role in cellular senescence and SKP2 could cause a decrease in the level of P27 expression [13, 20], we next verified whether P27 signaling pathway is implicated in the progress of LX-2 senescence. As illustrated in Fig 1, Western blot analysis showed that SEA markedly increased the expression of P27, but decreased the SKP2 protein level. Furthermore, the expression of P-AKT, the upstream of P27, was also significantly decreased under SEA exposure, although the total expression of AKT was not affected.
Apart from the regulation of the level of P27 by SKP2, P27 is also regulated by the FoxO3a protein at the transcriptional level [14]. Also, FoxO3a could be regulated by AKT and 14-3-3 protein [21]. Thus, we further investigated whether FoxO3a was involved in the senescence of LX-2 cells induced by SEA. The results of Western blot indicated that FoxO3a was significantly inhibited by SEA stimulation in the LX-2 cells (Fig 2A). Besides, cell immunofluorescence assay confirmed that FoxO3a was transferred from the nucleus to the cytoplasm after SEA treatment (Fig 2B and S1 Fig). These results suggested that FoxO3a was implicated in the SEA-induced senescence in LX-2 cells.
In order to further verify the role of P27 in the SEA-induced senescence in LX-2 cells, P27 specific small interfering RNA was used to knockdown the expression level of P27 protein in LX-2 cells. As illustrated in Fig 3, the SA-β-Gal staining showed that the senescent LX-2 cells significantly increased accompanied with the upregulated P27 upon SEA stimulation. Nevertheless, the senescence of LX-2 cells induced by SEA was reversed by the P27 siRNA. These results suggest that P27 is a key regulator in the senescence of LX-2 cells induced by SEA.
Studies indicate that SKP2 plays an important role in the process of cellular senescence [17,20,22], thus, we explored whether SKP2 is a regulator in the SEA-induced senescence in LX-2 cells. We found that SEA inhibited the expression of SKP2 (Fig 1). In order to further investigate the potential mechanism of SKP2 in the process of SEA-induced senescence, specific SKP2 over expression plasmid was constructed and transfected into LX-2 cells, and then the efficiency was confirmed by Western blot analysis. The results showed that the SKP2 protein expression in LX-2 cells was enhanced after transfection with SKP2 over expression plasmid (Fig 4A), and the high expression of SKP2 could inhibit the senescence of LX-2 cells induced by SEA (Fig 4B). These results suggest that SKP2 can inhibit LX-2 cells senescence mediated by SEA.
To explore the mechanism of SKP2 on senescence in LX-2 cells, we also examined the expression of P27, and we found that the expression of P27 in LX-2 cells was significantly restricted after the overexpression of SKP2 (Fig 4A). On the contrary, the expression of SKP2 in LX-2 cells was not affected by the knockdown of P27 expression (Fig 3A).
It has been well accepted that activation of quiescent HSCs is responsible for the excessive production of ECM in liver fibrosis [23,24], and there has been increased recognition in utilizing functions of HSCs for therapeutic applications to reverse liver fibrosis [25]. Thus, preventing the activation of HSCs and increasing the clearance of activated HSCs are viewed as promising anti-fibrotic strategies [4,26,27]. Among these, induction of activated HSCs apoptosis and inhibition of activated HSCs proliferation are the common anti-fibrotic strategies to block liver fibrosis. For example, we found SEA could induce HSC apoptosis and inhibit activation of HSCs under some suitable conditions [19]. In addition, studies showed that the senescence of HSCs would block the development of liver fibrosis. Kong X et al. have demonstrated that IL-22 induced HSCs senescence and restricted the development of liver fibrosis in mice [10,28,29]. Which are different from quiescent HSCs, senescent HSCs often manifest as SA-β-Gal staining positive cells. In the previous study, our results showed that more SA-β-Gal staining positive cells could be found in SEA-treated LX-2 cells and SEA decreased the expression of α-SMA in LX-2 cells partially due to SEA-induced senescence [11].
It has been shown that P53, tumor suppressor protein, plays a critical role in the induction of senescence. We have recently shown that SEA induced HSCs senescence through STAT3/P53/P21 pathway. SEA increased the expression of P-STAT3, P53 and P21. And knockdown of STAT3 or P53 inhibited the SEA-induced senescence of HSCs [12]. Besides the P53-P21 and P16-Rb signaling pathways [30–32], there are other signaling pathways that promoting the development and progression of cellular senescence. The inactivation of retinal vascular tumor suppressor factor (VHL) can decrease the expression of SKP2 and increase the expression of P27, and then induce cellular senescence [33]. Consistent with this result, the overexpression of HTLV-1 Tax protein also reduced the expression of SKP2 and accompanied with the occurrence of cellular senescence in human T cells [34]. These results suggest that the decrease of SKP2 and the induction of P27 might play direct roles in cellular senescence. SKP2 is a member of the F box protein family, and the formation of the SKP2-SCF complex exhibits the E3 ligase activity. Li Z et al. showed that SKP2 regulates cell cycle and cell proliferation by degradation of its downstream molecules such as P27, a cell cycle inhibitor [35–37]. And recent studies have shown that inactivation of SKP2 induces cellular senescence, in which the cell cycle inhibitor P27 and P21 expression are enhanced [17]. Therefore, we suspect that SKP2 is involved in the process of SEA-induced LX-2 cell senescence. We found that SEA markedly inhibited the expression of SKP2, but enhanced expression of P27 (Fig 1). In order to further verify the role of SKP2 and P27 in the senescence of LX-2 cells, we transfected P27 siRNA to LX-2 cells to knockdown the P27 protein expression and transfected SKP2 overexpression plasmid to upregulate the expression of SKP2. These results further confirmed that SEA-induced cellular senescence was partially dependent of SKP2/P27 pathway (Fig 3 and Fig 4).
In addition to the regulation of the post transcriptional level of P27 by SKP2, P27 is also regulated by the FoxO3a protein at the transcriptional level [14]. The data shows that FoxO3a participates in the process of many kinds of cell senescence. Xu-Feng et al. found that FoxO3a can inhibit the senescence of cardiovascular endothelial cells by regulating the cell cycle mediated by ROS [14]. Similarly, in the experiment of Kyoung Kim H et al., FoxO3a also exhibited an inhibition effect on human dermal fibroblast senescence. The experimental results demonstrated that knockdown of FoxO3a could promote the cell senescence [38]. Therefore, we further verify the effect of FoxO3a on the senescence of SEA-induced LX-2 cells, and our experimental results are consistent with the above phenomena. In SEA-treated LX-2 cells, FoxO3a protein expression was significantly inhibited, and FoxO3a occurred nuclear transfer from the nucleus to the cytoplasm under the role of SEA (Fig 2B). Thus, FoxO3a is a key regulator in the SEA-induced senescence of LX-2 cells.
To our knowledge, AKT kinases are critical players in PI3K-mediated signal transduction pathways [39]. AKT phosphorylates downstream substrates to regulate cell growth, proliferation, apoptosis, senescence, and other processes [40]. Cong Fu et al. found that P-AKT expression was down-regulated during the process of cellular senescence induced by H2O2 [41]. Studies demonstrated that AKT phosphorylated FoxO proteins, leading to the negative FoxO regulation via triggering its nuclear exclusion [21]. In addition, AKT can also promote the degradation of P27 [42]. In the present study, our results showed that the expression of P-AKT was inhibited by the SEA stimulation (Fig 1).
In conclusion, SEA might slow down the progression of liver fibrosis by promoting HSCs senescence through the FoxO3a/SKP2/P27 pathway. Our previous and present findings provide evidence supporting a possible mechanism by which SEA induces senescence in LX-2 cells (Fig 5) and these provide a potential target of the clinical research of liver fibrosis.
|
10.1371/journal.pntd.0005004 | Can You Judge a Disease Host by the Company It Keeps? Predicting Disease Hosts and Their Relative Importance: A Case Study for Leishmaniasis | Zoonoses are an important class of infectious diseases. An important element determining the impact of a zoonosis on domestic animal and human health is host range. Although for particular zoonoses some host species have been identified, until recently there have been no methods to predict those species most likely to be hosts or their relative importance. Complex inference networks infer potential biotic interactions between species using their degree of geographic co-occurrence, and have been posited as a potential tool for predicting disease hosts. Here we present the results of an interdisciplinary, empirical study to validate a model based on such networks for predicting hosts of Leishmania (L.) mexicana in Mexico. Using systematic sampling to validate the model predictions we identified 22 new species of host (34% of all species collected) with the probability to be a host strongly dependent on the probability of co-occurrence of vector and host. The results confirm that Leishmania (L.) mexicana is a generalist parasite but with a much wider host range than was previously thought. These results substantially change the geographic risk profile for Leishmaniasis and provide insights for the design of more efficient surveillance measures and a better understanding of potential dispersal scenarios.
| Emerging and neglected zoonoses are an important global threat to public health. Host range, in particular, is a crucial factor in determining disease risk and the potential for adequate interventions. Here we show that Leishmania has a very wide host range and that Complex Inference Networks can be used to infer ecological relationships in the context of zoonoses, identifying both the potential hosts and their relative importance. These results substantially change the risk profile and potential control measures that can be used to combat the disease, allowing for the design of more efficient surveillance measures and a better understanding of potential dispersal scenarios.
| Zoonoses are an important class of neglected [1,2] or emerging infectious diseases [3–6], accounting for more than 60% of human infectious diseases. Wildlife species that are hosts for pathogens play a fundamental role in zoonoses, threatening domestic animal and human health and global biodiversity. Although for particular zoonoses some hosts have been identified [7–14], there have been few systematic empirical studies carried out to identify the host range and the relative importance of the different hosts within that range for a given zoonosis.
Additionally, most work on disease ecology over the last 20 years has focused on single-host, single-agent systems. Recently however, there has been increasing interest in the more complex case of multi-host systems [15–20], with the realization that many zoonoses have potentially ample host ranges.
The relative importance of a species as a disease host will be highly multi-factorial, with risk factors covering many different scales, from the micro to the macro. However, there are two particularly important elements that come into play: host competence (the ability to transmit the parasite to a new host or vector) and the frequency of contact between host and pathogen or, in the case or vector borne diseases, host and vector [21].
Although it is intuitively clear that the “relative importance” of a host will depend on both its competence and its frequency of contact with the vector, it is a somewhat ambiguous concept in that it depends, for instance, on whether we are talking about human transmission or of maintaining an enzootic transmission cycle. We will take here relative importance to be associated with the probability of infection of an individual of a potential host species. Abstractly, this is a highly multi-factorial function P(C | X1, X2,…, XN), where, for instance, X1 could represent host competence and X2 frequency of host-vector contact. Although a host may be highly competent, if it only has infrequent contact with any disease vector, then the frequency of infected individuals will be low. Conversely, a host and a vector may have frequent contact, but the host may have low competence. All else being equal, the most important hosts will be those that have frequent contact with the disease vectors and are competent. Unfortunately, gathering information about these two aspects, especially for emerging or neglected diseases, is difficult and resource intensive [7,8,22,23]. Furthermore, how these fundamental aspects interact in multi-host systems is quite distinct from their single-host analogues. For instance, the fact that host-pathogen competencies may differ greatly among the hosts can potentially lead to a dilution effect [24–26].
Another important differentiating factor is that, multiple-host systems provide for much richer and complex scenarios for the dispersion of a disease from one geographic region to another [27]. As the characteristics of the host range play a crucial role in the emergence risk of a novel human pathogen and of the optimal interventions for combating the zoonosis [16] the importance of predicting and identifying potential disease hosts has been widely recognized [28–30]. To do so by exhaustive, systematic search through all possible hosts would be prohibitively resource intensive. At the same time, good data often only exist for a few (presumed) focal species. As it is unknown what part of the host range has already been discovered, the undiscovered part constitutes a type of ‘epidemiological dark matter’ [31].
An early attempt at systematising the search for potential hosts [32], in the case of Ebola, considered a heuristic approach based on expert knowledge, which was used to then filter the list of potential candidates. As such, it is both subjective and subject to model bias. More recently, other methods have appeared: In [33] a small group of four suspected hosts was used as a starting point for including biotic effects indirectly by calculating the fundamental niche of these four mammal species and considering the geographical correspondence with the niche distributions of the vectors. This paper was more concerned with including information about a particular set of potential hosts into corresponding risk maps rather than identifying new hosts per se. In contrast, in [34], a classification model using a supervised learning technique was used to predict other potential rodent reservoirs based on the predictive value of a set of potentially distinguishing characteristics of already known ones. Note that this paper was concerned with the potential hosts of a large number of pathogens considered all together and therefore could not discriminate against potential hosts of one disease versus another. In this case only categorical and no spatial information was used. Moreover, as it is based on supervised learning it can be affected by bias in the data defining both the class and in the predictors. This is in evidence in that the most predictive factor found was the number of literature citations for a given species. In [35], the authors considered biotic factors as potential predictive variables for describing the geographic range of Ebola rather than trying to predict which mammals are the most likely hosts.
In contrast, in Stephens et al [36], a general framework was presented using Complex Inference Networks based on the degree of co-occurrence between different species, for inferring potential biotic interactions. The framework is also capable of including in other variable types, at distinct spatial resolutions, such as environmental layers normally associated with abiotic variables [37] allowing for a comparison of the relative importance of biotic versus abiotic factors.
The methodology differs from those of [32–35] by using as model inputs only purely spatial data, using point collection data to proxy spatial distributions of taxa and co-occurrences to infer potential biotic interactions. In particular, it uses no auxiliary information, such as expert knowledge, as in the case of [32]; fundamental niche distributions of taxa, as in the case of [33]; or specific categorical data associated with the relevant taxa, as in the case of [34].
Networks are an important tool in ecological studies [38–41]. However, their local structure—in the sense of two nodes and a link as the base element—represents an already known relation, such as in a food web [38], or in a contact network representing ticks, vertebrates and pathogens, as in [42]. In this case the local structure of the network, i.e., the individual nodes and links, only represent what is known. However, the global properties of the network can lead to new insights from an eco-systemic or community viewpoint and also to specific predictions. In contrast, Stephens et al. [36] use the local structure of networks to infer and discover previously unknown relations, such as the relation between vector and host. Although nodes are taxa, the local structure of the network is different to a traditional ecological network in that the links represent the degree of overlap between the distributions of the corresponding taxa with the idea that statistically significant degrees of co-occurrence can be an indicator of potential biotic interaction between the associated taxa, such as between a host and a vector.
In determining the host range of a zoonosis, an exhaustive empirical analysis of all potential hosts is prohibitively difficult, hence the importance of theoretical models, such as that presented in [36], for guiding observation and experiment. Although consistent with known results, it is important to note that the theoretical predictions of [36] have not previously been tested experimentally. A theoretical model needs to be validated by experiment, as this is the only way to truly determine if the model works. In this vein, most ecological modelling of zoonoses remains untested, in that theoretical predictions are not validated using a suitable experimental validation framework. This paper presents the results of an interdisciplinary study carried out to experimentally test the predictions of [36] using, as a test, the case of Leishmaniasis in Mexico.
Leishmaniasis is a significant, yet neglected tropical disease, with 350 million people in 98 countries worldwide living at risk of developing one of the many forms of the disease [43]. It is caused by infection with one of several different species of protozoan parasites of the genus Leishmania, which maintain their life cycle through transmission between an insect (sandflies—genus Lutzomyia) and a mammalian host. In Mexico, the most epidemiologically important species is Leishmania (L.) mexicana, though the presence of other species has been confirmed. Eleven species of Lutzomyia are considered to have potential medical importance. Of these, three are known vectors of either cutaneous or visceral Leishmaniasis, while four others have been found infected with L. (L.) mexicana [44].
Besides the clinical and social importance of Leishmaniasis [45] and the acknowledgment of its zoonotic nature [46], the identification of wildlife hosts for these parasites is sparse and non-systematic. Prior to the present study, only 8 mammalian species had been identified as hosts in Mexico [47–49]. This potential lack of knowledge of parasite hosts greatly increases the difficulty of formulating theoretical approaches to explaining and predicting disease spread or for planning better and sustainable control measures [50].
The collection of specimens was performed according to the guidelines of the American Society for the Use of Mammalogists of Wild Mammals in Research and under a collecting permit is- sued by the General Direction of Wildlife of Mexico (permission number SGPA/DGVS/04631/ 14). The infections in mice were carried out following the National Ethical Guidelines for laboratory animals NOM-062-ZOO-1999. The project was approved by the Institutional Ethics Committee of the Medical Faculty of the National Autonomous University of Mexico (UNAM) with the registration number FMED/CI/RGG/013/01/2008.
The general modelling methodology of [36], is based on the idea that biotic interactions can be inferred from the locations of taxa as a function of space and time. Although biotic and ecological interactions in general are very complex, it is reasonable to state that the spatio-temporal distributions of taxa, or other ecological variables, reflect all of the factors and their causal interactions that determine them. In [36], the degree of co-occurrence between taxa was taken as an observable measure with which potential interactions could be inferred. Although co-occurrence is not equal to biological interaction, a significantly non-random co-occurrence distribution is a necessary condition for a biotic interaction between taxa, and as such it can be used to formulate hypotheses that can be checked experimentally. However, it is clearly not a sufficient condition. In the spirit of niche modelling, a biotic variable that co-occurs with a target taxon can be understood as being a niche component in the same sense as any abiotic variable, such as temperature. In fact, one would generally expect a closer causal relation between biotic variables than with abiotic variables. For example, the distribution of prey species for a predator, such as a carnivore, should clearly influence the latter’s distribution more significantly than temperature or precipitation.
In the case of many zoonoses, the predominant interaction between vector and host is due to the former feeding on the latter. This obviously requires a coincidence in space and time. Species that offer blood meals can maintain the presence of vector populations independently of the capacity to harbour a given pathogen. In other words, the interaction between host and vector is a necessary but not sufficient condition for the transmission of the pathogen. The total number of encounters between vector and potential host depends on many factors, including the abundance of both species. However, a key factor is the geographical overlap between them, as the greater the overlap the greater the probability of an encounter. Thus, for two host species, identical in all respects except their relative geographical overlap with the vectors, the host species with the larger overlap will be epidemiologically more important. Thus, vector-mammal geographical overlap is a necessary but not sufficient condition for both a feeding interaction and a pathogen transmission interaction.
Of course, there may be geographical overlap between species due to other reasons than a direct biotic interaction. Even if species distributions were random there would be overlap. It is therefore necessary to measure overlap relative to a null hypothesis, such as that associated with random distributions. Additionally, it may occur that there is a non-random overlap due to the existence of one or more confounding factors; for example, an abiotic variable, such as temperature. This can only be quantified by controlling for the presence of such a factor. As it is obviously infeasible to control systematically for every potential factor, a logic must be presented for considering a particular candidate. In summary: although geographical overlap is not a sufficient condition for biological interaction, it is necessary, and as such can be used to construct models that can then be checked explicitly by experiment to see to what extent it is predictive.
The explicit example considered in [36] was the identification of potential hosts for Leishmaniasis by studying co-occurrences between the vector species and the potential host species. A Complex Inference Network summarising the co-occurrence distributions was deduced that showed the most important potential mammal hosts for each sandfly species. Although the full network contains a great deal of structure and information, in terms of experimental validation each network observable requires an experimental protocol to be able to measure it. In particular, to work at the species level for the vectors, and associate and confirm hosts for a given vector species, would require collecting sandflies and genotyping their blood meals, as well as collecting potential host species and confirming the presence of the pathogen. In the present experimental study, we restricted attention to only potential host species and tested them for the presence of the pathogen considering the vector at the genus level only.
The explicit model for predicting potential hosts was created using a database of point collections for one Class, Mammalia, and one genus, Lutzomyia, of the class Insecta. The mammal data set contains 37,297 unique point collections from geo-referenced localities for 419 terrestrial mammals occurring in Mexico—the full potential host range (GBIF; www.gbif.org, and CONABIO; www.conabio.gob.mx). For Lutzomyia, there were 270 collections points taken from published literature and from national collections: Instituto de Diagnóstico y Referencia Epidemiológica (InDRE, Mexico City), the Colección Entomológica Regional, Universidad Autónoma de Yucatán (UADY, Mérida) and the Laboratorio de Medicina Tropical at the Universidad Nacional Autónoma de México (UNAM, Mexico City).
First, we divide up a geographic region of interest into spatial cells, xα,–in the present case Mexico–here we used a uniform grid of 3,337 rectangular cells of size 25km x 25km. The choice of an appropriate cell size is known in geography as the “modifiable areal unit problem”. In terms of forming a spatial grid, there are at least two important considerations: The sizes of the statistical samples of the variables and their degree of correlation. Too fine a grid and there will be no co-occurrences, too rough and there will be little to no discrimination. It was checked explicitly in [36] that the relative ranking of mammals by the model was quite insensitive to the cell size over the range 5km to 100km. See also [51]. One then counts co-occurrences in each spatial cell between different taxa, or other variables. In the present case, the co-occurrences are between the presence of Lutzomyia, Bi, and the presence of each distinct mammal species, Ik.
We take the taxon distribution, Bi (Lutzomyia), and a subset of potential niche variables. We are interested in the probability P(Bi | I′) = NBiAND I′ /NI′, where NBiAND I′ is the number of spatial cells where there is a co-occurrence of the taxon Bi and the niche variables I′, which we take here to be biotic variables, and NI′ is the number of cells where the niche variables take their stated values. The niche profile I′(xα) associated with a spatial cell xα then determines the probability of the distribution variable, Bi(xα), in that cell, and one now has a predictive model. The problem of calculating P(Bi | I’) directly is that both NBi AND I′ and NI′ are likely to be zero when the number of taxa or niche variables considered simultaneously is large, as there will tend to be no co-occurrences of so many variables. This can be ameliorated by considering a reduced number of both class and feature variables. For instance, P(Bi | Ik) is determined by the number of co-occurrences of the taxon Bi and the particular niche variable Ik and, in principle, allows us to find the most important statistical associations between the niche variables and the taxa distributions. However, P(Bi | Ik) being a probability does not account for sample size. For example, if P(Bi | Ik) = 1, this may be as a result of there being a coincidence of Bi and Ik in one spatial cell or 1,000. Obviously, the latter is more statistically significant. To remedy this we consider the following test statistic
ε(Bi|Ik)=NIk(P(Bi|Ik)−P(Bi))(NIkP(Bi)(1−P(Bi)))1/2
(1)
a binomial test which measures the statistical dependence of Bi on Ik relative to the null hypothesis that the distribution of Bi is independent of Ik and randomly distributed over the grid, i.e., P(Bi) = NBi/N, where NBi is the number of grid cells with point collections of species Bi and N is the total number of cells in the grid. The sampling distribution of the null hypothesis is a binomial distribution where, in this case, every cell is given a probability P(Bi) of having a point collection of Bi. The numerator of eq (1) then, is the difference between the actual number of co-occurrences of Bi and Ik relative to the expected number if the distribution of point collections were obtained from a binomial with sampling probability P(Bi). As we are talking about a stochastic sampling the numerator must be measured in appropriate “units”. As the underlying null hypothesis is that of a binomial distribution, it is natural to measure the numerator in standard deviations of this distribution and that forms the denominator of eq (1). In general, the null hypothesis will always be associated with a binomial distribution as in each cell we are carrying out a Bernoulli trial (“coin flip”). However, the sampling probability can certainly change.
The quantitative values of ε(Bi |Ik) can be interpreted in the standard sense of hypothesis testing by considering the associated p-value as the probability that |ε(Bi |Ik) | is at least as large as the observed one and then comparing this p-value with a required significance level. In the case where NIk > 5–10 then a normal approximation for the binomial distribution should be adequate, in which case ε(Bi |Bk) = 1.96 would represent the standard 95% confidence interval. When a normal approximation is not accurate then other approximations to the cumulative probability distribution of the binomial must be used.
As ε increases monotonically with the frequency of co-occurrence, we interpret a statistically significant positive correlation as inferring a potential biotic interaction. Here, between sandflies and the corresponding mammal, which in this ecological setting one would naturally interpret as the mammal being a blood source for the sandfly, and therefore a potential host. The higher the value of (P(Bi | Ik)—P(Bi)) the greater the degree of spatial overlap between the species distributions and therefore the greater the risk posed by the corresponding mammal. Negative values of ε correspond to spatial overlaps that are less than one would expect from the null hypothesis.
The 419 mammal species were ranked according to ε. The resultant list serves as a predictive risk model, with the hypothesis that the highest ranked mammals correspond to the most important hosts, where, in the absence of other information, we assume that host competence is equal and importance is associated with the degree of spatial overlap between sandflies and mammal. All else being equal more overlap means more vector-host encounters. It should be noted that the method is not determining the physiological capacity of a mammal species to be a host but, rather, its potential epidemiological importance given that presence of mammal hosts is a necessary condition for the presence of the pathogen.
A corresponding biotic geographic risk model can be computed by calculating the probabilities P(Bi |I′), or proxies thereof, for each spatial cell. When I′ is of high dimension, this can be done using different classification models, such as neural networks, discriminant analysis, etc. A particularly transparent, simple and effective approximation is the Naive Bayes approximation:
P(Bi|I)=P(I|Bi)P(Bi)P(I)=∏k=1NP(Ik|Bi)P(Bi)P(I)
(2)
where, in the first equality, Bayes rule has been used, and in the second it has been assumed that the niche variables Ik are independent. The product here is over the N niche variables under consideration as conditioning factors for Bi. In the case of the relationship between Lutzomyias and mammals, N represents the number of mammal species. A score function that can be used as a proxy for P(Bi |I′) is
S(Bi|I′)=∑k=1NS(Bi|Ik)=∑k=1Nln(P(Ik|Bi)P(Ik|Bi__))
(3)
where Bi¯ is the complement of the set Bi. For example, if Bi is the set of cells with presence of taxon Bi then represents the set of cells without presence. S(Bi | I′) is a measure of the probability to find the distribution variable Bi when the niche profile is I′. It can be applied to a spatial cell xα by determining the niche profile of the cell, I′(xα). As an example, for two biotic niche variables, B2 and B3, that take values 1 (corresponding to the fact that there is a point collection associated with that cell) and 0 (there is no point collection associated with the cell), the four possible biotic niche profiles of any cell are (B2, B3) = (0,0), (0,1), (1,0) and (1,1). The score contributions of each biotic variable are S(Bi|B2) and S(Bi|B3), calculated using the above formula. Hence, S(Bi | I′) = S(Bi | B2, B3) = S(Bi|B2) + S(Bi|B3). Thus, for any given spatial cell xα one can assign a niche profile, i.e. values of B2 and B3, from whence it is possible to assign a corresponding score. If there is no statistical association between Bi and B2 or B3 then the corresponding score contributions are zero.
An overall zero score signifies that the probability to find Bi is the same as would be found if Bi were distributed randomly. If the score is positive then there is a higher than random probability to find Bi present and on the contrary if the score is negative. As each niche factor is treated separately in ε(Bi |Ik) or S(Bi|Ik) we can thus evaluate the relative contribution of any given niche factor and compare it to the contribution of any other. By determining the set of presence/no presence attributes in a spatial cell, eq (3) can be applied to each cell thereby determining the relative risk of that cell for the presence of Lutzomyias. As taking the ranked list as a predictive model involves several important assumptions; it is essential to test the model with experimental data. Obtaining the relevant data required sampling the spatial grid by collecting mammals and sandflies from different geographic points and then testing them for the presence of the parasite.
The sampling sites were selected as follows: a 25 x 25 km grid (as discussed above) was superimposed on a map of Mexico and this was used as template to determine the sampling sites. The sampling was stratified according to altitude so that only grid squares at < 2000 masl were used. Also, we excluded any grid square with >50% of water or urban cover. Sometimes the 25 x 25 km grid square selected was sub-divided in order to have more than one sampling site per square. Once the grid was established we selected 52 localities at random in 10 Mexican states and covering many different eco-regions and associated with a large selection of vegetation types. A random sample of spatial cells was necessary in order to fairly validate the model. If we had targeted the sampling to only those cells predicted to be highest risk then we would not be able to discriminate between high and low risk in the validation. Of course, once the model is validated it can be used with confidence in the future to preferentially identify those regions of highest risk where, for example, surveillance efforts should be targeted. The sampling was carried out by four field groups who, over a period of two years, collected 922 taxonomically identified specimens of 70 distinct species (for more details of animal sampling see S1 Text).
Tissue samples were taken to the Laboratorio de Inmunoparasitología of the Unidad de Investigación en Medicina Experimental of the National Autonomous University of Mexico, UNAM, where PCR tests were carried out to identify the presence of the pathogen Leishmania (L.) mexicana.
In Table 1, we show the list of collected species with the number of individuals that tested positive for the presence of the parasite Leishmania (L.) mexicana and the number that tested negative. Of the 70 mammal species collected, approximately 1/6 of all species present in Mexico, 24 (34%) had one or more samples that tested positive for the presence of the Leishmania (L.) mexicana parasite. Thirteen species of bats, and one of squirrel, were identified for the first time as Leishmania hosts in Mexico. Of the total number of collected individuals (N = 922), 62 tested positive, yielding an average infection rate across all species of 6.7%, although infection rates varied greatly, both temporally and spatially, exhibiting considerable heterogeneity, from 0% to 60%, across distinct collection sites and season of the year. In addition to the percentage that tested positive we also include the 95% confidence interval limits using the Wilson score interval [55,56] for that percentage relative to the null hypothesis that all mammal species had the same baseline infection rate of 6.7%. However, for those species where the number of positives is zero we also calculate the exact probability to obtain this result, (1-p)N, where p is the baseline infection rate and N is the number of negative collections.
Fig 1 shows the ranked values of ε, the statistical measure of degree of co-occurrence used to infer potential biotic interactions (see Eq (1) of the Methods section) for all mammal species as determined from the complex network exhibited in [36] (S1 Table) and considering Lutzomyia as a genus. The horizontal axis represents the null hypothesis that sandflies are distributed randomly with respect to mammal distributions, in other words P(Bi | Ik) = P(Bi) and hence ε = 0. To determine the extent to which collection biases can influence the overall distribution we randomly redistributed all collections over those spatial cells that had at least one collection. This has the effect of removing correlations between one species and another while at the same time preserving any bias associated with under sampling of certain geographical areas. This random re-assortment was repeated 50 times and average values for ε determined for each species.
To test the predictive power of the risk model, and to better visualize the relationship between ε and the percentage of species that tested positive, we group the list of mammal species ranked by ε into deciles [57] each decile corresponding to 10% of the list, and compute the average value of ε for each decile. The result can be seen in Fig 2. The relative proportion of positives, P, is a strongly increasing function of ε. Note that this regression is only to demonstrate the predictive power of the underlying classification model based on ε, i.e., the statistical significance of co-occurrence. Similarly, in Fig 3 we see the relative correlation between the average value of ε and the percentage of individuals identified as positive. Once again, the relative proportion of positives is a strongly increasing function of ε. Of course, P is a multi-factorial function, P(X1, X2,…, XN) that depends on many factors, such as host competencies. Essentially, here we are considering a regression model for P(X1, X2,…, XN) with respect to the variable X1 = ε and ignoring the rest as we do not have the relevant information to include them. Seen as a logistic regression at the species level, the associated relation is: Logit P = -1.415 + 0.186 * ε, with a p value of 0.03 on the regression coefficient. This confirms the statistically significant relation between ε as a statistical measure of geographical overlap and the probability to be a host of Leishmania (L.) Mexicana.
In Fig 4 we show a graph of prevalence at the species level versus ε for the 24 positive species, while in Fig 5, we compare the disease risk maps determined by using only the eight previously confirmed hosts of Leishmania versus one using the set resulting from our analysis, where by risk we mean probability of presence of the vector. A clear distinction can be seen between the areas of higher risk between the two models with the present model indicating a much higher degree of risk of presence of Lutzomyia in other than the southeast of Mexico.
Host range is an important factor in the dynamics of a zoonosis, both as a variable that affects the overall risk of presence of the disease as well as in terms of determining optimal interventions. An essential factor in determining the importance of a host is its co-distribution with the disease vector. In this paper, we reported the results of an extensive, interdisciplinary, empirical investigation carried out to test the predictions of a model to predict the relative importance of mammal hosts for the pathogen Leishmania (L.) mexicana associated with the emerging disease Leishmaniasis in Mexico. Once again, we emphasise that importance here is an “all else being equal” notion, i.e., that the greater the overlap between species the greater the probability of a vector-host interaction and therefore a greater number of infected individuals and a greater probability of transmission. Of course, many other factors—host competence, host abundance etc. will influence the epidemiological importance of a given host.
As the distribution of Fig 1 predicts the most important potential mammal hosts of Leishmania (L.) mexicana, the deviation of the set of real ε values from the two random benchmarks considered, the ε = 0 line and the average random ε curve, shows that the distribution of Lutzomyia is strongly, positively correlated with a large number of mammal species, and, further, that this fact is not explainable by collection biases. Furthermore, the strong asymmetry of the distribution, with 149 potential hosts associated with statistically significant positive correlations with sandflies, compared with only two species with statistically significant negative correlations, is consistent with a hypothesis that sandflies are generalists that are capable of feeding on, and potentially infecting, whatever potential mammal species are available, as has been suggested for several mosquitoes [58]. If sandflies were specialists, associated with a few focal species, one would expect to see high ε values only for those particular species, due to the fact that the focal species are an important and necessary biotic element in the niche of the sandfly, while the other mammals would be incidental and therefore one would expect to see either random association or a positive association mediated by, say, abiotic variables.
The most important results of this paper are in Figs 2 and 3, where we clearly see the significant correlation between the probability for a species or individual to be a host and ε as a measure of the statistical significance of the degree of overlap between vector and host distributions. This is evidence that although many other factors, such as species abundance, species competence etc., enter, the fact that a vector and host must co-occur in order to interact leaves a significant predictive imprint. In other words, the more overlap the more opportunity for host-vector interactions. The multi-factorial nature of the complex relation between host and vector is implicit in that the relation between host probability and ε, although statistically significant, is not characterized by a very high value of R2. Especially noteworthy is the decile 1 result, where 3 out of 7 species were identified as hosts—Baiomys taylori, Peromyscus maniculatus and Peromyscus eremicus. These were collected from sites in Jalisco and Nuevo Leon, states from where collections of Lutzomyia were previously scarce or non-existent. However, recently, it has been confirmed that various species of Lutzomyias are relatively common in these areas. Thus, we believe that a part of the low ranking of these species is due to a systematic bias in the historic collection of Lutzomyias towards the southern part of Mexico. The impact of this on the relation between the percentage of positive host species or individuals and ε is substantial. A regression using only the first 9 deciles yields an R2 = 0.92 versus 0.44 for all 10 deciles for the species-ε relation, while for the individuals-ε relation the corresponding figures are 0.65 and 0.39. Figs 2 and 3 also illustrate what we mean by host importance. The fact that the percentage of infected individuals decreases as a function of epsilon is consistent with the fact that in the higher deciles (higher ε values) there is a higher probability of a vector encountering an infected host than in the lower deciles (lower ε values). This does not imply, of course, that the pathogen is transmitted. It is however, once again, a necessary if not sufficient condition.
Besides validating both the general methodology as well as the specific model for Leishmaniasis of [36], our results also provide an extensive list of new hosts for Leishmania (L.) mexicana in Mexico that substantially changes what we know about the transmission cycle of the pathogen and the potential efficacy of interventions and/or surveillance efforts. As 33% of collected species tested positive for the presence of L. (L.) mexicana our results are completely consistent with the prediction that L. (L.) mexicana is a very generalist pathogen and that Lutzomyia is a very generalist genus. Furthermore, if we consider the probability that a species is a true negative at the 95% confidence level with respect to the null hypothesis of a 6.7% infection rate across all species, then we see that there are no true negatives. The closest is Heteromys desmarestianus with N = 30 and a probability of 88% of being a true negative. Interestingly, though, Heteromys desmarestianus has previously been identified as a host [49,59]. We also note that, of the 24 identified host species, only four—Phyllostomus discolor, Peromyscus eremicus, Glossophaga commissarisi and Glossophaga soricina—are associated with a prevalence that, with 95% confidence according to the Wilson score interval, is greater than the average prevalence across all species. Moreover, the first two species were associated with only one collection so that one would not expect the Wilson score interval to be reliable. The exact probability for an observed 100% prevalence with N = 1 is 6.7%.
Our principle result, as stated previously, is the validation of the methodology and the explicit model of [36] for Leishmaniasis. However, we may also further analyse our experimental results. We have noted that they are quite consistent with the hypothesis that most host species have prevalence values that are compatible with the null hypothesis of a constant prevalence of 6.7% across species. For those positive species with N > 20 in fact prevalence is very stable. Of course, heterogeneities are to be expected. This can be due not only to intrinsic differences in host competence but also, potentially, to spatial heterogeneity associated with epidemiological “hotspots” where many variables together may be favourable for transmission. In Fig 4 we see that there is no noticeable relation between prevalence and ε. We would argue that there should be a dependence if prevalence is averaged for a given species over several geographical locations that systematically sample the range of that species. However, the sampling intensity of the present data is not capable of showing such effects at a species by species level. The fact that the species by species infection rates are consistent with a relatively constant 6.7% prevalence is compatible with the hypothesis that the competence of the different mammal host species is relatively homogeneous. However, as can be seen in Fig 4, the degree of dispersion of the data is large. Partly this is due to the fact that several of the observed prevalences are associated with very small sample sizes. For instance, the species with prevalence of 100% are associated with only one individual. These can be considered as outliers. This heterogeneity in sample size at the species level is one reason why a coarse grained analysis, as observed in Figs 2 and 3, is more appropriate. Taken at face value, the relative homogeneity of prevalence would also argue against any potential dilution effect due to higher biodiversity as this depends on strong competence heterogeneity amongst hosts. In fact, these results would be consistent with the fact that Lutzomyias do not differentiate very much among different potential sources of blood meal [60].
The results of this study stand in stark contrast to our previous understanding of the hosts of different species of Leishmania: Of the more than 2000 mammal species on the American continent only about 50 (2.5%) have been identified as hosts of Leishmania [44], while in Mexico, 8 out of 419 (2.1%) have been identified as hosts [but see 61]. If our result that 33% of collected mammal species are hosts is extrapolated to other non-collected species then potentially hundreds of mammal species could be implicated as hosts of Leishmania. Of the collected species that tested positive, 13 were bats, identified for the first time as hosts of Leishmania (L.) mexicana in Mexico [61]. We also identified the grey squirrel as a peri-domestic host species with a high degree of contact with human settlements [62].
If sandflies are generalists, in that they feed off a large number of species across many genera, and pathogen competencies are relatively uniform, as is consistent with our observations, then one would also hypothesize that Leishmania (L.) mexicana is also a generalist in that it can infect a large number of potential host species. In the case of Chagas disease, the impressive genetic plasticity and associated adaptability of Trypanosoma cruzi has been amply studied [63–65], as well as its epidemiological implications. Our results suggest that L. (L.) mexicana should also demonstrate a high degree of genetic plasticity and adaptability to be able to infect such a wide array of mammal species. Recent work on the genetics of Leishmania seems to be consistent with this viewpoint [66,67].
Given that previously there were only eight confirmed host species of Leishmania in Mexico, the results of this paper change the risk landscape for this neglected disease in Mexico, both geographically, and in terms of its control or elimination, with similar potential consequences for other countries. Although it is known that Lutzomyias have an ample geographic range and therefore are a risk element for Leishmaniasis in many areas of Mexico, what the present work demonstrates is that there are a large number, and potentially very many more, of hosts involved in the transmission cycle of Leishmaniasis. This complicates both interventions and surveillance. In terms of surveillance we would argue that our model yields a good first approximation as to which host species to survey—those that have been identified as hosts and have the highest ε values. Of course, further field and laboratory work should be carried out to better understand the underlying factors influencing host competence, both at the species level and across different geographical locations for the same species. It is also necessary to test species that are high on the model list but up to now have not been checked. In terms of interventions, with such a large host range, that spans both sylvatic and peri-domestic species, it will be very difficult to eliminate any enzootic transmission cycle, with consequential difficulties in the long term elimination of the vector. Geographically, as can be seen in Fig 5, a risk map derived from the distributions of the eight host species known before the results of this study is completely different to a risk map associated with the 30 species of host that we have now confirmed indicating much higher risk in states such as Jalisco and Nuevo Leon that have until recently been considered low risk compared to the south east.
Our results show that within species collection data there is a great deal of useful information about interspecific interactions and community structure that may be deduced with innovative modelling techniques, such as complex inference networks, and applied to important problem areas such as multi-host diseases. It also shows the importance of the systematic and unbiased collection of data associated with the distributions of potential vectors and potential reservoirs. For instance, the models created showed a higher risk for presence of Lutzomyia in the north of Mexico, along the border with the US, than had previously been the case. In fact, recent human cases have been reported in this region, and recent field work has shown that the presence of Lutzomyia is more extensive than had been previously thought [44,68].
|
10.1371/journal.pgen.1005030 | The Complex Contributions of Genetics and Nutrition to Immunity in Drosophila melanogaster | Both malnutrition and undernutrition can lead to compromised immune defense in a diversity of animals, and “nutritional immunology” has been suggested as a means of understanding immunity and determining strategies for fighting infection. The genetic basis for the effects of diet on immunity, however, has been largely unknown. In the present study, we have conducted genome-wide association mapping in Drosophila melanogaster to identify the genetic basis for individual variation in resistance, and for variation in immunological sensitivity to diet (genotype-by-environment interaction, or GxE). D. melanogaster were reared for several generations on either high-glucose or low-glucose diets and then infected with Providencia rettgeri, a natural bacterial pathogen of D. melanogaster. Systemic pathogen load was measured at the peak of infection intensity, and several indicators of nutritional status were taken from uninfected flies reared on each diet. We find that dietary glucose level significantly alters the quality of immune defense, with elevated dietary glucose resulting in higher pathogen loads. The quality of immune defense is genetically variable within the sampled population, and we find genetic variation for immunological sensitivity to dietary glucose (genotype-by-diet interaction). Immune defense was genetically correlated with indicators of metabolic status in flies reared on the high-glucose diet, and we identified multiple genes that explain variation in immune defense, including several that have not been previously implicated in immune response but which are confirmed to alter pathogen load after RNAi knockdown. Our findings emphasize the importance of dietary composition to immune defense and reveal genes outside the conventional “immune system” that can be important in determining susceptibility to infection. Functional variation in these genes is segregating in a natural population, providing the substrate for evolutionary response to pathogen pressure in the context of nutritional environment.
| Previous studies have indicated that dietary nutrition influences immune defense in a variety of animals, but the mechanistic and genetic basis for that influence is largely unknown. We use the model insect Drosophila melanogaster to conduct an unbiased genome-wide mapping study to identify genes responsible for variation in resistance to bacterial infection after rearing on either high-glucose or low-glucose diets. We find the flies are universally more susceptible to infection when they are reared on the high-glucose diet than when they are reared on the low-glucose diet, and that metabolite levels genetically correlate with quality of immune defense after rearing on the high-glucose diet. We identify several genes that contribute to variation in defense quality on both diets, most of which are not traditionally thought of as part of the immune system. The genetic variation we observe can be important for evolved responses to pathogen pressure, although the effectiveness of natural selection will be partially determined by the host nutritional state.
| There is strong intuition that dietary nutrition affects the quality of immune defense, and this intuition is well supported scientifically. Starvation increases susceptibility to infection in insects as well as humans [1,2], and specific dietary components such as vitamins, carbohydrates, and proteins have been implicated in shaping immunity to bacterial infection [3–7]. Elevated dietary protein relative to sugar increases standing levels of immune activity in Drosophila melanogaster [8], and diets deficient in protein increase susceptibility to infection by Salmonella typhimurium in mice [6]. Nutrition alters development in ways that may have immunological import [9–11], and insects and other animals alter their feeding behavior in response to infection [12,13]. There is growing evidence that the ratio of protein to carbohydrates (P:C) in the diet may specifically influence several life history traits[11,14–18], including some that may predict resistance to infection. For example, the African army worm Spodoptera exempta becomes more susceptible to infection by the bacterium Bacillus subtilis when supplied with diets high in sugar relative to protein, and infected caterpillars will actively choose to eat diets higher in protein without increasing sugar intake [13]. These and other such observations have led to the suggestion that “nutritional immunology” should be employed to identify ideal dietary compositions for the combat of infection [4]. However, despite the increasingly clear impact of diet on resistance to infection, we have remarkably little insight into how nutrition alters infection outcomes, and whether or why individuals in natural populations differ genetically in their immunological response to diet.
Natural populations are rife with genetic variation for traits that determine health and evolutionary fitness, and both human and Drosophila populations are genetically variable for the ability to fight bacterial infection [19,20]. Such variation may occur in intuitively evident genes, such as those that make up the immune system [21,22], but phenotypically important variation may also map to less obvious genes that shape host physiological context. Even traits that have strong genetic determination can be influenced by the environment, including the availability of nutrition [23,24]. Importantly, different genotypes can vary in their susceptibility to environmental influence, resulting in traits that are determined by the interaction between genotype and environment (GxE) [25]. In very few cases, however, have the genes underlying sensitivity to environment been determined, and it is indeed difficult to predict a priori what the genes for environmental sensitivity might be. The genetic variation that controls both direct trait determination as well as that that controls environmentally influenced phenotypic variation are critically important to the health and evolutionary potential of populations.
We have previously used candidate-gene based approaches to map the genetic basis for variation in Drosophila melanogaster resistance to bacterial infection [26–28]. These studies were successful in identifying naturally occurring alleles that shape defense quality, but they focused exclusively on genes in the immune system. While we may expect diet to shape resistance to infection, we have no particular expectation that the effects of diet act through the canonical immune system (i.e. Toll and IMD pathways [29,30]). Dietary composition has widespread metabolic and developmental consequences, and these consequences vary quantitatively and qualitatively among genetically diverse Drosophila [31]. There is evidence for crosstalk between metabolic signaling pathways such as insulin-like signaling and canonical immune pathways in Drosophila, both during development and in the initiation of an immune response [32–36]. Thus, it is plausible to imagine that the immunological effect of diet, and especially genetic variation in immunological response to diet (genotype-by-diet interaction), could be controlled by genes outside of what is typically conceived to be the “immune system.”
In the present study, we conduct an unbiased genome-wide association study to identify genes that shape variation in resistance to bacterial infection among D. melanogaster reared on either a high glucose or low glucose diet. Specifically, we deliver experimental infections with the bacterium Providencia rettgeri and measure systemic pathogen load 24-hours post infection. This time point both provides a robust estimate of infection intensity [37] and correlates strongly with risk of mortality [38]. Throughout the manuscript we will refer to pathogen load as “resistance” or “immune defense”. We find that flies reared on a high glucose diet harbor significantly higher pathogen loads and substantially altered metabolite levels, including elevated free glucose, glycogen and triglycerides. Although there is considerable natural genetic variation for resistance to infection on both diets, resistance is generally well correlated across the two diets. Nonetheless, we find evidence of genotype-by-environment interactions determining immune defense, as well as metabolic alterations that correlate genetically with resistance in flies reared on the high glucose diet. We are able to map and validate several genes that contribute to variation in resistance in both diet-independent and diet-dependent manners. Importantly, most of these are not typically considered part of the canonical immune system.
We found considerable natural genetic variation for immune defense segregating within the Drosophila Genetic Reference Panel (DGRP), where the quality of defense is defined as the ability to limit pathogen proliferation. We infected male flies from 172 of the complete genome-sequenced lines [39] with the Gram-negative bacterium Providencia rettgeri after rearing on either a high glucose or low glucose diet in a replicated block design (see Methods), then measured systemic pathogen load 24 hours later. Pathogen load was significantly predicted by line genotype and diet (Table 1; S1 Fig, p < 10-4 for both) as well as by a genotype-by-diet interaction (p = 0.0016), indicating that genotypes differ in their immunological sensitivity to dietary glucose. Nonetheless, pathogen load was highly correlated across the two diets (Pearson r = 0.69, p < 10-4; Fig. 1), indicating a strong main effect of genotype on immune performance. On average, flies reared on the high-glucose diet sustained systemic pathogen loads approximately 2.4 times higher than those of flies reared on the low-glucose diet.
We measured several indices of nutritional status in each Drosophila line after rearing on the high-glucose and low-glucose diets because we predicted that specific metabolite profiles might be associated with changes in immunity. We measured free glucose, glycogen stores, total triglycerides, free glycerol, soluble protein, and wet mass, as these provide an overall picture of an individual’s nutritional status. The Nutritional Indices (NIs) showed predictable responses to diet. For example, levels of glucose, glycogen, and triglycerides were substantially elevated by rearing on the high-glucose diet (Fig. 2; p < 10-4 in all cases), although wet weight and free glycerol were significantly reduced by rearing on high glucose (Fig. 2; p < 10-4 in both cases). The lines exhibited highly significant genetic variation for all NIs after rearing on either diet (p < 10-4 in all cases; Table 2). Each NI was significantly genetically correlated across diets (Fig. 3), indicating strong genetic determination of NIs regardless of diet. Surprisingly, only wet mass, glycogen and free glucose showed strong genotype-by-diet interactions (Table 1).
Since we found that increasing dietary glucose resulted in increased pathogen load as well as alteration of metabolic profile, we asked whether metabolic profile correlated with pathogen load across genotypes. The only NI that correlated with pathogen load was free glucose, which was slightly negatively correlated with P. rettgeri load on the high-glucose diet (Pearson’s r = -0.18, p = 0.033). This is somewhat surprising given that the general effect of increased dietary glucose is both elevated blood glucose and an increase in pathogen load, and may indicate that variation in pathogen load is associated with rates of conversion between molecules.
We hypothesized that genetic variation might shape the relationship between overall metabolic state and immune defense and that our nutritional indices might give more information about the overall metabolic status of the fly when considered in aggregate. We therefore performed a principal component analysis and tested whether the primary principal components (PCs) for each diet correlated with immune defense quality. The top five PCs summarizing the NIs on each diet each explain 8–41% of the total variance in nutritional state, with loadings of each NI given in Table 3. None of the metabolic PCs correlated with pathogen load on the low-glucose diet. The fourth PC on the high-glucose diet was significantly correlated with pathogen load (Pearson’s r = -0.27, p = 0.001; Table 3). This PC, which explains 11% of the total variance, is heavily positively loaded with free glucose (0.58) and soluble protein (0.36) and is negatively loaded with glycogen stores (-0.71), consistent with the observation that free glucose alone is negatively correlated with pathogen load. This is the only PC where free glucose and glycogen stores load in opposite directions, possibly indicating the rate of conversion between dietary glucose to glycogen. PC1 trends toward negative correlation with pathogen load on the high-glucose diet (Pearson’s r = -0.15; p = 0.06). This PC explains 37% of the variance and is positively loaded with all NIs, and presumably reflects overall fly mass although mass itself does not correlate with pathogen load (Fig. 3).
When we compared our data to previously published DGRP phenotype data [39], we found correlations between our NIs and three metabolism-related traits: starvation resistance, chill coma recovery, and startle response. Starvation resistance, as determined by Mackay et al. [20], is positively correlated with all of our NIs except soluble protein with correlation coefficients ranging from 0.169 (Table 4, p = 0.044) to 0.388 (p < 10-4). Overall, genotypes with greater energy reserves were better able to withstand the stress of starvation: measures of wet mass, soluble protein, and glycogen stores were significantly negatively correlated with time to recovery from chill coma as measured by Mackay et al. (chill coma recovery correlated with wet mass: r = -0.235, p = 0.005; soluble protein: r = -0.248, p = 0.003; glycogen: r = -0.20, p = 0.016). Our measures of free glucose levels and total triglycerides were weakly correlated with Mackay et al.’s measure of startle response (Pearson’s r < 0.20 and p < 0.05 for each). This consistency in related phenotypic measures and relationships across lab groups indicates the genetic robustness of the phenotypes.
The bacterial endosymbiont Wolbachia pipientis has been shown to confer protection against RNA viruses in Drosophila [40,41], but previous experiments have not uncovered any protective benefit of Wolbachia against secondary bacterial infection [42,43]. Richardson et al. [44] determined that 52% of the lines in the DGRP are infected with Wolbachia, and we find Wolbachia status to be a weakly significant predictor of P. rettgeri load on both diets (S2 Fig, low glucose: p = 0.0361; high glucose: p = 0.0327; data from both diets combined: p = 0.014), with lower average bacterial loads in the Wolbachia-infected lines than in the Wolbachia-uninfected lines.
Because the complete genomes have been sequenced for every line in the DGRP, we were able to conduct unbiased genome-wide association mapping for each of our measured phenotypes. We used mixed effect linear models to identify genetic polymorphisms that predict systemic pathogen load. Using a significance criterion of 10-6 we identified seven single nucleotide polymorphisms (SNPs) in six genes that associate with variation in pathogen load on the high-glucose diet, 11 SNPs in 9 genes that associate with load on the low-glucose diet, and 19 SNPs in 12 genes that associate with pathogen load when the data from both diets is pooled (Table 5; S3 and S4 Figs). This significance threshold corresponds to a false discovery rate of 5–10% (depending on the phenotypic distribution of the particular trait being evaluated and the details of the analytical model) and provided a reasonable number of SNPs for further characterization. Several of the mapped SNPs were common to multiple analyses. Overall, we mapped SNPs in the genes crinkled, defective proboscis extension response 6, diptericin, elk, fruitless, kinesin heavy chain 73, multiplexin, Scr64B, sema-1a, tout velu/CG12869, CG42524, CG7991, CG4835 and CG15544. We additionally mapped SNPs to 2 distinct regions annotated to encode small, nontranslated RNAs, potentially revealing variation for more complex regulation of the immune system [45].
Our mapped SNPs are highly enriched for lying within or adjacent to genes, with 21 of the total 23 (91.3%) lying within 5 kb of an annotated gene (Table 5). In contrast, only 55% of SNPs genome-wide lie within 5 kb of a known gene. Of our 5 mapped SNPs within gene coding regions, 2 are synonymous and 3 are nonsynonymous, again in stark contrast to the genome average, for which there are approximately 2.7 synonymous polymorphisms for every nonsynonymous polymorphism [46]. One of the two synonymous variants we mapped is in perfect linkage disequilibrium with an amino-acid-altering SNP in Diptericin. The other is in perfect disequilibrium with a 3′ UTR variant of kinesin heavy chain 73. Thus, both of our mapped synonymous SNPs can be considered to be redundant with more plausibly functional SNPs. We used RNAi to knock down 13 of the mapped genes, 9 of which resulted in significantly altered pathogen load either on a standard diet or in a diet-specific manner (S5 Fig, S1 Table). In contrast, only one of five control genes chosen by virtue of physical proximity to mapped genes yielded an altered bacterial load phenotype after RNAi knockdown.
To identify SNPs that have strongly diet-dependent effects on immunity, we first considered SNPs that had significant effects (p < 10-6) on one diet but not on the other (p > 10-4; Fig. 4). Only a few SNPs meet this criterion. One SNP in crinkled (2L.15045678) was significantly associated with variation in immunity on the high glucose diet (p = 1.94 x 10-7) but not on the low glucose diet (p = 0.0074). A SNP in Sema-1a (2L.8630728) was very on the brink of significance on the high glucose diet (p = 1.42 x 10-6) and nowhere near our significance threshold on the low glucose diet (p = 0.001). Reciprocally, one SNP in elk (2R.13779189) was significant on the low glucose diet (p = 6.75 x 10-7) but not on the high glucose diet (p = 4.49 x 10-4). All SNPs with p <10-4 on either diet were significant at p < 10-6 when the data from both diets were pooled.
Our second approach to finding genes with significant diet-dependent effects was to pool the data from both diets and evaluate the SNP-by-diet interaction in a second GWAS analysis. While this approach resulted in a somewhat liberal inflation in P-values (S4b Fig), it revealed SNPs in several genes has having diet-dependent effects at a nominal threshold of p < 10-6. Of the genes mapped with this second approach, we chose TepII, gprk2, and similar to test by RNAi, and confirmed the importance of these genes on suppression of P. rettgeri proliferation (see below).
To determine whether any gene function categories were enriched in our set of significantly mapped SNPs, we performed a GO enrichment analysis using GOWINDA [47], which corrects for gene size, on the reduced GO category list defined by GO Slim [48]. Because so few SNPs mapped significantly at our cutoff of p<10-6, we performed the GO analysis at a significance threshold of p<10-5. Categories related to immunity and metabolism were among the most enriched, but no functional categories were significantly enriched after multiple correction (S2 Table). GO analysis of GWAS results implicitly assumes a quantitative genetic model where many genes in every relevant functional process each contribute small but significant effects on the overall phenotype. We have no evidence that this is an appropriate conceptual model for our defense phenotype, so we did not pursue the GO analysis further.
We performed genome-wide association mapping of each of the nutritional indices, yielding several hits in or near genes with reasonable links to metabolic status [49]. We found no overlap between SNPs significantly associated with variation in the NIs and those significantly associated with variation in immune defense. The genetic basis for altered nutritional status in response to diet will be the subject of an independent paper [49].
Diptericin. Diptericin is a antimicrobial peptide that is produced in response to DAP-type peptidoglycan that makes up the cell walls of Gram-negative bacteria such as P. rettgeri [50,51]. Two SNPs in perfect linkage disequilibrium (2R.14753586—synonymous, 2R.14753589—nonsynonymous) are significantly associated with variable suppression of P. rettgeri infection in flies reared on both diets (p = 9.03 x 10-7 for each SNP on high glucose and p = 2.93 x 10-7 on low glucose) as well as when data from the two diets are pooled (p = 7.04 x 10-08 for each SNP). While it might seem intuitive that an antibacterial peptide gene would map in an immunity screen, this result was surprising as we have not identified any marked effect of Diptericin in previous association studies using other Gram-negative bacterial infections [19,27]. Indeed, it is generally believed that there is enough redundancy in AMPs that mutations in a single peptide would have little effect on organism-level immunity [e.g. 52]. The nonsynonymous SNP (2R.14753589) results in a serine versus arginine polymorphism segregating in the population. In the DGRP, the more resistant serine allele is carried by 82% of lines and the more permissive arginine by 14% of lines (4% of lines are heterozygous at the SNP). Two of the DGRP lines are homozygous for a premature stop codon in Diptericin at position 2R.14753502. While this stop codon did reach not our minor allele frequency threshold for consideration in the study, we thought it was notable that two lines carrying the premature termination exhibited the absolute highest bacterial loads across the entire DGRP mapping panel. Both of these lines carried the higher-resistance serine variant at 2R.14753589, thereby slightly decreasing the statistical significance of the independent contrast between the serine and arginine variants. If these two lines are excluded from the analysis, the P-value for 2R.14753589 is 4.43x10-9. Interestingly, we found that serine and arginine are also segregating in Drosophila simulans through an independent mutation at the same codon, suggesting the possibility of convergent balancing selection (Unckless et al. in prep.; see Discussion).
Multiplexin. Multiplexin (mp) encodes a collagen protein. Multiplexin is a huge gene (55 kb) with 15 annotated transcripts. Annotated molecular functions include carbohydrate binding and motor neuron axon guidance [53]. Loss-of-function mutants have smaller larval fat bodies than wild-type flies [54], which may be relevant since the fat body is the primary tissue that drives systemic immunity to bacterial infection. Three intronic SNPs in mp are significantly associated with variation in P. rettgeri load in flies reared on either the high glucose (p = 2.25 x 10-7) or low glucose (p = 4.12 x 10-6) diet, as well as when data from both diets are pooled (p = 3.9 x 10-7). Ubiquitous RNAi knockdown of mp resulted in significantly decreased P. rettgeri load after infection relative to controls with wild-type mp expression (p = 0.017). The relationship between resistance and the larval fat body phenotype in multiplexin mutants may suggest a role for the humoral immune response in this phenotype. Mutant flies may have altered antimicrobial peptide expression.
Defective proboscis extension response 6. An intronic SNP (3L.10039434) in Defective proboscis extension response 6 (Dpr6) was associated with variation in P. rettgeri load in flies reared on the low glucose diet (p = 9.54 x 10-8) and when data from flies reared on both diets were pooled (p = 3.51 x 10-7). Dpr6 belongs to a family of genes thought to be involved in sensory perception of chemical stimulus, including gustatory perception of food, and contains an immunoglobulin domain that may be involved in cell-cell recognition [55]. Ubiquitous RNAi knockdown of dpr6 resulted in a significant decrease in P. rettgeri load after infection (p = 0.0097).
Crinkled. An intronic SNP (2L.15045678) in crinkled mapped for variable resistance specifically on the high glucose diet (p = 1.94 x 10-7), and less significantly when data from both diets were pooled (p = 4.29 x 10-5), but was not significantly associated with variation in P. rettgeri load when flies were reared on the low glucose diet (p = 0.0075). Crinkled encodes myosin VIIa, an actin-dependent ATPase. RNAi knockdown experiments for crinkled suggest that it does influence immunity in a diet-dependent manner. We used ubiquitous RNAi to knock down ck in flies reared on either the high glucose or low glucose diet. The knockdown had no significant effect on P. rettgeri load of flies reared on the low glucose diet (p = 0.45) but was marginally significant when flies were reared on the high glucose diet (p = 0.07; Fig. 5a). Further exploring the diet dependence, we found that Principle Component 4 of our nutritional indices measured on the high glucose diet correlated with P. rettgeri load in a ck allele-dependent manner. PC4 is strongly negatively correlated with bacterial load in flies homozygous for the A allele (r = -0.417, P = 0.0014) but is uncorrelated with load in flies bearing the T allele (r = – 0.115, P = 0.329; Fig. 5b). Since PC4 is loaded primarily with glucose, protein and glycogen, we also examined correlations between these NIs and P. rettgeri load within each ck allele in flies reared on the high glucose diet (S6 Fig). Mirroring the overall phenotypic data, free glucose levels trended toward negative correlation with P. rettgeri load in flies bearing the A allele (r = -0.23, p = 0.088) but not in flies bearing the T allele (r = -0.12, p = 0.32). Glycogen levels trended toward positive correlation with P. rettgeri load within the A allele (r = 0.20, p = 0.132) but not within the T allele (r = -0.07, p = 0.58).
Sema-1a. An intronic SNP (2L.8630728) in Sema-1a, which encodes a semaphorin protein, fell just below our significance threshold on the high glucose diet (p = 1.42 x 10-6) and was much less significant on the low glucose diet (p = 0.0011). The dramatic difference in effect on the different diets suggested to us that Sema-1a might have diet-dependent effects on immunity. Semaphorins tend to be highly pleiotropic and play major roles in developmental processes [56]. RNAi knockdown of Sema-1a resulted in flies with marginally significantly higher P. rettgeri loads than controls on the low glucose diet (p = 0.081), on the high glucose diet (p = 0.088), and when the data from both diets were combined (p = 0.025).
CG12869. A SNP (2R.10477114) 1594 bp upstream of functionally unannotated gene CG12869 was significantly associated with P. rettgeri load in flies reared on the low glucose diet (p = 7.65 x 10-7) and approached significance in flies reared on the high glucose diet (p = 8.87 x 10-5) and when the data from both diets were combined (p = 1.49 x 10-6). While little is known about CG12869, the encoded protein is predicted to have carboxylesterase activity. RNAi knockdown of CG12869 in flies reared on either the high-glucose or low-glucose diet resulted in modestly increased P. rettgeri loads when the data from both diets were combined (ppooled = 0.047), although not when either diet is considered independently (low glucose: p = 0.133, high glucose: p = 0.164).
G protein-coupled receptor kinase 2. Gprk2 was previously associated with defense response to bacteria through interaction with cactus and is required for normal AMP production [57]. It is also involved with several biological processes that might be influenced by nutritional environment including hedgehog signaling and regulation of appetite {Cheng:2012kd, Chatterjee:2010df}. An intronic SNP in Gprk2 (3R.27273757) yielded a significant SNP-by-diet interaction predicting P. rettgeri load (p = 7.75 x 10-8). RNAi knockdown of Gprk2 resulted in increased P. rettgeri load relative to control flies (p = 0.016), consistent with the results of Valanne et al. [57], who found that Gprk2 disruption reduced resistance to infection by Enterococcus faecalis. We found no distinction between knockdown on high-glucose versus low-glucose diets (pknockdown = 0.08, pdiet = 0.04, pinteraction = 0.94).
Thioester-containing protein 2. Thioester-containing proteins (TEPs) are opsonins that promote phagocytosis and parasite killing in invertebrates, including phagocytosis of Gram-negative bacteria [58]. TEPs are homologous to vertebrate complement C3 and macroglobulins, and Drosophila TepII has previously been shown to evolve under adaptive positive selection in the presumptive pathogen-binding domain [59]. P. rettgeri load was determined by a significant diet*SNP interaction for four nonsynonymous SNPs in TepII (p = 5.97 x 10-7) and an additional synonymous SNP in tight disequilibrium (p = 7.00 x 10-7). RNAi knockdown of TepII resulted in reduced immune defense (p = 0.0017), independent of diet (p = 0.94), which is consistent with the known role of TepII in insect immunity.
Similar. An intronic SNP in similar (3R.25909307) showed a significant Diet*SNP interaction (p = 9.17 x 10-7). Sima is involved in protein dimerization and signal transduction and has been associated with response to stress. RNAi knockdown of similar resulted in increased P. rettgeri load after infection of flies reared on the low glucose diet (p = 0.005) but not on the high glucose diet (p = 0.28). Variants of similar may influence how an individual responds to a nutrient-poor diet which in turn may influence their ability to resist infection.
Kinesin heavy chain 73 and Src64B. A synonymous SNP and a SNP in the 3′ UTR of Khc-73 were associated with variation in bacterial load when data from both diets are pooled (3.1 x 10-7 and 5.48 x 10-8, respectively), and an intronic SNP (3L.4603286) in Src64b mapped highly significantly on each diet (low glucose: p = 1.41 x 10-7; high glucose: p = 2.50 x 10-7) and when the data from both diets were pooled (p = 3.34 x 10-9). Khc-73 is a microtubule motor protein [53] and Src64b is a tyrosine kinase with a wide range of reported phenotypes including cellular immune response [60]. RNAi knockdown of either gene did not result in any significant change in systemic P. rettgeri load after infection (Khc-73: p = 0.67, Src64b: p = 0.97). Thus, neither of these mapped genes validated by our RNAi knockdown criteria. This could be because the two genes are false positive map results or because the RNAi failed to adequately block protein synthesis in the knockdown experiment.
Other candidate genes. We mapped SNPs associated with variation in post-infection P. rettgeri load in the genes elk, fruitless, tout velu, CG42524, CG7991, CG4835 and CG15544 (Table 5), but we were unable to establish RNAi knockdowns for these and were thus unable to test whether disruption of these genes influences resistance to infection.
Nearest-neighbor negative controls. It is unknown what proportion of genes in the genome could conceivably yield immune defense phenotypes when ubiquitous RNAi disrupts their expression. To estimate the false-positive rate on our RNAi knockdowns, we additionally knocked down several arbitrary genes that are physically adjacent to our mapped genes but are not known to have any immune function. Whereas 9 out of 13 of our mapped candidate genes yielded defense phenotypes upon RNAi knockdown, only one out of the five arbitrary neighboring genes yielded an immunity phenotype. Little is known about the function of that arbitrary gene whose knockdown resulted in a modest decrease in P. rettgeri load (p = 0.029) (CG34356), but it has been shown to be involved in protein phosphorylation [61]. Our rate of 9 in 11 positive knockdown experiments among the mapped candidates is a significant excess over the 1 in 5 negative control genes that gave immune phenotypes (Fisher’s Exact Test: p = 0.018), giving us confidence that the majority of our mapped genes are true positive results.
Diptericin is a classical immunity gene with a large effect in our study. We reasoned that genotype at Diptericin might mask genes with smaller effects, and that we could increase power to detect diet-dependent variants by controlling for Diptericin genotype. Furthermore, there was significant linkage disequilibrium between SNPs in Diptericin and other mapped SNPs (S7 Fig). We therefore re-conducted the genome-wide association analysis with the addition of Diptericin genotype as a covariate that could take on three possible states: the arginine versus serine variants at position 2R.14753589 and the premature stop codon (although the two DGRP lines carrying the premature stop codon also carried the serine variant, we classified them separately because they were phenotypically so extreme). Lines carrying residual heterozygosity at Dpt were treated as having missing data for the Dpt genotype. All GWAS results and knockdown experiments reported to this point were mapped without Dpt genotype as a covariate. Unexpectedly, instead of revealing new genes that predict immune phenotype, inclusion of Dpt genotype in the mapping model caused the number of significant SNPs (p < 10-6) to drop from 19 to only 4 when the data from both diets were pooled, from 11 to 2 on the low glucose diet only, and from 7 to 1 on the high glucose diet only. Inclusion of Dpt as a covariate greatly improved the observed fit of our q-q plots to the null expectation, eliminating experiment-wide p-value inflation (S4 Fig). We observed an increase in the number of SNP-by-diet interactions from 77 to 88 when Dpt genotype is included as a covariate (S4 Fig, S3 Table). Only one SNP (2L.13072327; located in a small RNA) was significant in both the original models and when Dpt genotype was used as a covariate. For the SNPs significant for the interaction effect, 66 were significant in both methods, 14 were specific to mapping without Dpt as a covariate, and 25 were specific to mapping with Dpt as a covariate. As shown in S4 Fig, there is generally good agreement between the two methods for interaction, although both are quite inflated.
To assess whether mapping with Diptericin genotype as a covariate provided reliable results, we performed the same RNAi knockdown experiments as described above with two new candidates genes. Both resulted in increased pathogen load when knocked down (S5 Fig, S1a Table). Briefly, these genes were CG33090, a beta-glucosidase, and CG6495, a gene of unknown function that was significantly induced upon infection in a previous study [62]. We additionally chose to validate CG12004, which mapped with a P-value that missed our significance threshold (p = 4.72 x 10-6), but that has been previously shown to be involved in defense response to fungus [63]. Knockdown of CG12004 resulted in a marginally significant increase in pathogen load (P = 0.0514).
We reexamined the correlations between nutritional indices and bacterial load when Dpt variant was included as a covariate in the regression. In all cases, the model with Dpt variant was a better fit than the model that did not include Dpt genotype (S4 Table). On the high glucose diet, the correlation between free glucose level and bacterial load becomes slightly less significant (p = 0.078 vs. p = 0.033 previously), while the correlation between soluble protein and bacterial load became more significant (p = 0.036 vs. p = 0.061 previously). No principal components on the low glucose diet became significant with Dpt as a covariate. However, on the high glucose diet, PC3 became marginally significant (p = 0.052 vs. 0.194 without considering Dpt genotype) and PC4 remained significant (p = 0.007 vs. 0.001 without considering Dpt genotype).
We found the Drosophila Genetic Reference Panel to be highly variable for resistance to P. rettgeri infection. We also determined that the severity of bacterial infection increased dramatically when flies were reared on a high-glucose diet, and the flies became hyperglycemic and hyperlipidemic. Relative quality of immune defense was highly correlated across the two diets, indicating strong genetic determination of the defense phenotype. However, we also observed a significant genotype-by-diet interaction shaping defense. Specifically, there were several lines that suffered disproportionately severe infections after rearing on the high-glucose diet, although these lines fell closer to the center of the resistance distribution when they were reared on the low-glucose diet. We did not find any lines that showed markedly higher resistance on the high-glucose diet. We were able to identify several genes that contributed to variation in resistance on both diets.
Not only did severity of infection increase with elevated dietary glucose, the flies became hyperglycemic, hyperlipidemic, and had elevated glycogen stores after rearing on the high-glucose diet. Because both glucose levels and infection severity increased with rearing on the high glucose diet, we predicted that those two traits would also be genetically correlated. Unexpectedly, however, free glucose levels were negatively correlated with severity of infection across genotypes when flies were reared on the high glucose diet (the two traits were uncorrelated on the low glucose diet). We observed an even stronger correlation between resistance and a principal component that was positively loaded with free glucose and negatively loaded with glycogen stores. Because metabolic measurements were taken from uninfected flies, they indicate genetic capacity to assimilate or manage the excess dietary glucose in the absence of the pathogen. The genetic correlation with infection severity indicates that resistance to P. rettgeri infection is linked in some way to glucose metabolism, uptake, and/or conversion to and from glycogen. Our map results suggest that this effect is partially mediated by the crinkled gene, which encodes a myosin VIIa cytoskeletal ATPase. We identified a polymorphism in crinkled that highly significantly predicted bacterial load when flies were reared on the high glucose diet, although not on the low glucose diet. Flies bearing the rarer allele show strongly negatively correlated glucose levels and pathogen loads. Furthermore, independently determined expression of the crinkled gene [64] correlates with resistance to P. rettgeri and our observed glucose level. Full characterization of the mechanism by which crinkled shapes immunity and glucose metabolism will require future study.
We were more generally able to map several genes that contribute to phenotypic variation in immune performance, both in diet-specific and diet-independent manners. The mapped polymorphisms were highly significantly enriched for being nonsynonymous and for lying within or very near genes. RNAi knockdown confirmed roles for the mapped genes in resistance to P. rettgeri, with 82% of the knockdowns of mapped genes resulting in altered pathogen loads. In contrast, we found defense phenotypes after knockdown of only 17% of negative control genes that are chromosomally linked to mapped genes but were otherwise arbitrary.
Only a small handful of the mapped genes had annotated immune function. Instead, we identified genes encoding proteins annotated in processes such as feeding behavior and cytoskeletal trafficking. This is a fully expected outcome of the experiment, and such genes are precisely what GWAS studies are designed to detect. Functional variation in dedicated immune genes is probably subject to strong natural selection in the wild, and most variation is probably quickly purged from the population. In contrast, however, populations may retain genetic variation that results in smaller effects on resistance, especially when the primary selection on the gene is for a function other than immune defense. Such genetic variants can then cause a large proportion of the observed phenotypic variance in natural populations, and in mapping panels derived from natural populations, such as the DGRP. The effect on immune defense of knocking down the mapped genes by RNAi was small relative to what might be expected from disruption of core components of the immune system. For this reason, it is unsurprising that these genes have not been discovered in previous mutation screens for susceptibility to bacterial infection. That we are able to map and confirm many of these non-conventional genes opens the possibility of whole new avenues of research and illustrates the value of unbiased genome-wide mapping relative to candidate gene based studies. This result also suggests that resistance to infection, especially in the context of dietary variation, is best viewed as a synthetic trait of the whole organism phenotype and is not determined solely by the canonical immune system. Genes that influence any number of developmental or metabolic processes may carry variation that directly or indirectly influences the ability of the organism to resist infection.
At the outset of this experiment, we might have hypothesized that the genetic basis for immunological sensitivity to diet would map to stereotypical metabolic processes, either because of crosstalk between metabolic and immune signaling, varied ability to incorporate metabolites during development, or variation in the capacity to sequester nutrients from pathogens. However, our mapping did not uncover the most obvious potential metabolic processes, such as insulin-like signaling, carbohydrate metabolism, or energetic storage. Instead, we identified genes with highly diverse function, which indicates a much more nuanced and complex interaction between dietary intake and immune defense. Importantly, because the flies in our study were reared from egg-to-adult on the experimental diet of interest, we do not distinguish between defense-impacting effects that arise during development versus those that manifest during the response to infection. It is important to bear in mind that the effects of allelic variation in the mapped genes could manifest at any stage of development or in any aspect of host physiology that may ultimately influence antibacterial defense. Determining the mechanisms by which the mapped genes influence resistance will require considerable additional study. In most cases, the RNAi knockdowns of mapped genes confirmed an effect of immune phenotype, but did not necessarily recapitulate diet-specific effects on resistance. While RNAi knockdown is a useful tool for confirming the role of mapped genes in immune defense, it is expected that the effect of RNAi knockdown will be much larger than the difference in phenotype between two alleles of the gene. Thus, where the SNP variants may cause modest modification of defense phenotype—perhaps revealed only under certain dietary environments—the RNAi knockdowns are more of a sledgehammer whose effects will be seen under all dietary conditions.
One of the variants that most significantly predicted pathogen load irrespective of diet was an amino acid polymorphism in the canonical antibacterial peptide Diptericin. This was surprising to us, as previous candidate gene studies had failed to detect major effect of allelic variation in Diptericin or any other antimicrobial peptide gene on resistance to Serratia marcescens, Enterococcus faecalis, Lactococcus lactis, or Providencia burhodogranariea [26–28]. Our interpretation had been that AMPs are plentiful and functionally redundant [52], such that minor variation in any one peptide would not have major effect on organism-level resistance. However, in followup experiments we have confirmed that the Serine/Arginine variant mapped in the present study is a strong predictor of resistance to some but not all Gram-negative bacterial pathogens (Unckless et al in prep). Thus, it would appear that the relative importance of Diptericin, and by extension presumably other antibacterial peptides, depends on the agent of infection. Moreover, we have found an independent mutation in natural populations of Drosophila simulans that converges on a Serine/Arginine polymorphism at the same Diptericin codon, with the same consequence for relative resistance to this suite of bacteria. Surprisingly, natural populations of both D. melanogaster and D. simulans are additionally polymorphic for apparent loss-of-function mutations at Diptericin, and flies carrying these variants are highly susceptible to infection by P. rettgeri and other bacteria [38](Unckless et al in prep). The collective data indicate a complex evolutionary history of Diptericin that includes convergent evolution of selectively balanced polymorphisms in two species, with variation in relative resistance to a subset of pathogens.
We found that infection by the endosymbiont Wolbachia pipientis is associated with modest but significant resistance to infection by P. rettgeri. Previous studies have not found differences in Wolbachia-infected vs. uninfected flies in immune system activity or resistance to infection by secondary bacteria, including P. rettgeri [42,43,65]. Our present study is substantially larger than these others, and therefore may have greater power to detect small protective effects of Wolbachia infection. Unlike previous studies which have compared Wolbachia-infected flies to genetically matched lines which were cured of Wolbachia using antibiotics, our present study cannot fully distinguish between the effects of Wolbachia and host genotype. For example, Wolbachia infection status could be associated with general health of the lines and therefore resistance to P. rettgeri infection, or Wolbachia infection could be significantly associated with a genetic polymorphism that also predicts resistance to P. rettgeri. Presence of Wolbachia was weakly associated with a decrease in soluble protein in the present study (p = 0.046), and has been previously shown to alter fly physiology by buffering the effects of excess or deficit in dietary iron [66] and by modulating other metabolic processes including insulin signaling [67]. These physiological impacts may suggest indirect mechanisms by which Wolbachia infection could confer weak protection against infection by pathogens like P. rettgeri.
In summary, we have shown that natural genetic variation for immune defense can be attributed to variation in several genes, with both diet-dependent and diet-independent effects. We also find that metabolic indices are correlated with immune defense when flies are reared on a high glucose diet. Importantly, several of the mapped genes would not be considered conventional “immune” genes, yet we confirm with RNAi knockdown that they pleiotropically contribute to immune defense. The genes mapped in this study harbor allelic variation that shapes the quality of immune defense, and thus may be instrumental in the evolution of resistance to bacterial infection in natural populations experiencing varied dietary environments.
The Drosophila Genetic Reference Panel (DGRP; [39]) is a collection of 192 lines that have been inbred to homozygosity and whose complete genomes have been sequenced. Each line is derived from an independent wild female captured in a fruit market in Raleigh, NC, USA in 2003. We used 172 of the most robust lines for this study, though the exact number and composition of lines varied slightly among replicate blocks of the experiment.
Bacterial infections were performed using Providencia rettgeri strain Dmel, which was isolated as an infection of a wild-caught D. melanogaster [68]. P. rettgeri are Gram-negative bacteria in the family Enterobacteriaceae, and are commonly found in association with insects and other animals. Injection of the Dmel strain of P. rettgeri into D. melanogaster under the conditions used here results in a highly reproducible initial dose of bacteria that proliferates 100–1000 fold over the first 24 hours post-infection, depending on the host fly genotype, with low to moderate host mortality [51]. Bacterial load at 24 hours post-infection correlates strongly with risk of host mortality [38], but pathogen load as a phenotype does not confound resistance and tolerance mechanisms in the way that survivorship does.
We used two experimental diets that varied in glucose content but otherwise had the same composition. The base diet was composed of 5% weight per volume Brewer’s yeast (MP Biomedicals, Santa Ana, CA) and 1% Drosophila agar (Genesee Scientific, San Diego, CA). The high-glucose diet contained 10% glucose (Sigma-Aldrich Corp., St. Louis, MO) while the low-glucose diet contained 2.5% glucose. All diets were supplemented with 800 mg/L methyl paraben (Sigma-Aldrich Corp., St. Louis, MO) and 6 mg/L carbendazim (Sigma-Aldrich Corp., St. Louis, MO) to inhibit microbial growth in the food. RNAi knockdown experiments for SNPs significant when data from both diets were pooled were performed on the “standard diet” which contained 8.2% glucose and 8.2% Brewer’s yeast. Each DGRP line was split and raised in parallel on both diets for at least three generations prior to the start of the experiment to control parental and grandparental effects within dietary treatments, and experimental flies were reared egg-to-adult in the dietary condition being assayed. We recognize that our diets differ in total caloric content as well as protein to carbohydrate ratios [4,13,14]. It is possible that Drosophila change their feeding behavior on the two diets, and that there may even be genetic variation for feeding behavioral response to diet. Our goal in this study is to determine the consequences of excess dietary glucose while remaining agnostic as to the precise cause of any altered nutritional assimilation.
Providencia rettgeri strain Dmel was grown overnight to stationary phase in Luria-Bertani (LB) broth at 37°C prior to each infection day. On the morning of infections, stationary cultures were diluted in sterile LB broth to A600 = 1.0. Male flies from each DGRP line were infected in the lateral scutum of the thorax by pricking with needles (0.10mm, Austerlitz Insect Pins, Prague, CR) that had been dipped in the diluted bacterial suspension, delivering approximately 1000 bacteria to each infected fly. Infections were performed in three blocks for each diet with each block containing all or nearly all DGRP lines under study. Each block for each diet was performed on a different day, with replicate blocks for the two diets interspersed on alternating weeks. Three researchers performed the infections on each experimental day, with lines assigned randomly to infectors within each block. Males aged 3–6 days post-eclosion were infected from each line. All flies were maintained in an incubator at 24°C on a 12-hour light/dark cycle. Infections were delivered approximately 2–4 hours after “dawn” from the perspective of the flies.
Approximately 24 hours after infection, males were homogenized in groups of 3 in 500 ul sterile LB broth. The homogenate was plated on standard LB agar plates using a robotic spiral plater (Don Whitley Scientific). Plates were incubated overnight at 37°C, and the resultant colonies were counted using the ProtoCOL system associated with the plater. P. rettgeri grows readily on Luria agar at 37°C, but the endogenous microbiota of D. melanogaster does not. Thus, we were able to capture colonies derived from viable infecting P. rettgeri without interference from the Drosophila gut microbiota. Counted colonies were visually inspected for morphology consistent with P. rettgeri, and homogenates from sham-infected flies always failed to yield bacterial colonies within the assay period. We used systemic pathogen load at 24 hours post-infection as our measure of immune defense.
In total, 6–9 data points representing 18–27 flies were collected from each line on each diet (high and low glucose). The total experiment consists of 1429 data points representing 4287 flies on the low glucose diet, and 1396 data points representing 4188 flies on the high glucose diet.
We queried a series of nutritional indices in flies reared on each diet. Each metabolite was assayed in three replicates on flies reared on each diet. Males were aged 3–6 days post-eclosion, then 10 live males were weighed using a MX5 microbalance (Mettler-Toledo, Columbus, OG) and homogenized in 200 μL buffer (10 mM Tris, 1 mM EDTA, pH 8.0 with 0.1% v/v Triton-X-100) using lysing matrix D (MP Biomedicals, Santa Ana, CA) on a FastPrep-24 homogenizer (MP Biomedicals, Santa Ana, CA). An aliquot of 50 microliters were frozen immediately while 150 microliters were incubated at 72 degrees C for 20 minutes to denature host proteins. Nutrient assays were performed with minor modifications of the procedures described in [69] using the following assay kits from Sigma-Aldrich (St. Louis, MO): glucose with the oxidase kit (GAGO-20); glycogen using the glucose kit and amyloglucosidase from Aspergillus niger (A7420) in 10 mM acetate buffer at pH 4.6; free glycerol and triglycerides using reagent kits F6428 and T2449, respectively. Soluble protein was assayed with the DC Protein Assay (BIO-RAD, Hercules, CA). Each metabolite was assayed on each pool of weighed and homogenized flies.
Mixed effect linear models were used to test for genetic and other contributions to phenotypic variation in systemic pathogen load and nutritional indices. Overall genetic main effects on systemic pathogen load were tested with the model
Yijklmno= μ+ Wolbi+ Dietj+ Linek(Wolbi) + Infectorl+ Platerm+ Blockn(Dieto) + Dieto*Linek(Wolbi) + εijklmno
where Y is the natural log-transformed measure of pathogen load for each data point, Wolbi (i = 1,2) has a fixed effect and indicates whether the line is infected with the endosymbiotic bacterium Wolbachia pipientis, Dietj (j = 1,2) has a fixed effect and indicates which of the two diets the flies were reared on, Infectorl (l = 1,3) has a fixed effect and is used to test whether the experimentor performing the infections influenced ultimate pathogen load, and Platerm (m = 1,2) has a fixed effect and indicates which of two spiral platers were used to plate the sample. Blockn(Dieto) (n = 1,3) has a fixed effect nested within the effect of Diet, and is used to test for differentiation among the three replicate blocks for each dietary treatment. Linek(Wolbi) (k = 1,172) is assumed to have a random effect, and is used to test the influence of genetic line on pathogen load within Wolbachia-infected and Wolbachia-uninfected classes. The interaction Dieto*Linek(Wolbi) is considered to have a fixed effect and tests whether genetic lines differ in their responsiveness to the two diets. This model was run in SAS 9.3 (Cary, NC) using the “mixed” procedure.
We determined line means for each nutritional index using abundance of metabolite per fly. The model used was analogous to that used for bacterial load:
Yijklmno= μ+ Wolbi+ Dietj+ Linek(Wolbi) + Blockn(Dieto) + Dieto*Linek(Wolbi) + εijklmno
Again, all factors were considered to be fixed except Line(Wolb) and Diet*Line(Wolb) and best linear unbiased predictors (BLUPs) were extracted for further analysis. For comparisons between diets, the model used was Nutrient/fly~Wolb+Line(Wolb)+Block.
To determine whether there was a genetic signature of a “metabolic syndrome” that may influence immune defense, we performed principal component analysis using the BLUPs extracted for each nutritional index. This analysis was implemented in R with the prcomp function with tol = 0.1 and unit variance scaling on. The principal component values for were then tested for correlation with bacterial load. This analysis was done for each diet individually.
The set of SNPs for mapping was described in Huang et al. (in revision) and consists of only SNPs with minor alleles present in at least four of the lines (MAF >2%; 2415518 total SNPs). For bacterial load (Ln CFU), we used SAS to run the following model:
LnCFU = m+SNPi+Dietj+SNPi*Dietj+Blockk(Dietj)+Wolbl+Infectorm+Platern+Lineo(SNPi)+eijklmno, where all factors were fixed except Line(SNP). P-values for the main effect of SNP and the SNP*Diet interaction were obtained for each SNP. We also ran the model separately on data obtained from flies reared on each of the two diets to obtain significance values for each SNP on each diet independently. These models were LnCFU = m+SNPi+Blockj+Wolbk+Infectorl+Platerm+Linen(SNPi)+eijklmn. We considered SNPs that mapped with significance level of p < 10-6 to be nominal positive hits and candidates for RNAi knockdown experiments. This p-value corresponds to a false discovery rate of 5–10% depending on the precise analysis being performed.
To correct for gene size, we used GOWINDA [47] to test for the enrichment of particular functional groups. Here we relax our significance threshold to include all SNPs with p<10-5. This allows for more power through the inclusion of additional SNPs. Relaxing the P-value threshold even further had little effect on GO enrichment results. Significantly associated SNPs for each treatment (low glucose, high glucose, main effect) were used with a background SNP set consisting of all SNPs used in the GWAS. GO slim [48] terms were used to reduce redundancy in GO categories. GOWINDA was run using gene mode, including all SNPs within 1000bp of a gene, a minimum gene number of 5, and with 100,000 simulations. We report all GO terms with a nominal P-value less than 0.1.
For all SNPs with P-values meeting our significance threshold and falling within 1000 bp of an annotated gene, we performed the infection assay described above on RNAi knockdown lines from the Vienna Drosophila RNAi Center (Vienna, Austria), if available.
To test the effect of the gene on resistance to infection, we crossed each RNAi line to a line carrying the ubiquitous driver (Act5C-Gal4/Cyo or da-Gal4) and infected F1 offspring of the knockdown genotype. We compared the immune defense in these F1 offspring to that of F1 progeny from the driver line crossed to the background genetic line of the RNAi transformant. Unless otherwise indicated, we performed RNAi knockdown experiments using a standard diet (1:1 glucose to yeast ratio, but more calorie dense than our high and low glucose diets—see methods). For those SNPs that showed a diet-specific effects, we performed RNAi knockdown experiments on the experimental high and low glucose diets.
It is completely unknown what proportion of genes throughout the genome might yield an immune phenotype when expression is repressed. To test whether genes containing our significantly associated SNPs were more likely to have an immune phenotype than a set of arbitrary genes from the genome, we also performed RNAi knockdown experiments on the genes that were physically close to those of interest but not known to be involved in immunity and not essential for viability. We refer to these as “nearest neighbor controls”.
|
10.1371/journal.ppat.1000321 | Genetic Variation in OAS1 Is a Risk Factor for Initial Infection with West Nile Virus in Man | West Nile virus (WNV) is a re-emerging pathogen that can cause fatal encephalitis. In mice, susceptibility to WNV has been reported to result from a single point mutation in oas1b, which encodes 2′–5′ oligoadenylate synthetase 1b, a member of the type I interferon-regulated OAS gene family involved in viral RNA degradation. In man, the human ortholog of oas1b appears to be OAS1. The ‘A’ allele at SNP rs10774671 of OAS1 has previously been shown to alter splicing of OAS1 and to be associated with reduced OAS activity in PBMCs. Here we show that the frequency of this hypofunctional allele is increased in both symptomatic and asymptomatic WNV seroconverters (Caucasians from five US centers; total n = 501; OR = 1.6 [95% CI 1.2–2.0], P = 0.0002 in a recessive genetic model). We then directly tested the effect of this SNP on viral replication in a novel ex vivo model of WNV infection in primary human lymphoid tissue. Virus accumulation varied markedly among donors, and was highest for individuals homozygous for the ‘A’ allele (P<0.0001). Together, these data identify OAS1 SNP rs10774671 as a host genetic risk factor for initial infection with WNV in humans.
| When humans are exposed to infectious agents, the outcome may vary: some remain uninfected, some who become infected remain asymptomatic, and symptomatic individuals may develop clinical manifestations that vary in number and severity. Both host and environmental factors are thought to influence outcome. Here we show that a specific human mutation that results in reduced function in a known antiviral gene named OAS1 is associated with increased rate of infection with West Nile virus, a member of the flavivirus family. OAS1 codes for a component of the type I interferon signaling system; its homologue in mice has been shown previously to be important for flavivirus susceptibility. Individuals carrying two copies of the hypofunctional variant of OAS1 were found more frequently than expected among both asymptomatic and symptomatic WNV-positive individuals. To investigate further, we developed a novel model of WNV replication using human tonsil tissue in which high variability in replication was observed among donors. We found that the highest virus accumulation occurred in donors with two copies of the hypofunctional OAS1 variant. Together these data suggest that OAS1 activity may influence the probability of initial infection, but not the severity or symptomatic quality of infection, after WNV exposure in man.
| West Nile virus (WNV) is a re-emerging flavivirus transmitted by mosquitoes to several species of birds. Humans and several other mammalian species may be accidental dead-end hosts. First isolated in Uganda from a febrile woman in 1937 [1], WNV has caused sporadic outbreaks in the Middle East, Africa, Western Asia, Europe, and Australia. In the Western Hemisphere, it was first isolated from a patient during an outbreak of meningoencephalitis in New York in 1999 [2],[3],[4]. Since then it has rapidly spread across the United States into Canada and Central and South America, and has caused annual outbreaks of disease. Through November 18, 2008, there have been 28,906 US laboratory-confirmed symptomatic WNV-seropositive cases reported to the Centers for Disease Control and Prevention (CDC), with 1121 (3.9%) WNV-induced deaths (www.cdc.gov). In addition to human infections, the virus has caused significant morbidity and mortality in birds and horses.
In the US, seroprevalence is ∼3% in the general population and 20–30% of infected individuals have been estimated to become symptomatic [5]. Clinical manifestations include West Nile fever (WNF) and WNV-induced neuroinvasive disease (WNND), which manifests primarily as meningitis and encephalitis [6]. To date, no specific antiviral agents or vaccines have been approved by the FDA for human WNV infection, and treatment is supportive. Survivors of WNND may develop long-term neurologic sequelae.
Viral infections in man are typically controlled in part by early induction of type I interferon (IFN), which initiates a cascade of innate mechanisms that impair virtually all aspects of the viral life cycle. Genetic studies in mice have identified several type I IFN-regulated anti-viral effector pathways that control WNV replication, including the double-stranded RNA-dependent protein kinase (PKR) mechanism of protein synthesis inhibition, and the 2′–5′ oligoadenylate synthetase (OAS) pathway of RNaseL-mediated RNA degradation [7]. OAS enzymes catalyze the synthesis of 2′–5′-linked oligoadenylates (2–5A) from ATP, which can then bind and activate latent RNaseL resulting in the degradation of host and viral RNAs [8].
Targeted gene disruption in mice has revealed a critical role for the type I IFN Receptor, PKR and RNaseL for survival following WNV infection [7],[9]. A critical role for OAS emerged from genetic analysis of the mouse locus flv, which determines in an autosomal dominant manner susceptibility to WNV and other flaviviruses [10],[11]. Flv has recently been mapped by two groups working independently [12],[13] to a missense mutation in the 2′–5′ oligoadenylate synthetase 1b (oas1b) gene, which results in a truncated protein. How this mutation actually works has not yet been fully delineated. It clearly is not redundant with the multiple other oas1 genes found in the mouse genome [14]. However, there is evidence that protection may not actually be mediated through RNase L [15], and it is not yet known whether the oas1b gene product is enzymatically active [16]. The genetic data suggesting a role for oas1b in protection against WNV in mice are supported by in vitro experiments showing that WNV replication in cells expressing wild-type oas1b is less efficient than in cells expressing the truncated form [17],[18]. Furthermore, genetic knock-in of the resistant oas1b allele into a susceptible mouse strain resulted in resistance to Yellow Fever Virus, a related flavivirus [19]. WNV susceptibility has not been reported in this knock-in mouse. Taken together, the combined in vivo and in vitro data suggest an important role of oas1b in innate resistance to flavivirus infection in the mouse.
Thus, we hypothesized that polymorphisms in the human ortholog of oas1b, OAS1 [14], could influence outcome in humans exposed to WNV. Established risk factors for human WNV disease include age, immunosuppression, and genetic deficiency in the chemokine receptor/HIV-1 coreceptor CCR5, due to homozygous inheritance of the defective allele CCR5Δ32 [20],[21]. Despite the strong epidemiologic association with CCR5Δ32, only ∼4% of Caucasian symptomatic WNV-infected individuals can be accounted for by CCR5 deficiency, indicating that other genetic risk factors for WNV disease may exist. A small case-control study (n = 27 cases) failed to find evidence of association of severe WNV disease with two OAS1 SNPs [22]. Instead, the investigators found a strong association with a silent variant (rs3213545) in OAS-Like (OASL), a paralogue of OAS1 located approximately 8 megabases away that has not been found to encode a protein with OAS activity [8]. In this report, we attempted to validate the OASL association with WNV, but could not. Instead, we found a consistent association of an OAS1 SNP and infection with WNV from samples collected from five independent US centers, and with WNV replication in human lymphoid tissue ex vivo.
To investigate the role of the OAS system in controlling outcome after WNV exposure in man, we scanned the OAS1 gene for known polymorphisms that occur in this region. Unlike the mouse genome, where 8 copies of oas1 have been identified (oas1a-h), there is only one copy of OAS1 in the human genome, although alternative splicing gives rise to several isoforms [14]. All OAS1 splice variants include exons 1–5, but vary with regard to downstream exons, and thus produce proteins of various sizes, including the p42, p44, p46, p48, and p52 forms [14],[23]. We identified common SNPs in the OAS1 gene region in the CEPH cohort using data from the International HapMap Project (www.hapmap.org). Human OAS1 is one of three genes in the OAS cluster, ordered OAS1, OAS3 and OAS2 on chromosome 12 [24]. As shown in Figure 1, at least eleven SNPs in the OAS1 region have been found to cluster into two blocks of linkage disequilibrium (LD). The first block contains four SNPs, three of which are found within the first intron (rs7956880, rs10744785, and rs4766662) and one of which (rs2158390) is found in the 5′UTR. The second block contains seven SNPs, with rs3741981 in exon 3; rs2057778, rs2285934 and rs2285934 in intron 3; rs10774671 located at the last nucleotide in intron 5; rs1051042 and rs2660 in exon 7; and rs7135577 in the 3′UTR. The SNPs found in block 1 are not in coding regions. The six SNPs in block 2 excluding rs3741981 are in tight LD (Figure 1, the black region in block 2) and form two haplotypes that have been previously shown by Bonnevie-Nielsen and colleagues to be highly associated with the level of OAS enzymatic activity measured in peripheral blood mononuclear cells [23]. These investigators concluded that rs10774671 is most likely to be the functional SNP in this block because it sits at the last nucleotide of intron 5 in the OAS1 gene and serves as a splice acceptor site for exon 7. The G allele is predicted to allow splicing to occur resulting in the production of a p46 form with high enzymatic OAS activity. The A allele is predicted to prevent splicing at this site; instead, splicing occurs further downstream, resulting in two other forms, designated p48 and p52 associated with lower OAS enzymatic activity. Heterozygotes were reported to have intermediate OAS activity [23]. Although other splice variants of this gene exist, including p42 and p44, the splice variants that arise from the use of exon 7 (p46, p48, and p52) are controlled by rs10774671 [23],[24]. Thus, this polymorphism is a biologically plausible genetic probe to investigate the role of OAS1 in human WNV disease.
To test whether this SNP is associated with WNV infection, we obtained serum or plasma from five independent US sources of Caucasian WNV+ individuals (n = 501). We collected symptomatic WNV-seropositive patient samples (n = 331) from the following state public health departments and outbreak years: Arizona (2004), California (2005), Colorado, (2003) and Illinois (2005–6) as detailed in Table 1. These samples were from individuals who sought medical attention and where WNV was considered in the differential diagnosis because of compatible signs and symptoms (fever, meningitis, encephalitis) and confirmed by serologic testing. We also obtained asymptomatic WNV-seropositive blood donor samples (n = 170) from the American Red Cross (ARC) national blood supply screening program, identified as WNV nucleic acid amplification test (NAT) reactive and confirmed by serologic testing. These individuals remained symptom-free, according to self-assessment in a follow-up questionnaire, for at least 2 weeks post-donation [25],[26].
For comparison, we obtained two groups of US Caucasian control samples (n = 552). The first was comprised of healthy US Caucasian random blood donors (RBD) collected prior to the introduction of WNV into the US in 1999 (n = 360). The second consisted of healthy US blood donors collected by the ARC who were identified during routine blood screening as WNV false positives (initial reactivity by WNV NAT that could not be replicated and WNV-seronegative (n = 192)).
We limited the analysis to self-reporting Caucasian individuals because the allele frequency of the OAS1 rs10774671 SNP varies according to race (www.ncbi.nlm.nih.gov/projects/SNP/). In both cases and controls, the overall genotyping success rate was >96.0%. 156 patient samples were re-tested using a second genotyping method with 100% concordance. Within each group (WNV+ and control), the OAS1 genotypes were in Hardy-Weinberg equilibrium (HWE; p = 0.51 and p = 0.93, respectively). In our control samples (n = 552), the frequency of the OAS1 rs10774671 AA genotype shown in Table 1 (38.0%) was consistent with the observed AA genotypic frequency in other published Caucasian cohorts representing over 5000 individuals predominantly from Europe (36.8–41.1%) [27],[28],[29] as well as the published report in the CEPH cohort in the HapMap Project (39%, n = 59; www.hapmap.org).
A genotypic contingency analysis (3×2) for OAS1 rs10774671 revealed a statistically significant association when WNV-positive samples were compared to controls (p = 0.0008). In our analysis, we considered the dominant, recessive and additive genetic models (Table 2), and all were statistically significant (p<0.05). Using the recessive model when all cases were grouped together, the AA genotype was significantly greater in WNV-positive subjects than in controls (Odds Ratio (OR) = 1.6 [95% Confidence Interval (CI) 1.3–2.1, p = 0.0001; Table 2). A similar trend in OR was observed in each of the 5 centers when considered separately, ranging between 1.5 and 1.7. Analysis of the data using the additive model also showed a highly significant association (OR = 1.4 per allele [95% CI 1.2–1.7], p = 0.0003; Table 2) indicating that having more A alleles was significantly greater in WNV-positive individuals than in controls. The OR for each center when considered separately was less consistent in this model, spanning between 1.2 and 2.0. To determine which model fits the data best, we analyzed the predicted probability of WNV infection by copies of A allele under the three models (Figure 2A). Superimposed are the observed proportions of WNV-positive individuals by copies of A (square boxes) along with the 95% CI. By comparing the observed to predicted proportion of cases by number of alleles using chi-square goodness-of-fit tests, we determined that the recessive and additive models are approximately equally supported by the data, with the dominant model having a decidedly poorer fit. We also analyzed each genotype separately, using GG as the reference genotype, since the G allele is the ancestral allele (from published Pan troglodytes sequence data, www.ncbi.nlm.nih.gov). As shown in Figure 2B, the strongest association with WNV infection was observed when AA was compared to the GG genotype (OR = 1.8, 95% [CI] 1.2–2.7, p = 0.002). There was also a statistically significant increase between the AG and AA genotypes (OR = 1.5 95% [CI] 1.2–2.0, p = 0.002). The comparison between the GG and AG genotypes was not statistically significant (OR = 1.2, 95% [CI] 0.8–1.7, p = 0.38), suggesting a better fit for the recessive model. Taken together, these data show that the AA genotype of OAS1 at SNP 10774671 is significantly elevated in individuals seropositive for WNV.
We next asked whether the elevated frequency of the A allele in WNV-seropositive subjects varied according to clinical outcome. The AA genotypic frequency for the four groups of symptomatic WNV+ subjects ranged between 47.1–51.1% (average 49.5%, n = 331). The AA genotypic frequency for asymptomatic WNV+ subjects from the ARC fell within this range at 48.8% (n = 170). In contrast, the AA frequency in the control samples was 38.0% (n = 552) (Figure 3A). Analysis of symptomatic patient samples for whom clinical outcome data were available (n = 294) showed a similar distribution of OAS1 genotypes in both the WNF and WNND clinical subgroups (Table 3 and Figure 3A). With regard to death as an outcome, the sample size is underpowered for analysis: only 13 of all 294 cases for which clinical outcome data were available experienced fatal WNV infection. These data are most consistent with the hypothesis that OAS1 at SNP 10774671 is a risk factor for seroconversion after exposure to WNV, but not for specific clinical manifestations.
We previously demonstrated that CCR5Δ32 homozygosity is a strong risk factor for symptomatic WNV disease [20],[21]. To investigate whether the association of OAS1 SNP 10774671 with WNV infection is independent of CCR5Δ32, we analyzed the distribution of OAS1 in WNV+ subjects from the four groups of symptomatic US Caucasians who were CCR5+/+ (n = 261), CCR5+/Δ32 heterozygotes (n = 53), and CCR5Δ32/Δ32 homozygotes (n = 17). The expected frequencies were calculated solely on OAS1 HWE considerations, since the increase in CCR5Δ32 homozygotes resulted in skewing of HWE for this allele, as previously reported [20],[21]. The distribution of OAS1 genotypes according to CCR5 genotype did not differ significantly from expectation as shown in Table S1 (χ2 = 5.2, p = 0.26) suggesting that these alleles assort independently.
Since age is also an accepted risk factor for symptomatic WNV infection, we determined OAS1 SNP 10774671 genotypes in all symptomatic WNV+ subjects stratified by age; <45 years old (n = 99), between 45 and 64 years old (n = 131), and >64 years old (n = 101) as shown in Table S2, but found no significant differences.
We next attempted to validate the association previously identified by Yakub et al. between OASL SNP rs3213545 and WNV infection [22]. Contingency analysis (3×2) did not reveal any significant association (p = 0.93) when the WNV-infected seropositive subjects (n = 501) were compared to WNV-seronegative controls (n = 552). Analysis of the additive model (OR = 1.0 [95%CI 0.8–1.3], p = 0.94), the recessive model (OR = 1.0, [95%CI 0.69–1.6], p = 0.83), and the dominant model (OR = 1.0, [95%CI 0.8–1.3], p = 0.82) revealed no association with WNV infection (Table S3). When samples collected from each of the 5 centers were analyzed separately, no significant association was found (Table S3). The frequency of TT homozygosity was not elevated above controls (Figure 3B) and did not vary between asymptomatic donors and symptomatic patients. Since the previous study by Yakub et al. showed an association between this SNP and severe WNV disease, we also looked for an association according to clinical outcome. As shown in Table 3 and Figure 3B, no association was found with either WNF or WNND, the more severe CDC clinical definition.
To directly assess the role of OAS1 during active WNV replication in human cells, we developed an experimental ex vivo model of WNV infection using explants of human tonsil tissue. This model may mimic the early events after infection that may occur in vivo in the draining lymph node, shown in mice to be an early site of viral replication that precedes CNS spread [30]. Since infection in primary human cells in vitro typically requires exogenous stimulation or activation [31],[32], this system is more natural in that the architecture and composition of the primary lymphoid tissue are preserved and no exogenous stimuli are added. Explanted tissue from patients undergoing tonsillectomies was sectioned and inoculated with WNV. For each experimental condition, 27 blocks of tissue were sectioned per donor to control for variation in the number of susceptible cells that may occur between individual slices of tissue from the same donor. Supernatant from the cultured tissue was tested every third day for virus using a focus forming unit (FFU) assay. As shown in Figure 4A, virus was detected in the culture supernatant at the earliest time point (day 3) and increased throughout the 12 day culture for each of the 21 donors tested. WNV-infected cells were visible by direct immunostaining of the tissues (Figure 4B). Tonsils from all donors supported WNV replication to high levels that varied widely among donors, particularly at early time points post-infection (e.g. greater than 4 logs variability at day 3). This model may be useful not only for investigating WNV pathogenesis but also for testing the efficacy of potential WNV therapeutics.
The variation in virus production we observed among donors provided an ideal opportunity to test the dependence of viral replication on OAS1 genotype in intact human tissue, assuming that OAS1 mRNA was expressed in the tissue. Twelve of the 21 tissues tested for WNV replication were available for RNA extraction to investigate this. First we examined the effect of WNV infection on expression of all OAS1 transcripts (Figure 5A), using primers located between exons 2 and 3. OAS1 mRNA was detected in uninfected tissue from all 12 donors (data not shown), and expression of total OAS1 mRNA was induced ∼2 fold by WNV compared with donor-matched uninfected control tissue at day 3 post-infection (p = 0.01; Figure 5A). This increased to ∼4.6 fold induction for both day 6 (p = 0.002) and day 12 (p<0.0001) after infection. Since OAS1 is an IFN-stimulated gene, we also measured IFNβ1 in these samples and observed a significant increase above control at day 6 (p = 0.002) and day 12 (p = 0.007) post infection (Figure 5B). These 12 samples were also tested by RT-PCR for OAS1 rs10774671 A and G allele-specific transcripts using a previously established method [23]. Both A and G allele-specific transcripts (319 and 417 bp amplicons for A allele; 416 bp amplicon for G allele) were detected in appropriate uninfected tissue from all 12 donors (Figure 5C). As shown in Figure 5D, total OAS1 SNP rs10774671-specific mRNA modestly increased in a time-dependent manner after infection. This was due at least in part to induction of OAS1 SNP rs10774671 A allele-specific RNA, as shown by analysis of tissue from AA homozygotes (n = 5; Figure 5E). Since RNA generated from the A and G alleles differ by only 1 nucleotide, we were unable to quantitate the relative induction of A and G allele-specific RNA in WNV-infected tissue from AG heterozygotes. We observed ∼1.5-fold induction of G allele-specific mRNA in infected tissue from the one GG homozygote that was available. Thus both OAS1 SNP rs10774671 variants of interest are present at the RNA level throughout the course of WNV infection of human tonsil tissue, allowing us to ask whether virus replication in this system is OAS1 genotype-dependent.
To test whether OAS1 variation affects WNV infection in human lymphoid tissue ex vivo, we genotyped patient samples for the OAS1 rs10774671 SNP. Of the 21 donors tested, 7 were homozygous for the A allele (33.3%), 2 were homozygous for the G allele (9.5%), and the remaining 12 were AG heterozygotes (57.2%). This distribution is similar to the genotypic distribution observed in controls. When viral production was analyzed as a function of OAS1 genotype, we found that tissues from individuals homozygous for the A allele supported higher levels than tissues from patients with AG or GG genotypes (Figure 6). The most significant difference was observed 3 days post infection (p<0.0001) with a mean viral titer of 3.9×104 ffu in tissues from donors with the AA genotype versus those from individuals who were either AG or GG (1.1×103 ffu). At day 6, the mean viral titer was still greater in AA than in non-AA tissues (1.7×106 versus 5.0×104 ffu, respectively; p = 0.002). At day 9, the differences in viral titer had decreased (1.3×106 ffu for AA versus 3.8×105 ffu for non-AA; p = 0.02). By day 12 post-infection, no significant difference was observed (p = 0.80). Thus the allele associated with higher OAS enzymatic activity (G) is associated with lower viral production for at least 75% of the period of observation after infection in lymphoid tissue.
The present study provides the first evidence in support of the hypothesis that OAS1 influences risk of infection in response to WNV exposure in humans. This is based on association of the OAS1 ‘A’ allele at SNP rs10774671, which has previously been associated with low OAS enzymatic activity, with seropositivity for WNV using recessive and additive genetic models on patient samples collected from five US centers. We found that the AA genotype was elevated in WNV-positive individuals regardless of whether these patients remained asymptomatic or developed symptomatic disease, including WNF and WNND. We also observed that homozygosity for the A allele correlated with increased levels of WNV replication in primary lymphoid tissue from human donors. Moreover, the results are consistent with the previously reported identification of the flavivirus resistance locus flv in mice as a truncation mutation in oas1b [12],[13], the ortholog of human OAS1 [14]. Our data do not confirm the previously published association of OASL and WNV infection in man. Together, these data suggest that OAS1 activity modulates early viral replication, and that decreased OAS1 activity is associated with increased risk of infection and seroconversion. These data suggest that carriers of the G allele, which is associated with high OAS1 activity, may effectively control WNV replication early after infection in humans through innate mechanisms. Furthermore, since oas1b in the mouse has been implicated in resistance not only to WNV, but to several other flaviviruses, this association may also be relevant for other medically important neurotropic flaviviruses, such as Japanese encephalitis virus, Yellow Fever virus, and Tick-borne encephalitis virus.
There are several limitations to our study. First, like most gene association studies, the analysis is retrospective. Second, due to a limitation of sample availability the association of WNV infection with OAS1 SNP 10774671 was determined only in North American Caucasians. Third, the epidemiologic data, while consistent with a role for OAS1 in controlling WNV infection, do not establish causality versus linkage disequilibrium with an as yet unidentified causal variant. Importantly though, the variant tested has been previously associated with OAS function, and is therefore biologically plausible [23].
Additional work will be needed to delineate the precise mechanisms by which OAS1 SNP rs10774671 variation may affect WNV infection. We were able to detect the OAS1 A and G allele transcripts in uninfected human tonsil tissue as predicted by donor genotype, and we found that WNV infection was able to induce total OAS1 RNA in the tissue as well as RNA specific for the OAS1 A and G alleles. However, additional work will be needed to determine OAS1 allele-specific activity in tonsil tissue as a function of WNV infection, and the relative contribution of each of the OAS1 isoforms present in the tissue. While OAS activity in human PBMCs has been reported to vary in an OAS1 SNP rs10774671-dependent manner, with the rank order GG>AG>AA, OAS enzymatic activity of purified protein encoded by OAS1 A and G allele variants has not yet been determined [23]. Furthermore, the mechanism of action of oas1b in mice is not yet clear, nor is there evidence that oas1b is enzymatically active [16] or that it functions through activation of RNaseL [15].
Population stratification as a cause of skewed genotypic frequency is a concern in all gene association studies. Given the limited availability of genomic DNA from the small samples of serum available to us from WNV+ subjects, we were unable to address this issue directly. However, we believe this is unlikely to confound our results since the five WNV+ groups were from geographically distinct US populations yet had similar AA genotypic frequencies ranging from 47.1% to 51.4% with an average of 49.3%, whereas the two US WNV-negative control populations that we genotyped, which also were from distinct geographic regions, had much lower AA genotypic frequencies (36.7 and 40.6%) that both fell within a narrow range reported in previously published studies, (36.8–41.1%) [27],[28],[29]. In particular, Smyth et al reported an OAS1 SNP rs10774671 AA genotypic frequency of 41.1% for 4,735 subjects from the 1958 British Birth Cohort (www.cls.ioe.ac.uk/studies.asp?section=000100020003)[29]; Fedetz et al reported a frequency of 36.8% for 424 Caucasians from Spain [27]; and the International HapMap Project reported a frequency of 39% in US Caucasians (n = 59; www.hapmap.org). It is important to note that a fourth published study, by Field et al., reported an AA genotypic frequency of 49.9% [28]. This anomaly is most likely due to the use of a non-random population, namely the healthy siblings of diabetics. In particular, the A allele frequency in non-transmitted alleles of the parents, which would be expected to have nothing to do with disease, was 0.61 in this study, which would give an expected frequency of 0.61×0.61 = 37% for the AA genotype (L. Field, personal communication), compatible with the frequency found in all five other control populations. Of note, no difference in frequency was observed between cases and controls in a genetically unlinked but related variant in OASL (www.hapmap.org), where the odds ratio for all genetic models tested was 1.0, p≥0.82.
Animal studies have shown conclusively that the type I IFN system exerts potent antiviral effects critical for survival from viral infections, including WNV infection [9],[30],[33],[34]. However, there is limited evidence in humans supporting a role for endogenous type I IFN signaling at the level of viral pathogenesis. Our study suggests a potentially important role of this type I IFN-inducible pathway at the level of initial infection and containment of virus during the early phases of natural human infection. Our data show that WNV infection is less likely to result in seroconversion in individuals with the GG genotype (higher OAS enzymatic activity), suggesting that individuals with the AA genotype (lower enzymatic activity) are less likely to control initial WNV infection with innate immunity and initiate an adaptive immune response, signaled by seroconversion.
Genetic variation in OAS1 has been previously associated with other diseases, including Type I Diabetes, SARS, and Hepatitis C. The associations in Type I Diabetes and SARS have been inconsistent in different studies [28],[29],[35],[36]. For Hepatitis C, a relative of WNV also in the flavivirus family, a polymorphism in the 3′-untranslated region of OAS1 has been associated with persistent infection [37]. This is particularly relevant since this polymorphism is in strong linkage disequilibrium with polymorphism rs10774671 which we have associated with WNV seroconversion [23].
A clinical implication of our results is that pharmacologic induction of OAS1, which has been shown to occur in response to IFNα treatment [38],[39] could be efficacious in the context of preventing WNV infection in man. While treatment with IFNα has been reported to be effective in WNV-infected patients, the world experience is limited to a small number of scattered case reports, and treatment failure has also been documented [40],[41],[42]. Our data support continued clinical research of IFNα as a therapeutic option in WNV disease. Finally, our data indicate that the great majority of risk of symptomatic WNV disease remains unexplained at the genetic level, and support continued research into variation in the type I IFN system as a factor contributing to the heterogeneity of outcome in this disease.
The study was approved by the Office of Human Subjects Research of the NIH. Samples of symptomatic WNV-seropositive patient cases from four US states (AZ, CO, CA, and IL) were collected as previously described [20],[21]. Since the allele frequency of this SNP varies by race (www.hapmap.org), and since information about race was not available for all subjects in the WNV-seropositive patient cases, we analyzed only self-reporting Caucasians (total n = 331 in the 4 US states combined). Symptomatic WNV-seropositive patients are defined as individuals who came to clinical attention with symptoms consistent with WNV disease (primarily WNF, WNND), where WNV was confirmed using serological tests. Of the WNV-seropositive samples received, sufficient DNA from self-reported Caucasian patients was available for analysis of 345 samples. Genotypes for both OAS1 and OASL were obtained for 331 symptomatic WNV-seropositive subjects with an overall genotyping success rate of 95.9%. Serum samples were collected from the following states and years: 1) Arizona from the 2004 epidemic (n = 135); 2) Colorado from the 2003 epidemic (n = 72); 3) California from the 2005 epidemic (n = 87); 4) Illinois from the 2005/2006 epidemic (n = 37). Plasma samples from asymptomatic WNV-seropositive blood donors from the American Red Cross (ARC WNV+) were collected between 2003–2008. Asymptomatic WNV-seropositive blood donors (n = 170) are defined as Caucasian random blood donors who tested reactive for WNV nucleic acid twice and were WNV IgM seropositive who remained asymptomatic for at least 2 weeks post donation as assessed by follow-up questionnaire as described previously [25].
Two healthy US Caucasian control groups were established: 1) healthy unrelated US Caucasian random blood donors (RBD) from the NIH Department of Transfusion Medicine (n = 360) collected under an IRB-approved protocol [20]; and 2) healthy Caucasian blood donors from the American Red Cross (n = 192) who tested WNV nucleic acid reactive upon initial screen at the time of blood donation, but were negative for WNV nucleic acid upon retesting and were also WNV-seronegative for IgM (false positives). The following information was provided if available: age, gender, self-reported racial group, date of sample collection, and CDC-defined clinical presentation at the time of sample collection: WNF, WNND, and death. For all study samples, investigators were blinded to unique patient identifiers.
200 µl of serum collected from Arizona, Colorado, California, and Illinois were thawed for genomic DNA purification using the QiaAmp 96 DNA Blood Kit according to the manufacturer's instructions (Qiagen, Valencia, CA). 250 µl of plasma collected from test subjects by the American Red Cross was purified using NucliSENS EasyMAG automated nucleic acid extraction technology (Biomerieux, Inc). Purified DNA was eluted into 100 µl of the recommended buffer and stored at 4°C until further use. DNA from NIH random blood donors was isolated from peripheral blood leukocytes as previously described [43]. SNPs were genotyped using the ABI 7900HT PCR System with TaqMan primer/probe mix predesigned and validated by ABI (C___2567433_10 and C__11753831_1_). PCR-RFLP was used as a second genotyping method for OAS1 SNP rs10774671 as previously described [27].
Contingency tables were utilized to calculate genotypic and allelic frequency differences between cases and controls by comparing numbers of expected (Hardy-Weinberg equation) and observed individuals using chi-square tests of significance to obtain a two-sided p value using either 1 (2×2 table) or 2 (3×2 table) degrees of freedom. ORs were calculated using a recessive genetic model (i.e. AA verses AG plus GG for OAS1 and TT versus TC plus CC for OASL) or the dominant genetic model (i.e. AA plus AG versus GG for OAS1 and TT plus TC versus CC for OASL) by cross tabulation and 95% CI were estimated using the approximation of Woolf (GraphPad Software version 4.0b, San Diego, CA). The additive model OR values and 95% CI were calculated using JMP software. Chi-square was used unless otherwise indicated. ORs and CIs based on the additive model were estimated using JMP (SAS, version 7). Chi-square goodness-of-fit statistics were used to compare dominant, recessive, and additive models. The two-sided unpaired t-test was used to calculate statistical significance of viral replication after log10 transformation of the data for each time point and a two-sided paired t-test was used to calculate significance of gene expression analysis. P values were considered significant if <0.05 and the reported values have not been corrected for multiple comparisons. Linkage disequilibrium of OAS1 SNPs and resultant haplotypes were examined using Haploview 4.0 (available at www.broad.mit.edu/mpg/haploview/index) using previously obtained data from the International HapMap project (www.hapmap.org) on the CEU panel using release 21 data.
Human tonsils from routine tonsillectomy performed at the Children's Hospital National Medical Center in Washington, DC were collected under an Institutional Review Board-approved protocol from NICHD within a few hours after surgery. The mean age of the children used in this study was 5±3. Tonsils were dissected in blocks of approximately 2 mm. 9 tissue blocks per well in triplicate wells per condition were placed on top of collagen sponge gels floating in six-well plates as previously described [44]. All 27 tissue blocks from each donor were then individually inoculated with 5 µl containing 500 FFU WNV strain NY99, or left uninfected. We assessed productive WNV infection by measuring FFU as described above in culture medium pooled from 27 blocks during the previous 3 days between successive media changes. FFU at each time point represent the cumulative total of infectious virions produced. Coefficient of variation for each donor and each time point was calculated to be 13.3%. All tonsil donors were analyzed for the CCR5Δ32 mutation as previously described [20] and found to be wild type at this allele.
Confluent Vero cells were grown in a 12 well plate in OptiPro SFM (Invitrogen) with 2% FBS (Hyclone) and 50 µg/ml gentamicin (ATCC). Supernatants from tonsil cultures were diluted and incubated on the cells for 1 hour at 37°C prior to overlaying with 2 ml Opti-MEM (Invitrogen) with 8 g/L methylcellulose (Sigma), 2% FBS and 50 µg/ml gentamicin. Plates were incubated for 2 days at 37°C, then washed 3 times with PBS. 500 µl of diluted anti-WNV anti-sera/well (HMAF, ATCC #VR-82) was incubated for 1 hour at 37°C. Plates were then washed 3 times and 500 µl of diluted (1∶10) anti-mouse/anti-rabbit HRP labeled polymer (DAKO Cytomation) was added. Cells were incubated for 1 hour at 37°C, washed 3 times and focus forming units (FFU) of WNV were visualized by addition of 1 ml diaminobenzidine (DAB) mixture (4.5 mg DAB (Sigma) /10 ml PBS+4.5 µl 30% H202 /10 ml). Viral titers are expressed as FFU/ml.
RNA was extracted from lymphoid tissue using the RNeasy Tissue kit according to manufacturer's protocol (Qiagen). Reverse transcription was performed using Superscript III first strand synthesis Supermix (Invitrogen) with random hexamers. OAS1 (Hs00242943_m1), IFNβ1 (Hs00277188_s1) and GAPDH (Hs00266705_g1) primer/probe sets were obtained from ABI and cycled on an ABI 7900HT PCR System. Each sample was normalized to GAPDH and fold change in infected samples were compared to uninfected donor-matched samples.
4–5 µm sections of paraffin-embedded slides were stained with anti-WNV hyperimmune mouse ascites fluid (HMAF, ATCC #VR-82) diluted 1∶100 in antibody diluent, background reducing agent (DAKO Cytomation). After incubation with anti-mouse polymer-horse radish peroxidase (DAKO Cytomation), slides were developed with streptavidin and diaminobenzidine liquid (DAKO Cytomation) and counterstained with hematoxylin.
Tissues from 12 tonsil donor samples were tested for OAS1 SNP-specific mRNA. cDNA from uninfected and infected donor-matched samples for each time point amplified by PCR as previously described [23]. Briefly, primers were designed to flank the splice acceptor site (F-ggcggaccctacaggaaact; R-acaccagctcactgaggagc). The presence of a G allele resulted in a 416 bp amplicon, while the presence of the A allele resulted in 319 bp and 417 bp fragments. PCR products were resolved on a 2% agarose gel, visualized using ethidium bromide, and quantitated by densitometry normalized to β-actin using Image J software version 1.38×. Fold change was calculated by comparing normalized WNV-infected measurements to normalized uninfected measurements for each time point for each donor.
|
10.1371/journal.pgen.1003031 | A Targeted Glycan-Related Gene Screen Reveals Heparan Sulfate Proteoglycan Sulfation Regulates WNT and BMP Trans-Synaptic Signaling | A Drosophila transgenic RNAi screen targeting the glycan genome, including all N/O/GAG-glycan biosynthesis/modification enzymes and glycan-binding lectins, was conducted to discover novel glycan functions in synaptogenesis. As proof-of-product, we characterized functionally paired heparan sulfate (HS) 6-O-sulfotransferase (hs6st) and sulfatase (sulf1), which bidirectionally control HS proteoglycan (HSPG) sulfation. RNAi knockdown of hs6st and sulf1 causes opposite effects on functional synapse development, with decreased (hs6st) and increased (sulf1) neurotransmission strength confirmed in null mutants. HSPG co-receptors for WNT and BMP intercellular signaling, Dally-like Protein and Syndecan, are differentially misregulated in the synaptomatrix of these mutants. Consistently, hs6st and sulf1 nulls differentially elevate both WNT (Wingless; Wg) and BMP (Glass Bottom Boat; Gbb) ligand abundance in the synaptomatrix. Anterograde Wg signaling via Wg receptor dFrizzled2 C-terminus nuclear import and retrograde Gbb signaling via synaptic MAD phosphorylation and nuclear import are differentially activated in hs6st and sulf1 mutants. Consequently, transcriptional control of presynaptic glutamate release machinery and postsynaptic glutamate receptors is bidirectionally altered in hs6st and sulf1 mutants, explaining the bidirectional change in synaptic functional strength. Genetic correction of the altered WNT/BMP signaling restores normal synaptic development in both mutant conditions, proving that altered trans-synaptic signaling causes functional differentiation defects.
| Glycans are sugar additions to proteins. Surrounding all eukaryotic cells, secreted and membrane glycans form a glycocalyx that regulates cell–cell signaling. However, the mechanisms controlling glycan-dependent intercellular communication are largely unknown. In the nervous system, glycans play important roles in the development and regulation of synapses mediating intercellular communication. The Drosophila neuromuscular junction serves as a genetically tractable synapse in which expression of glycan-related genes can be systematically knocked down to investigate effects on synaptic morphology and function. This study employs a transgenic RNAi screen to characterize the synaptic requirements of 130 glycan-related genes. From this screen, two functionally paired genes (hs6st and sulf1) that add or remove a sulfate at the 6-O position on heparan sulfate proteoglycans (HSPGs) were identified as being critically important for synaptic functional development. Removal of each gene produces an opposite effect on neurotransmission strength, weakening and strengthening communication, respectively. This mechanism controls the synaptic expression of two HSPGs, which act as co-receptors to control the abundance of anterograde WNT and retrograde BMP signals, which drive intracellular signal transduction pathways regulating gene transcription to control synaptic functional development. This screen serves as a platform for systematic investigation of glycan mechanisms regulating synaptic development.
| Glycans coat cell surfaces, and glycosylation decorates secreted molecules of the pericellular space and extracellular matrix (ECM) [1], [2]. It is well known that glycan modifications mediate critical functions of intercellular signaling and regulate interactions of numerous growth factors with the ECM [3], [4]. The synthesis, modification and degradation of glycoconjugates, including O/N-linked glycoproteins, glycosaminoglycan (GAG) proteoglycans and glycan-binding lectins, is controlled by a dedicated cadre of genes [5], [6]. In the nervous system, these glycan-related genes play key roles in development, including neuron fate specification, migration, formation of axon tracts and synapse maturation [7]. At synapses, glycosylated ECM molecules, membrane receptors and outer-leaflet glycolipids together form the highly specialized synaptomatrix interface [4], [8], which interacts with trans-synaptic signals to modulate synaptogenesis [9].
A prime example is the classic Agrin proteoglycan, which bears heparan sulfate (HS) chains, O/N-linked glycans and also a glycan-binding lectin domain that binds other glycoconjugates [10], [11], [12]. Reduction of GAG sulfation perturbs the Agrin signaling that drives postsynaptic acetylcholine receptor (AChR) cluster maintenance at the neuromuscular synapse [13]. Likewise, Galbeta1,4GlcNAc and Galbeta1,3GalNAc glycans inhibit Agrin signaling by suppressing muscle specific kinase (MuSK) autophosphorylation, a key step during synaptogenesis [14]. Analogous glycan-dependent mechanisms at the Drosophila neuromuscular synapse involve the secreted Mind-the-Gap (Mtg) lectin, which assembles the glycosylated synaptomatrix between presynaptic active zone and postsynaptic glutamate receptor (GluR) domains [15]. This glycan mechanism induces GluR clustering, synaptic localization of integrin ECM receptors, and shapes trans-synaptic signaling by controlling ligand/receptor abundance [16], [17], [18]. Thus, many long-term studies in vertebrate and invertebrate genetic models suggest that glycan mechanisms are a core foundation of synapse development.
In the current study, we conducted a broad transgenic RNA interference (RNAi) screen of synaptic glycan function, assaying requirements in both structural and functional development of the Drosophila neuromuscular junction (NMJ). We tested 130 genes from 8 functional categories: N-glycan, O-glycan and GAG biosynthesis; glycosyltransferases and glycan modifying/degrading enzymes; glycoprotein and proteoglycan core proteins; sugar transporters and glycan-binding lectins. We found that RNAi-knockdown of genes in all eight categories affects synaptic morphological development, with gene-specific effects on branching, bouton differentiation and synapse area. Likewise, all eight categories regulate synaptic functional development, with gene-specific effects both weakening and strengthening neurotransmission. Interestingly, only a few genes affect both structure and function, suggesting separable roles for glycans in regulating these synaptogenic pathways. The results of this genomic transgenic screen are presented as a platform from which to pursue systematic investigation of glycan mechanisms in synaptic development.
Two genes were selected for screen validation and mechanistic characterization; functionally-paired HS 6-O-endosulfatase (sulf1) and HS 6-O-sulfotransferase (hs6st). RNAi knockdown and null mutants identically alter synaptic functional development in a bidirectional manner; loss of sulf1 elevates neurotransmission strength, whereas loss of hs6st weakens it. Heparan sulfate proteoglycan (HSPG) targets Dally-like Protein (Dlp) and Syndecan (Sdc) [19], [20] are mislocalized in sulf1 and hs6st null synapses. In other developmental contexts, the sulfation state of these HSPG co-receptors strongly regulates WNT and BMP intercellular signaling [20], [21], [22]. At Drosophila synapses, WNT (Wg) is a key anterograde [23], [24] and BMP (Gbb) a key retrograde [25], [26] trans-synaptic signal. Consistently, loss of sulf1 and hs6st differentially changes synaptomatrix levels of Wg and Gbb, and downstream signaling into muscle and motor neuron nuclei, respectively. Glutamate release and receptor machinery is thereby bidirectionally altered in the two nulls. Genetic restoration of Wg/Gbb signaling to control levels restores the bidirectional changes in synaptic functional strength and pre-/post- synaptic differentiation in both sulf1 and hs6st nulls. We conclude that extracellular HSPG sulfation state in the synaptomatrix is a point of intersection between WNT/BMP trans-synaptic signaling pathways that drive functional development of the neuromuscular synapse.
Synaptic glycans play important roles as ligands, modulators and co-receptors regulating cell-matrix and intercellular communication [3], [27], [28]. Differential glycan distribution on pre- and postsynaptic surfaces, and in the cleft, of numerous protein classes, strongly suggests that glycan mechanisms mediate synaptic structural and functional development [29], [30], [31]. To test the genomic scope of this requirement, we used confocal imaging and electrophysiological recording at the well-characterized Drosophila glutamatergic neuromuscular junction (NMJ) [32], [33], [34] to screen the Vienna Drosophila RNAi Center (VDRC) library of glycan-related genes [35]. We induced UAS-RNAi knockdown using the ubiquitous UH1-GAL4 driver [15], [36]. We assayed morphological defects by co-labeling for pre- and postsynaptic markers, and assayed functional defects with two-electrode voltage clamp (TEVC) recording of neurotransmission strength. A summary of the screen results is shown in Figure 1. Full numerical results of the screen are shown in Table S1.
Candidate glycan-related genes were identified and classified into eight functional categories using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database [37] (Figure 1). Additional genes were added to the screen based on ortholog identification using the Information Hyperlinked over Proteins (iHOP) database [38]. The candidate gene list was expanded and verified using Flybase [39]. From this list, genes were cross-referenced with available VDRC UAS-RNAi transgenic lines to generate a final candidate list containing 130 genes within eight functionally-defined categories (Figure 1): N-glycan, O-glycan and glycosaminoglycan (GAG) biosynthesis; glycan core proteins (HSPG core proteins/glycoproteins); sugar transporters; glycosyltranferases; glycan modification genes (modification and degradation of glycans); and glycan-binding lectins. On genetic knockdown, 103 lines were viable until the wandering 3rd instar, whereas 27 lines showed developmental lethality at embryonic and early larval stages of development. From the 103 genetic lines characterized by confocal microscopy and TEVC electrophysiology in the 3rd instar (Figure 1), 21 exhibited pupal stage developmental lethality. Interestingly, >50% of pupal lethal lines displayed statistically significant defects in NMJ synaptic morphology and function.
For all 103 larval-viable lines, synapse morphology and function was quantified at the wandering 3rd instar NMJ (Figure 1; Table S1). Each UAS-RNAi line driven by UH1-GAL4 in the w1118 background was compared to the genetic control of w1118 crossed to UH1-GAL4 (UH1-GAL4×w1118) [35]. All morphological and functional assays were done blind to genotype, with values reported as fold-change compared to genetic control, as well as statistical significance calculated using one-way ANOVA analyses (see color scheme; P<0.05 (*), P<0.01 (**); Figure 1). The data represents ≥6 NMJs from ≥3 animals from every genotype. Synapse morphology was imaged by co-labeling with presynaptic marker anti-horse radish peroxidase (HRP) and postsynaptic marker anti-Discs Large (DLG). A synaptic bouton was defined as a varicosity of ≥2 µm in minimum diameter labeled by both HRP and DLG, and a synaptic branch was defined as a process containing at least two boutons [40]. NMJ branch number was the least affected morphological parameter, with only 2 of 103 genes showing a statistically significant change (Figure 1). Many more genes were involved in bouton development. All 27 genes showing a statistically significant change compared to genetic control exhibited elevated bouton numbers (Figure 1), suggesting that glycan mechanisms primarily limit morphological growth. Synapse area was determined by outlining the terminal area labeled by DLG using the thresholding function in ImageJ. The majority of gene knockdown conditions showed a decrease in NMJ area compared to control (Figure 1). 7 RNAi lines exhibited a statistically significant decrease in area, whereas only 2 lines exhibited a statistically significant increase in synaptic area. All raw values of measured morphological parameters are included in Table S1.
To assay functional differentiation, the motor nerve was stimulated with a suction electrode while the evoked excitatory junctional current (EJC) was recorded in the muscle (Figure 1) [41]. Nerve stimulation was applied at 4 V for 0.5 ms at a frequency of 0.2 Hz, with the muscle clamped at −60 mV. EJC amplitudes were calculated from recorded traces in the ubiquitously-driven RNAi lines (w1118 background) compared to the w1118; UH1-GAL4/+ control. Recordings were obtained from ≥3 independent trials for each RNAi knockdown condition. All electrophysiological screening was done blind to genotype, with values reported as fold-change and statistical significance calculated by one-way ANOVA analyses (see color scheme; P<0.05 (*), P<0.01 (**); Figure 1). Genes from all eight glycan classes were identified to produce changes in neurotransmission strength upon genetic knockdown. For the 103 larval-viable lines tested, 26 lines showed a trend towards increased transmission strength, and 12 were statistically elevated compared to genetic control (Figure 1). 4 gene knockdowns showed a trend towards decreased transmission strength, of which only 1 line reached statistical significance. 73 of the 103 lines tested showed no change in functional strength (Figure 1). Interestingly, only 6 RNAi lines showed statistically significant effects on both NMJ morphology parameters and EJC amplitude: CG1597, CG6657, CG7480, CG4451, CG6725 and CG11874 (Figure 1). This suggests that glycan effects on synapse morphological and functional development are largely separable. All raw values of EJC measurements are included in Table S1.
To validate results, a secondary screen was conducted using independent RNAi lines obtained from the VDRC and Harvard TRiP collections (Table S2). Of the 44 genes that showed morphological and functional defects in the primary screen, 33 were retested using independent RNAi lines, with the others lacking available secondary lines from any source. Using the same screen of morphological and functional characterization, we determined that ∼80% of retested secondary lines showed the reported structural (bouton number) and functional (EJC) phenotypes consistent with primary screen (Table S2). These primary and secondary RNAi screen results now represent a resource for the systematic characterization of glycan mechanisms underlying synaptic structural and functional development. Screen results were further studied by comparing synaptogenesis phenotypes of RNAi knockdown with defined genetic nulls for two genes, CG6725 and CG4451, from the glycosaminoglycan biosynthesis class (Figure 1). The RNAi screen of functional strength as measured by EJC amplitudes indicated opposite effects for these two lines, with CG6725 (RNAi-sulf1) knockdown exhibiting an increase in transmission strength and CG4451 (RNAi-hs6st) knockdown producing a decrease (Figure 1). Along with our goal to identify interesting glycan-related genes involved in synapse development, we show here characterization of null alleles of two genes obtained from screen results and define the associated mechanisms driving the bidirectional regulation of synaptic functional development.
The RNAi screen identified two functionally-paired genes, sulf1 (CG6725) and hs6st (CG4451), with similar effects on morphological development but opposite effects on synaptic functional differentiation (Figure 1). Our goal was to use these genes as a test case from the completed glycan screen, by assaying phenotypes in recently characterized null mutants of both genes [42], [43]. The gene products Sulfated (Sulf1), an HS 6-endosulfatase, and Hs6st, an HS 6-O-sulfotransferase, drive opposing changes in sulfation state of the same C6 carbon of the repeated glucosamine unit in GAG modified heparan sulfate proteoglycans [43], [44]. Viable null mutants are available for both genes, e.g. sulf1 (sulf1Δ1) and hs6st (hs6std770) [42], [43], but requirements have never been assayed in the nervous system or neuromusculature. We therefore first compared phenotypes of RNAi knockdown and null alleles at the NMJ synapse by confocal imaging of synaptic morphogenesis and TEVC recording of synaptic functional neurotransmission.
Using double-labeling for HRP (presynaptic) and DLG (postsynaptic), NMJ structural parameters including bouton number, branch number and synaptic area were quantified in sulf1 and hs6st null alleles. The mutant results closely recapitulated the RNAi knockdown findings from the screen (Table S1). To consistently compare RNAi and null mutant conditions, both animal groups were simultaneously reared and processed to visualize the NMJ (Figure S1). Structural quantification showed an increased bouton number with RNAi-mediated sulf1 knockdown (sulf1-RNAi×UH1-GAL4; 36.4±1.6, n = 10) and hs6st knockdown (hs6st-RNAi×UH1-GAL4; 35.1±1.96, n = 10) compared to the transgenic control (w1118×UH1-GAL4; 21.9±1.84, p<0.001, n = 10; Figure S1A, S1B). Consistently, increased bouton number was observed in both sulf1 (31.9±1.37, n = 10) and hs6st (36.25±2.58, n = 8) null mutants compared to genetic control (w1118, 19.3±1.69, p<0.001, n = 10; Figure S1C, S1D). In contrast, no significant change in branch number was exhibited with sulf1 knockdown (3.22±0.28, p>0.05, n = 9) or hs6st knockdown (3.22±0.22, p>0.05, n = 9) compared to control (w1118×UH1-GAL4; 2.64±0.06, n = 11). Similarly, no significant change was observed in the synaptic branch number in sulf1 (2.8±0.33, p = 0.27, n = 10,) and hs6st (3.63±0.38, p = 0.115, n = 10) nulls compared to control (w1118; 3.4±0.46, n = 8). Further, there was no significant difference in synaptic area in sulf1 (138.16±5.82, p>0.05, n = 10,) and hs6st (138.48±13.38, p>0.05, n = 8,) mutants compared to the control (w1118; 118.04±8,38, n = 10), however a slight increase in synaptic area was observed in sulf1 knockdown (178.68±10.64, p<0.05, n = 9), while no change was observed for hs6st knockdown (164±8.47, p>0.05, n = 10) as compared to control (w1118×UH1-GAL4; 134.57±11.95, n = 10). Based on these imaging studies, we conclude morphological differences in synaptic architecture observed in both sulf1 and hs6st null allele conditions are consistent with both RNAi knockdown conditions.
Functional development was next tested with electrophysiological recording to compare RNAi and null mutant phenotypes (Figure 2). Representative TEVC records are shown as an average of 10 consecutive nerve stimulus responses in 1.0 mM extracellular Ca2+ for each transgenic genotype in Figure 2A; sulf1 knockdown (UH1-GAL4×sulf1-RNAi), hs6st knockdown (UH1-GAL4×hs6st-RNAi) and genetic control (UH1-GAL4×w1118). There was a striking ∼80% difference in EJC amplitude between sulf1 and hs6st knockdown conditions, with sulf1 elevated by ∼30% and hs6st reduced by ∼30% compared to control. Quantification of EJC amplitudes showed both knockdown conditions to be highly significantly different from control and each other (control, 286.22±8.56 nA; sulf1-RNAi, 365.01±9.502 nA, p<0.001; hs6st-RNAi, 199.19±11.84 nA, p<0.001; sulf1-RNAi vs. hs6st-RNAi, p<0.001; Figure 2B). These opposite effects on neurotransmission strength were confirmed in characterized null alleles for both genes [42], [43]. Representative traces from sulf1Δ1 and hs6std770 null mutants compared to w1118 control are shown in Figure 2C. Quantification of EJC amplitudes showed null mutants to be highly significantly different from control and each other (w1118, 256.14±7.38 nA; sulf1Δ1, 372.86±18.49 nA, n = 11, p<0.001; hs6st, 209.66±13.44 nA, n = 14, p<0.01; sulf1Δ1 vs. hs6st, p<0.001; Figure 2D). These results were confirmed in an independent sulf1 null allele (sulf1ΔP1), which shows comparable elevation compared to control (w1118, 244.91±9.04 nA; sulf1ΔP1, 282.28±13.59, p<0.05, n = 22), as well as the hs6st null (hs6std770) over deficiency (Df(3R)ED6027), which shows comparable depression compared to control (w1118, 256.14±7.38 nA; hs6st/Df(3R)ED6027, 224.06±7.65 nA, p<0.05, n = 18). These results reveal a critical role for sulf1 and hs6st genes in synaptic functional development.
Given the functionally-paired nature of sulf1 and hs6st activities on 6-O-S modification, and the epistatic function of hs6st to sulf1, we predicted that knocking both genes down would produce a phenotype similar to knockdown of hs6st alone. Consistently, hs6st and sulf1 double knockdown produced EJC amplitudes significantly lower than control (w1118×hs6st-RNAi; sulf1-RNAi (control), 225.17±6.28 nA, n = 12; hs6st-RNAi, sulf1-RNAi×UH1-GAL4, 198.22±9.77 nA, n = 15, p<0.05; Figure S2). Cell-specific knockdown in neural (elav-GAL4), muscle (24B-GAL4) and glia (repo-GAL4) also support the observed opposite effects in neurotransmission strength. With sulf1 knockdown in muscle, EJC amplitude was significantly elevated compared to control (w1118×sulf1-RNAi (control), 199.97±21.86 nA; 24B-GAL4×sulf1-RNAi (knockdown), 222.88±25.78 nA, p<0.01, n = 10), but no change occurred with neural knockdown (elav-GAL4×sulf1-RNAi, 196.09±25.08 nA, p = 0.72, n = 10) or glial knockdown (repo-GAL4×sulf1-RNAi, 208.40±32.45 nA, p = 0.53, n = 7). Moreover, only neural knockdown of hs6st caused a decrease in EJC amplitude (w1118×hs6st-RNAi (control), 211.496±22.142 nA, elav-GAL4×hs6st-RNAi (knockdown), 184.68±28.97 nA, p<0.05, n = 16), while no change occurred with muscle knockdown (24B-GAL4×hs6st-RNAi, 209.92±24.74 nA, p = 0.88, n = 9) or glial knockdown (repo-GAL4×hs6st-RNAi, 216.38±37.80 nA, p = 0.32, n = 7). We conclude that HSPG sulfation state strongly modulates NMJ functional development, with contributions from both motor neuron and muscle, but not glia. The clear next step was to test for differences in the localization and abundance of synaptic HSPG targets known to regulate NMJ synaptogenesis.
Both GPI-anchored HSPG glypican Dally-like (Dlp) and transmembrane HSPG Syndecan (Sdc) are clearly expressed at the Drosophila NMJ (Figure S3), where they are known to regulate synaptogenesis [45]. We detect no enrichment of the secreted HSPG perelcan (Trol) at the NMJ, although it is abundantly expressed in the motor nerve leading up to the synaptic terminal and present in lower levels throughout the muscle (Figure S4). We therefore hypothesized that membrane-associated Dlp and Sdc HSPGs are targeted by sulf1 and hs6st activity to regulate their synaptic distribution and/or function. To test this hypothesis, we assayed both Dlp and Sdc under non-permeabilized, detergent-free conditions to examine their cell surface expression at the NMJ synaptic interface of sulf1 and hs6st null mutants compared to control. These data are summarized in Figure 3.
In the genetic background control (w1118), Dlp shows a punctate expression pattern strongly concentrated in a halo-like array around the anti-HRP labeled presynaptic membrane (Figure 3A, top; Figure S3). In sulf1 mutants there was a clear and consistent increase in Dlp abundance, with more numerous and intense punctae at the synaptic interface surrounding NMJ boutons, while at hs6st mutant synapses there was an opposing decrease in Dlp abundance (Figure 3A). This bidirectional and differential effect on Dlp abundance was quantified as fluorescence intensity normalized to the internal HRP labeling control. There was a significant Dlp increase in sulf1 compared to control (∼40% elevated over control; p<0.05; n = 11), and a significant Dlp decrease in the hs6st null synapse (∼15% reduced compared to control; p<0.05; n = 11; Figure 3B). Importantly, the difference between sulf1 and hs6st nulls was very highly significant (p<0.001). In comparison, cell surface Sdc labeling also showed a dense halo-like localization around NMJ synaptic boutons labeled with cell adhesion marker Fasciclin II (FasII; Figure 3C; Figure S3). Synaptic Sdc labeling intensity was consistently greater in both sulf1 and hs6st nulls compared to control (Figure 3C). Quantification of fluorescence intensity normalized to HRP revealed that Sdc abundance was greatly increased in sulf1 null synapses compared to control (∼35% elevated over control; p<0.01; n = 17) and, to a greater degree, also in hs6st nulls (∼50% elevated over control; p<0.001; n = 12; Figure 3D). Thus, both Dlp and Sdc HSPGs are strongly altered in sulf1 and hs6st null NMJ synapses, with Dlp bidirectionally misregulated and Sdc differentially elevated in the two mutant conditions.
HSPGs act as co-receptors for WNT and BMP intercellular signaling ligands in many developmental contexts, acting to modulate extracellular ligand abundance and downstream signaling [46], [47]. Drosophila WNT Wingless (Wg) distribution and signaling is known to be modulated by Dlp, which retains Wg at the cell surface in a mechanism that is enhanced by HS GAG chains [48]. Specifically, Wg ligand abundance and signaling activity along the dorso-ventral axis of the developing Drosophila wing disc is elevated in sulf1 mutants [22]. Likewise, BMP ligands in other cellular contexts are closely regulated by HSPG co-receptors [20]. Specifically, Dlp has been suggested to similarly regulate Drosophila BMP Glass Bottom Boat (Gbb) [20]. We therefore hypothesized that altered HSPG co-receptors Dlp and/or Sdc in sulf1 and hs6st null synapses regulate Wg and Gbb abundance to drive differentially altered trans-synaptic signaling across the synaptic cleft.
Classical WNT and BMP morphogens act locally at synapses to fine tune synaptogenesis [49], [50]. At the Drosophila NMJ, the WNT Wg is well-characterized as an anterograde trans-synaptic signal modulating synaptogenesis [23], [24], [51]. Similarly, the BMP Gbb is well-characterized as a retrograde signal driving synaptic development [25], [26], [52]. A third trans-synaptic signaling pathway, presynaptically-secreted Jelly Belly (Jeb) to postsynaptic Alk receptor [17], has no known interaction with HSPGs and therefore would not be expected to be affected in sulf1 and hs6st nulls, providing a comparison for specificity. To test the hypothesis that the observed alterations of HSPG co-receptor abundance will drive specific changes in WNT and BMP intercellular pathways, we labeled NMJ synapses with antibodies under non-permeablized conditions to reveal extracellular trans-synaptic signaling ligands (Figure S5), and compared protein abundance and distribution in controls, sulf1 and hs6st null mutants. The data are summarized in Figure 4.
NMJ synapses were first labeled with Wg antibody (green) together with anti-HRP (red) to label the presynaptic membrane (Figure 4A). In control animals (w1118), external Wg localized at large type Ib synaptic boutons in a dynamic pattern of punctuate distribution at the synaptic interface between motor neuron and muscle (Figure 4A, top; Figure S5). In sulf1 and hs6st mutants, Wg was consistently elevated and concentrated uniformly in the extracellular domain adjacent to, and overlapping with, the anti-HRP-labeled presynaptic membrane (Figure 4A, middle and bottom). The elevated Wg levels in mutants were clearly observed at the level of individual synaptic boutons, as shown in the magnified insets in Figure 4A. To examine changes in Wg spatial distribution, cross-sectional planes were examined in single confocal line scans through the diameter of individual synaptic boutons (Figure 4A, white lines). Representative distribution plots for membrane-marker HRP (red) and external Wg (green) are shown in Figure 4B. In all genotypes, extracellular Wg was closely associated with the HRP-labeled presynaptic membrane, but both sulf1 and hs6st nulls displayed a consistent increase in Wg label intensity and broadening of the spatial domain occupied by the secreted Wg ligand (Figure 4B, middle and bottom). To quantify changes in extracellular Wg abundance, the mean fluorescent signal intensity was normalized to the internal HRP co-label, and then normalized to analogous control intensity ratios. In sulf1Δ1 nulls, there was very highly significant elevation of Wg compared to control (∼90% increased; p<0.001; n = 16; Figure 4C). A similar increase was observed in the independent sulf1ΔP1 null (p<0.001; n = 11). The hs6st null displayed a smaller significant increase in Wg abundance (∼40% increased; p<0.001; n = 15; Figure 4C), which was again recapitulated in hs6st null over deficiency (Df(3R)ED6027) condition. Importantly, Wg abundance is differentially elevated in sulf1 vs. hs6st mutants (p<0.01, Figure 4C).
To test whether the sulf1/hs6st mechanism might coordinately regulate multiple trans-synaptic signals, we next assayed the BMP Gbb, a muscle-derived retrograde signal [25]. A barrier to previous Gbb analyses has been the absence of an anti-Gbb antibody. We therefore generated a specific anti-Gbb antibody for this study (see Methods). As above, labeling was done under non-permeabilized conditions to reveal only the extracellular Gbb, together with labeling for HRP or the cell adhesion molecule marker FasII to reveal the presynaptic membrane (Figure S5). In the control (w1118), extracellular Gbb concentrated in a ring of punctate domains around boutons (Figure 4D, top). Gbb was similarly punctate in sulf1 and hs6st nulls, but consistently more extensive and denser (Figure 4D, middle and bottom; see magnified insets). To examine Gbb spatial distribution, cross-sectional planes of confocal line scans were made through individual synaptic boutons (Figure 4D, white lines). Representative plots for FasII (green) and Gbb (red) show extracellular Gbb closely associated with the FasII-labeled presynaptic membrane in all genotypes (Figure 4E). However, sulf1 and hs6st nulls consistently displayed increased Gbb intensity and broadened expression compared to the control. Upon quantifying signal intensity of Gbb normalized to HRP co-label, sulf1Δ1 exhibited a significantly higher Gbb abundance than control (65% increased; p<0.01; n = 12; Figure 4F). The independent sulf1ΔP1 null allele showed a similar increase (p<0.001; n = 12). The hs6st null also showed Gbb elevation compared to control (59% increased; p<0.01; n = 11; Figure 4E), which was confirmed in hs6st null over deficiency (Df(3R)ED6027; p<0.05; n = 23).
To test further whether extracellular Wg and Gbb abundance was sensitive to the sulfation state of GAGs, a biochemical approach was next used to determine effects on Wg and Gbb trans-synaptic signals (Figure S6). Specifically, NMJs were acutely exposed to heparin, the most sulfated form of GAG [53], and then synaptic Wg and Gbb abundance was measured by immunolabeling as above. We found that both trans-synaptic signals were rapidly altered by heparin incubation in a dose-dependent manner. Specifically, incubation with increasing concentrations of heparin caused a reciprocal decrease in Wg labeling intensity in the NMJ synaptic domain (Figure S6A, S6C), with a significant decrease first detected with 0.315 mg/ml heparin incubation (∼50% less than control, p<0.01, n = 4). Interestingly, increasing heparin concentrations caused a parallel increase in Gbb abundance in the NMJ synaptic domain (Figure S6B, S6C) in a dose-dependent manner, with significant increases again first detected at 0.315 mg/ml heparin (∼25% greater than control, p<0.05) and rising further at 0.625 mg/ml heparin (∼40% greater than control, p<0.001). These results indicate that HSPG sulfation state does indeed affect trans-synaptic signal abundance, supporting the observed alterations in Wg and Gbb abundance in mutants of heparan sulfate modifying genes, sulf1 and hs6st.
To examine effects on other trans-synaptic signaling pathways in the sulf1 and hs6st mutant synapses, we also assayed for changes in Jeb [17] and FGF [17] signaling. In both control and mutants, extracellular Jeb labeling was tightly associated with NMJ type Ib boutons and, like other trans-synaptic ligands, occupied an extracellular domain closely associated with the presynaptic membrane (Figure S7A). However, in stark contrast to Wg and Gbb ligands in the same extracellular synaptomatrix domain, no change was observed in Jeb abundance or spatial distribution in sulf1 null (p = 0.99, n = 10) or hs6st null (p = 0.36, n = 8) compared to control (w1118) NMJ synapses (Figure S7B). FGF signaling is also well established to be affected by HSPGs [54], and one pioneering study has investigated roles for FGF signaling at the Drosophila NMJ [55]. The probe used in the previous study was an antibody against the FGF receptor Heartless (Htl) [56]. Using this antibody, we confirmed that the Htl receptor beautifully localizes to NMJ boutons to mediate FGF signaling (Figure S8A). However, Htl receptor synaptic abundance and distribution was very similar for the sulf1 (p = 0.89, n = 9) and hs6st (p = 0.69, n = 7) mutants compared to control (w1118) (Figure S8B). Unfortunately, no antibody probes are available for Drosophila FGF ligands, so these signals have not yet been queried. Together, these results show that both WNT (Wg) and BMP (Gbb) ligand abundance is coordinately upregulated by the sulf1 and hs6st mechanism at the NMJ synapse, but that a spatially overlapping signaling ligand (Jeb) and at least FGF receptor expression are unaffected. These results strongly predict that Wg and Gbb trans-synaptic signaling controlled by sulf1 and hs6st activity regulates synaptic functional development.
Wg and Gbb serve as anterograde and retrograde trans-synaptic signals, respectively, activating cognate receptors to initiate downstream signaling cascades and nuclear import pathways in muscles and motor neurons, respectively [24], [26], [50], [51]. The anterograde Wg signal drives dFrizzled-2 (dFz2) receptor internalization in the postsynaptic domain followed by cleavage of the receptor C-terminus, which then enters the muscle nuclei [57]. The muscle-derived retrograde Gbb signal activates presynaptic receptors to drive phosphorylation of the Mothers Against Decapentaplegic (Mad) transcription factor, and then P-Mad enters the motor neuron nuclei to regulate transcription [25], [26], [58]. Given the differential change in both HSPG co-receptor and Wg/Gbb ligand abundance in sulf1 vs. hs6st mutants, we hypothesized that these signaling pathways would be differentially affected during synaptogenesis. We therefore quantitatively assayed the paired muscle and motor neuron nuclear import pathways to determine whether and how trans-synaptic signaling may be modulated by sulf1 and hs6st at the NMJ synapse.
Characterized antibodies specifically recognizing the N- and C-termini of the Wg dFz2 receptor allow measurements of the receptor at the NMJ synapse (dFz2N; Figure S9) and the cleaved fragment (dFz2C; Figure 5) imported into muscle nuclei [57], [59]. We first assayed dFz2 receptor abundance at the NMJ with the N-terminal specific antibody. The dFz2 receptor is closely associated with the synaptic cell membrane marker FasII and occupies a domain that envelopes all type Ib boutons (Figure S9A). In hs6st nulls, the dFz2 receptor domain was spatially extended as compared to controls, however sulf1 alleles showed no detectable change in the receptor. Likewise, fluorescence intensity measurements showed no significant difference between control and sulf1 nulls, but hs6st null synapses displayed a ∼25% increase in dFz2 receptor abundance, a very significant elevation (p<0.01, n = 12; Figure S9B) in synaptic dFz2 abundance. Thus, importantly (see Discussion), significantly more dFz2 receptors occur in the hs6st null compared to sulf1 null synapse.
To assay downstream signal transduction, the cleaved Fz2C fragment imported into muscle nuclei was quantified using the established method of counting dFz2C-positive punctae in nuclei proximal to the NMJ (Figure 5) [59]. In genetic control (w1118), most muscle nuclei contained a small number (1–3) of detectable dFz2C punctae, but some nuclei contained more and others were devoid of detectable dFz2C (Figure 5A, top). More than 100 muscle nuclei were quantified in >7 different animals to determine the control level of dFz2C nuclear import. In sulf1 and hs6st mutants, there was a clear and consistent bidirectional difference in the number and size of dFz2C punctae in muscle nuclei (Figure 5A, middle and bottom). Null sulf1 nuclei showed a highly significant decrease in number of dFz2C punctae per nuclei (>50% decreased; p<0.01; n = 163; Figure 5B). In contrast, hs6st nulls had an opposing highly significant increase in dFz2C punctae per nuclei (>60% increased; p<0.01; n = 163; Figure 5B). The difference between sulf1 and hs6st null mutants was very highly significant (p<0.001), with a differential change in signaling paralleling the bidirectional change in synaptic functional differentiation (Figure 2).
A characterized antibody specifically recognizing phosphorylated Mad (P-Mad) allowed independent measurements of Gbb signaling in the presynaptic terminal and P-Mad import into the motor neuron nuclei as a transcriptional regulator (Figure 6) [25], [60]. To assay this transduction pathway, P-Mad fluorescent intensity normalized to FasII was first assayed in presynaptic boutons [61], [62]. In the genetic control (w1118), P-Mad labeling was bounded by the synaptic cell adhesion molecule marker FasII, with P-Mad localized in numerous punctate domains (Figure 6A, arrows). In sulf1 and hs6st nulls, both the intensity and size of P-Mad positive punctae were obviously and consistently greater than in controls (Figure 6A, middle and bottom). In fluorescence intensity quantification, sulf1 null synapses displayed a significant increase in synaptic P-Mad (45% increased; p<0.05; n = 10; Figure 6C). An increase in P-Mad was also observed in the hs6st null boutons (42% greater than control; p<0.01; n = 15; Figure 6C). The motor neuron nuclei at the ventral nerve cord (VNC) midline accumulate P-Mad transcription factor downstream of Gbb signaling at the NMJ [25], [61], [62]. In genetic control (w1118), P-Mad nuclear labeling was consistently detected in these motor neuron nuclei (Figure 6B, arrows). A similar P-Mad distribution was observed in motor neuron nuclei of sulf1 and hs6st nulls, but the intensity of P-mad expression was clearly and consistently elevated in both mutants compared to control (Figure 6B, middle and bottom). In fluorescence intensity quantification, sulf1 null neuronal nuclei displayed a very significant increase in P-Mad accumulation (15% increased; p<0.01; n = 14; Figure 6D), paralleling increased P-Mad signaling at the NMJ (Figure 6C). Likewise, hs6st null motoneuron nuclei exhibited a smaller but still significant elevation in P-Mad accumulation (9% elevated over control; p<0.05; n = 21; Figure 6D), again paralleling the observed P-Mad signaling change at the NMJ (Figure 6C). We conclude that both anterograde WNT (Wg) and retrograde BMP (Gbb) trans-synaptic signaling in muscle and motor neuron nuclei, respectively, is differentially regulated by the sulf1 and hs6st HSPG sulfation mechanism.
In the sulf1 and hs6st nulls we identified a bi-directional change in synaptic functional differentiation, measured as evoked junction current amplitudes increased in sulf1 and decreased in hs6st null synapses (Figure 2). We therefore hypothesized that these functional changes are driven by the differential Wg and Gbb trans-synaptic signaling defects characterized above in sulf1 and hs6st mutants (Figure 3, Figure 4, Figure 5, Figure 6). We reasoned that correcting Wg and Gbb levels in sulf1 and hs6st nulls should restore neurotransmission to control levels. To test this hypothesis, we crossed heterozygous wg/+ and gbb/+ mutants into both sulf1 and hs6st homozygous null backgrounds, both singly and in combination, and compared them to both positive and negative controls. The resulting 9 genotypes were all assayed with TEVC electrophysiology to compare EJC transmission strength. A summary of these data is given in Figure 7.
Representative transmission records are shown as an average of 10 consecutive EJC responses (1.0 mM extracellular Ca2+) for the genotypes in Figure 7A, with quantification of mean peak amplitudes in all genotypes shown in Figure 7B. First testing sulf1 nulls, we examined the consequences of heterozygous genetic reduction of Wg and Gbb, alone and in combination. Compared to the elevated EJC amplitude of the sulf1 null condition (381.28±1 62.24 nA, p<0.01, n = 9; Figure 7B), genetic reduction of Wg (wg/+; sulf1/sulf1) caused very significantly reduced transmission, similar to genetic reduction of Gbb (gbb/+; sulf1/sulf1) with a comparable effect, restoring EJC amplitude to control levels (267.16±16.33, p<0.01, n = 9; Figure 7B). Combinatorial genetic reduction of both Wg and Gbb in the sulf1 null (wg/gbb;sulf1/sulf1) similarly returned EJC amplitudes to control levels (278.78±23.17, n = 7; Figure 7B). Secondly testing hs6st nulls, genetic reduction of either Wg or Gbb alone was not sufficient to significantly change the depressed synaptic function (Figure 7B). In this case, combinatorial genetic reduction of both Wg and Gbb in the hs6st null (wg/gbb;hs6st/hs6st) was required to raise the depressed EJC amplitude, a very significant increase back to control levels (272.98±18.58, p<0.01, n = 8; Figure 7B). Therefore, we conclude that combinatorial Wg and Gbb trans-synaptic signaling defects are causative for the observed bi-directional effects on synaptic functional differentiation in the sulf1 and hs6st null mutant conditions.
The consequence of WNT (Wg) and BMP (Gbb) trans-synaptic signaling is nuclear import and transcriptional regulation in both synaptic partner cells [49], [51]. We therefore hypothesized that sulf1 and hs6st null mutants would show bidirectional changes in pre- and postsynaptic molecular components that would explain the bidirectional change in synaptic functional differentiation (Figure 2 and Figure 7). To test this hypothesis, we examined a key component of the presynaptic active zone (Bruchpilot; Brp) [63], and an essential subunit of the postsynaptic glutamate receptor (Bad Reception (Brec); GluRIID) [64]. In parallel, we also performed a miniature EJC (mEJC) analysis to compare functional presynaptic vesicle release probability and postsynaptic response amplitude. A summary of these data is shown in Figure 8.
First, NMJ synapses were double-labeled for GluRIID recognized with anti-Brec (green) and Brp recognized with anti-nc82 (red) to compare genetic control (w1118) with sulf1 and hs6st nulls (Figure 8A). We found that GluRIID was very significantly elevated at sulf1 synapses compared to control (∼30% increased; p<0.01, n = 20; Figure 8B). In the opposing direction, hs6st null synapses showed a significant decrease in GluRIID abundance (∼15% reduced; p<0.05, n = 21; Figure 8B). The GluRIID field area per bouton and number of GluRIID punctae normalized to field area per synaptic bouton were also bidirectionally altered in the sulf1 and hs6st nulls (Figure 8C, 8D). GluRIID receptor field area was increased in sulf1 (∼30% greater; p<0.01, n = 47) but decreased in hs6st (∼25% reduced; p<0.01, n = 51). Conversely, measurements of GluRIID puncta normalized to field area per synaptic bouton were decreased in sulf1 (∼15% lower; p<0.05, n = 47), but increased in hs6st nulls (∼40% greater; p<0.01, n = 51, Figure 8D). The bi-directional differences between sulf1 and hs6st were very highly significant (p<0.001). The active zone protein Brp also showed opposite effects (Figure 8A). Although the difference between sulf1 null and control was not quite significant (p>0.05, n = 20), hs6st null synapses showed a very significant decrease in Brp compared to control (∼20% reduced; p<0.01, n = 21; Figure 8A).
Based on these results, we next tested pre- (Brp) and postsynaptic (Brec/GluRIID) changes in sulf1 and hs6st mutants with genetic reduction of Wg and Gbb (wg/gbb;sulf1/sulf1 and wg/gbb;hs6st/hs6st), as in Figure 7. Distribution changes of both pre- and postsynaptic components were assayed as measurements of glutamate receptor field and active zone areas (Figure S10A). To measure glutamate receptor distribution comparing wg/gbb;sulf1/sulf1 to matched control, we counted the number of GluRIID punctae per bouton (p = 0.73, n = 48; Figure S10B) and GluRIID area (p = 0.92, n = 48; Figure S10C), and found both corrected back to control levels. Likewise, for wg/gbb;hs6st/hs6st compared to control, GluRIID puncta number (p = 0.88, n = 48) and area (p = 0.41, n = 58) were both corrected to control levels. To measure Brp-positive presynaptic active zones comparing wg/gbb;sulf1/sulf1 to matched control, we counted the number of Brp punctae per bouton (p = 0.43, n = 48; Figure S10D) and Brp area (p = 0.39, n = 48; Figure S10D), and found both corrected back to control levels. Likewise, for wg/gbb;hs6st/hs6st compared to control, Brp number (p = 0.54, n = 58) and area (p = 0.19, n = 58) were also corrected back to control levels. These results provide strong genetic evidence that Wg and Gbb trans-synaptic signaling changes are causative for the pre- and postsynaptic molecular differentiation defects in the sulf1 and hs6st null mutants.
These bidirectional pre- and postsynaptic molecular changes parallel functional transmission changes in sulf1 and hs6st mutants (Figure 2). To assay function at the single synapse level, we finally assayed spontaneous synaptic vesicle fusion events. Representative mEJC traces for control compared to sulf1 and hs6st nulls are shown in Figure 8E. Consistent with observed bidirectional changes in evoked transmission, mEJC amplitudes in hs6st were ∼25% lower than in sulf1 nulls (hs6st, 0.60±0.02 nA vs. sulf1, 0.76±0.05 nA; p<0.5, n = 34; Figure 8F). Moreover, hs6st nulls had a ∼100% elevated mEJC frequency compared to sulf1 nulls (hs6st, 2.56±0.27 vs. sulf1, 1.30±0.09; p<0.001, n = 34; Figure 8G). Based on these mEJC measurements, there was a highly significant bidirectional change in quantal content between the two mutant conditions, with sulf1 quantal content ∼50% greater than hs6st (sulf1, 539.98±22.02 vs. hs6st, 350.69±8.92; p<0.001, n = 34; Figure 8H). Taken together, these results show a bi-directional change in presynaptic glutamate release machinery and vesicle fusion probability, as well as postsynaptic glutamate receptor levels and functional responsiveness. We conclude that these changes underlie the bi-directional switch in neurotransmission strength characterizing sulf1 and hs6st mutants.
It is well known that synaptic interfaces harbor heavily-glycosylated membrane proteins, glycolipids and ECM molecules, but understanding of glycan-mediated mechanisms within this synaptomatrix is limited [9]. Our genomic screen aimed to systematically interrogate glycan roles in both structural and functional development in the genetically-tractable Drosophila NMJ synapse. 130 candidate genes were screened, classified into 8 functional families: N-glycan biosynthesis, O-glycan biosynthesis, GAG biosynthesis, glycoprotein/proteoglycan core proteins, glycan modifying/degrading enzymes, glycosyltransferases, sugar transporters and glycan-binding lectins. From this screen, 103 RNAi knockdown conditions were larval viable, whereas 27 others produced early developmental lethality. 35 genes had statistically significant effects on different measures of morphological development: 27 RNAi-mediated knockdowns increased synaptic bouton number, 9 affected synapse area (2 increased, 7 decreased) and 2 genes increased synaptic branch number. These data suggest that overall glycan mechanisms predominantly serve to limit synaptic morphogenesis. 13 genes had significant effects on the functional differentiation of the synapse, with 12 increasing transmission strength and only 1 decreasing function upon RNAi knockdown. Thus, glycan-mediated mechanisms also predominantly limit synaptic functional development. A very small fraction of tested genes (CG1597; pgant35A, CG7480; veg, CG6657; hs6st, CG4451; sulf1, CG6725 and CG11874) had effects on both morphology and function. A large percentage of genes (∼30%) showed morphological defects with no corresponding effect on function, while only 7% of genes showed functional alterations without morphological defects, and <5% of all genes affect both. These results suggest that glycans have clearly separable roles in modulating morphological and functional development of the NMJ synapse.
A growing list of neurological disorders linked to the synapse are attributed to dysfunctional glycan mechanisms, including muscular dystrophies, cognitive impairment and autism spectrum disorders [65], [66], [67]. Drosophila homologs of glycosylation genes implicated in neural disease states include ALG3 (CG4084), ALG6 (CG5091), DPM1 (CG10166), FUCT1 (CG9620), GCS1 (CG1597), MGAT2 (CG7921), MPDU1 (CG3792), PMI (CG33718) and PPM2 (CG12151) [65]. Two of these genes, Gfr (CG9620) and CG1597, showed synaptic morphology phenotypes in our RNAi screen. Given that connectivity defects are clearly implicated in cognitive impairment and autism spectrum disorders [68], [69], it would be of interest to explore the glycan mechanism affecting synapse morphology in Drosophila models of these disease states. Glycans are well known to modulate extracellular signaling, including ligands of integrin receptors, to regulate intercellular communication [70], [71]. In our genetic screen, several O-glycosyltransferases mediating this mechanism were identified to show morphological (GalNAc-T2, CG6394; pgant35A, CG7480, O-fut2, CG14789; rumi, CG31152) and functional (pgant5, CG31651; pgant35A, CG7480) synaptic defects upon RNAi knockdown. These findings suggest that known integrin-mediated signaling pathways controlling NMJ synaptic structural and functional development [16], [41], [72], [73] are modulated by glycan mechanisms. Our screen showed CG6657 RNAi knockdown affects functional differentiation, consistent with reports that this gene regulates peripheral nervous system development [74]. The corroboration of our screen results with published reports underscores the utility of RNAi-mediated screening to identify glycan mechanisms, and supports use of our screen results for bioinformatic/meta-analysis to link observed phenotypes to neurophysiological/pathological disease states and to direct future glycan mechanism studies at the synapse.
From our screen, the two functionally-paired genes sulf1 and hs6st were selected for further characterization. As in the RNAi screen, null alleles of these two genes had opposite effects on synaptic functional differentiation but similar effects on synapse morphogenesis, validating the corresponding screen results. The two gene products have functionally-paired roles; Hs6st is a heparan sulfate (HS) 6-O-sulfotransferase [43], and Sulf1 is a HS 6-O-endosulfatase [75]. These activities control sulfation of the same C6 on the repeated glucosamine moiety in HS GAG chains found on heparan sulfate proteoglycans (HSPGs). At the Drosophila NMJ, two HSPGs are known to regulate synapse assembly; the GPI-anchored glypican Dally-like protein (Dlp), and the transmembrane Syndecan (Sdc) [45]. In contrast, the secreted HSPG Perlecan (Trol) is not detectably enriched at the NMJ [76], and indeed appears to be selectively excluded from the perisynaptic domain. In other developmental contexts, the membrane HSPGs Dlp and Sdc are known to act as co-receptors for WNT and BMP ligands, regulating ligand abundance, presentation to cognate receptors and therefore signaling [20], [48]. Importantly, the regulation of HSPG co-receptor abundance has been shown to be dependent on sulfation state mediated by extracellular sulfatases [77]. Consistently, we observed upregulation of Dlp and Sdc in sulf1 null synapses, whereas Dlp was reduced in hs6st null synapses. In the developing Drosophila wing disc, HSPG co-receptors increase levels of the Wg ligand due to extracellular stabilization [78], and the primary function of Dlp in this developmental context is to retain Wg at the cell surface [21]. Likewise, in developing Drosophila embryos, a significant fraction of Wg ligand is retained on the cell surfaces in a HSPG-dependent manner [79], with the HSPG acting as an extracellular co-receptor. Syndecan also modulates ligand-dependent activation of cell-surface receptors by acting as a co-receptor [19], [20]. At the NMJ, regulation of both these HSPG co-receptors occurs in the closely juxtaposed region between presynaptic bouton and muscle subsynaptic reticulum, in the exact same extracellular space traversed by the secreted trans-synaptic Wg and Gbb signals [45]. We therefore proposed that altered Dlp and Sdc HSPG co-receptors in sulf1 and hs6st mutants differentially trap/stabilize Wg and Gbb trans-synaptic signals at the interface between motor neuron and muscle, to modulate the extent and efficacy of intercellular signaling driving synaptic development.
HS sulfation modification is linked to modulating the intercellular signaling driving neuronal differentiation [80]. In particular, WNT and BMP ligands are both regulated via HS sulfation of their extracellular co-receptors, and both signals have multiple functions directing neuronal differentiation, including synaptogenesis [49], [50], [51]. In the Drosophila wing disc, extracellular WNT (Wg) ligand abundance and distribution was recently shown to be strongly elevated in sulf1 null mutants [22]. Moreover, sulf1 has also recently been shown to modulate BMP signaling in other cellular contexts [81]. Consistently, we have shown here increased WNT Wg and the BMP Gbb abundance and distribution in sulf1 null NMJ synapses. The hs6st null also exhibits elevated Wg and Gbb at the synaptic interface, albeit the increase is lower and results in differential signaling consequences. In support of this contrasting effect, extracellular signaling ligands are known to bind HSPG HS chains differentially dependent on specific sulfation patterns [82], [83], [84]. It is important to note that the sulf1 and hs6st modulation of trans-synaptic signals is not universal, as Jelly Belly (Jeb) ligand abundance and distribution was not altered in the sulf1 and hs6st null conditions [17]. This indicates that discrete classes of secreted trans-synaptic molecules are modulated by distinct glycan mechanisms to control NMJ structure and function.
At the Drosophila NMJ, Wg is very well characterized as an anterograde trans-synaptic signal [23], [24], [85] and Gbb is very well characterized as a retrograde trans-synaptic signal [25], [26], [50], [86]. In Wg signaling, the dFz2 receptor is internalized upon Wg binding and then cleaved so that the dFz2-C fragment is imported into muscle nuclei [57], [59], [85]. In hs6st nulls, increased Wg ligand abundance at the synaptic terminal corresponds to an increase in dFz2C punctae in muscle nuclei as expected. In contrast, the increase in Wg at the sulf1 null synapse did not correspond to an increase in the dFz2C-terminus nuclear internalization, but rather a significant decrease. One explanation for this apparent discrepancy is the ‘exchange factor’ model based on the biphasic ability of the HSPG co-receptor Dlp to modulate Wg signaling [48]. In the Drosophila wing disc, this model suggests that the transition of Dlp co-receptor from an activator to repressor of signaling depends on Wg cognate receptor dFz2 levels, such that a low ratio of Dlp∶dFz2 potentiates Wg-dFz2 interaction, whereas a high ratio of Dlp∶dFz2 prevents dFz2 from capturing Wg [48]. In sulf1 null synapses, we observe a very great increase in Dlp abundance (∼40% elevated) with no significant change in the dFz2 receptor. In contrast, at hs6st null synapses there is a decrease in Dlp abundance (15% decreased) together with a significant increase in dFz2 receptor abundance (∼25% elevated). Thus, the higher Dlp∶dFz2 ratio in sulf1 nulls could explain the decrease in Wg signal activation, evidenced by decreased dFz2-C terminus import into the muscle nucleus. In contrast, the Dlp∶Fz2 ratio in hs6st is much lower, supporting activation of the dFz2-C terminus nuclear internalization pathway. This previously proposed competitive binding mechanism dependent on Dlp co-receptor and dFz2 receptor ratios predicts the observed synaptic Wg signaling pathway modulation in sulf1 and hs6st dependent manner [48].
At the Drosophila NMJ, Gbb is very well characterized as a retrograde trans-synaptic signal, with muscle-derived Gbb causing the receptor complex Wishful thinking (Wit), Thickveins (Tkv) and Saxaphone (Sax) to induce phosphorylation of the transcription factor mothers against Mothers against decapentaplegic (P-Mad) [25], [26], [87]. Mutation of Gbb ligand, receptors or regulators of this pathway have shown that Gbb-mediated retrograde signaling is required for proper synaptic differentiation and functional development [25], [52], [61], [86], [88]. Further, loss of Gbb signaling results in significantly decreased levels of P-Mad in the motor neurons [25]. We show here that accumulation of Gbb in sulf1 and hs6st null synapses causes elevated P-Mad signaling at the synapse and P-Mad accumulation in motor neuron nuclei. Importantly, sulf1 null synapses show a significantly higher level of P-Mad signaling compared to hs6st null synapses, and this same change is proportionally found in P-Mad accumulation within the motor neuron nuclei. These findings indicate differential activation of Gbb trans-synaptic signaling dependent on the HS sulfation state is controlled by the sulf1 and hs6st mechanism, similar to the differential effect observed on Wg trans-synaptic signaling. Our genetic interaction studies show that these differential effects on trans-synaptic signaling have functional consequences, and exert a causative action on the observed bi-directional functional differentiation phenotypes in sulf1 and hs6st nulls. Genetic correction of Wg and Gbb defects in the sulf1 null background restores elevated transmission back to control levels. Similarly, genetic correction of Wg and Gbb in hs6st nulls restores the decreased transmission strength back to control levels. These results demonstrate that the Wg and Gbb trans-synaptic signaling pathways are differentially regulated and, in combination, induce opposite effects on synaptic differentiation.
Both wg and gbb pathway mutants display disorganized and mislocalized presynaptic components at the active zone (e.g. Bruchpilot; Brp) and postsynaptic components including glutamate receptors (e.g. Bad reception; Brec/GluRIID) [23], [86], [89]. Consistently, the bi-directional effects on neurotransmission strength in sulf1 and hs6st mutants are paralleled by dysregulation of these same synaptic components. Changes in presynaptic Brp and postsynaptic GluR abundance/distribution causally explain the bi-directional effects on synaptic functional strength between sulf1 and hs6st null mutant states. Alterations in active zone Brp and postsynaptic GluRs also agree with assessment of spontaneous synaptic activity. Null sulf1 and hs6st synapses showed opposite effects on miniature evoked junctional current (mEJC) frequency (presynaptic component) and amplitude (postsynaptic component). Further, quantal content measurements also support the observation of bidirectional synaptic function in the two functionally paired nulls. Genetic correction of Wg and Gbb defects in both sulf1 and hs6st nulls restores the molecular composition of the pre- and postsynaptic compartments back to wildtype levels. When both trans-synaptic signaling pathways are considered together, these data suggest that HSPG sulfate modification under the control of functionally-paired sulf1 and hs6st jointly regulates both WNT and BMP trans-synaptic signaling pathways in a differential manner to modulate synaptic functional development on both sides of the cleft.
We present here the first systematic investigation of glycan roles in the modulation of synaptic structural and functional development. We have identified a host of glycan-related genes that are important for modulating neuromuscular synaptogenesis, and these genes are now available for future investigations, to determine mechanistic requirements at the synapse, and to explore links to neurological disorders. As proof for the utilization of these screen results, this study has identified extracellular heparan sulfate modification as a critical platform of the intersection for two secreted trans-synaptic signals, and differential control of their downstream signaling pathways that drive synaptic development. Other trans-synaptic signaling pathways are independent and unaffected by this mechanism, although it is of course possible that a larger assortment of signals could be modulated by this or similar mechanisms. This study supports the core hypothesis that the extracellular space of the synaptic interface, the heavily-glycosylated synaptomatrix, forms a domain where glycans coordinately mediate regulation of trans-synaptic pathways to modulate synaptogenesis and subsequent functional maturation.
The glycan-related gene collection was generated using the KEGG glycan databases and Flybase annotation. The 163 UAS-RNAi lines tested were obtained from the Vienna Drosophila RNAi Center (VDRC) and Harvard TriP collection. Transgenic UAS-RNAi males were crossed to GAL4 driver females, with progeny raised at 25°C on standard food, controlling for density (3 ♀ crossed to 2 ♂). The UH1-GAL4 driver was used for ubiquitous knockdown of target gene expression [15]. Neural specific elav-GAL4 [90], muscle specific 24B-GAL4 [91] and glia specific repo-GAL4 lines [92] from Bloomington stock center were used to assay cell-targeted knockdown. The two sulf1 null alleles used were sulf1Δ1 [42] and sulf1ΔP1 [43]. The two hs6st null alleles used were hs6std770 and the deficiency Df(3R)ED6027 [93]. The wg allele wgI-12 [94] and gbb alleles gbb1 and gbb2 were used [25], [87]. Multiply mutant animals were made using standard genetic crosses. The trol-GFP line was obtained from Flytrap [76].
We generated a rabbit polyclonal anti-Gbb antibody using a 1∶1 combination of two Gbb-specific peptides (SHHRSKRSASHP, NDENVNLKKYRNMIVKSC) corresponding to amino acids 319–330 and 435–452 of Gbb (Young-In Frontier, Seoul, Korea). The antibody was purified by Protein A affinity chromatography, and antibody specificity demonstrated by examining immunoreactivity in the wandering third instar neuromusculature with gbb mutants and by expressing UAS-gbb9.1 under the control of the muscle driver BG57-GAL4 (Figure S11). Immunoreactivity in the wandering third instar neuromusculature was severely reduced in a strong hypomorphic gbb allele (gbb1/gbb2, UAS-gbb9.9), which has leaky expression of UAS-gbb9.9 in a null allelic combination [25], [87], [95]. In sharp contrast, the anti-Gbb signal was strongly elevated in BG57-GAL4/UAS-gbb9.1 relative to wildtype larvae.
Wandering third instars were dissected in Ca2+-free saline and then immediately fixed in either 4% paraformaldehyde for 10 minutes (all labels except anti-Dlp) or Bouin's fixative for 30 mins (anti-Dlp). Preparations were then washed in permeabilizing PBST (PBS+0.1% Triton-X) or detergent-free PBS for extracellular labeling only [16]. The following primary antibodies were used: rabbit or goat anti-HRP (1∶250; Jackson ImmunoResearch Laboratories); mouse anti-DLG (4F3; 1∶250; Developmental Studies Hybridoma Bank (DSHB)); mouse anti–Fasciclin II (1D4; 1∶5; DSHB); mouse anti-Dlp (13G8, 1∶5; DSHB) and rabbit anti-Syndecan (1∶200) [96]; mouse anti-Wg (4D4; 1∶2 DSHB) and rabbit anti-Gbb (1∶100); rabbit anti-PcanV (1∶1000) [97]; guinea pig anti-Jeb (1∶100) [17]; rabbit anti-dFz2-C (1∶500) and rabbit anti-dFz2-N (1∶100) [57]; rabbit anti-Htl (1∶100) [56]; rabbit anti-P-Mad (PS1; 1∶1000) [60]; rabbit anti-GluRIID (1∶500) [64] and mouse anti-BRP (1∶100; DSHB). Primary antibodies were incubated at 4°C overnight. Alexa-conjugated secondary antibodies (Jackson ImmunoResearch Laboratories) were used at 1∶250 dilutions for 2 hours at room temperature. Staining with propidium iodide (Sigma Aldrich) to visualize cell nuclei was done at 1∶100 dilution of 1 mg/ml propidium iodide incubated for 30 minutes at room temperature.
Images were taken with on an upright Zeiss LSM 510 META laser-scanning confocal using a Plan Apo 63× oil objective. For structural quantification, including NMJ synapse branch number, bouton number and area, preparations were double-labeled with anti-HRP and anti-DLG, with counts made at muscle 4 in segment A3. For nuclear import studies, nuclei were identified by propidium iodide staining with fluorescent punctae counted and intensity quantified [59]. For synaptic functional protein quantitation, glutamate receptor and Brp punctae were quantified for muscle 4, segment 3. Glutamate receptor number and field area was quantified in consecutive boutons of >3 µm diameter. All preparations were fixed, stained and processed simultaneously to allow for intensity comparisons. All analyses were done with ImageJ software (National Institutes of Health) using the threshold function to outline areas and Z-stacks made using the maximum projection function. Statistics were done either with one-way ANOVA analysis followed by Dunnett's post-test, student's t-test or Mann-Whitney test for non-parametric data. All analyses were done blind to genotypes during all stages of experimentation and analysis. All figure images were projected in LSM Image Examiner (Zeiss) and exported to Adobe photoshop.
Stock solution of heparin (Sigma, H3393) in 1×PBS was prepared and serially diluted to obtain concentrations (e.g. 0.625, 0.315 and 0.156 mg/ml). Dissected wandering third instar larvae were incubated with these heparin concentrations for 5 minutes at RT, followed by a 1 minute wash with 1×PBS and then 10 minute fix with 4% paraformaldehyde in 1×PBS. After fixation, anti-Wg or anti-Gbb antibodies were used as above with appropriate secondary antibodies. Processed animals were analyzed for changes in intensity measurements as above in the image quantification section. All fluorescence intensity measurements were compared to preparations treated identically with only 1×PBS and no heparin, and the processed simultaneously for immunolabeling, microscopy and quantification.
Two-electrode voltage-clamp (TEVC) records were made from the wandering third instar NMJ as previously described [41]. In brief, staged control, mutant and transgenic RNAi animals were secured on sylgard-coated coverslips with surgical glue (liquid suture), dissected longitudinally along the dorsal midline, and glued flat. The segmental nerves were cut near the base of the ventral nerve cord. Recording was performed in 128 mM NaCl, 2 mM KCl, 4 mM MgCl2, 1.0 mM CaCl2, 70 mM sucrose, and 5 mM Hepes. Recording electrodes (1-mm outer diameter capillaries; World Precision Instruments) were filled with 3 M KCl and had resistances of >15 MΩ. Spontaneous mEJCs were collected using continuous (gap-free) recording and evoked EJC recordings were made from the voltage-clamped (Vhold = −60 mV) muscle 6 in segment A3 with a TEVC amplifier (Axoclamp 200B; MDS Analytical Technologies). The cut segmental nerve was stimulated with a glass suction electrode at a suprathreshold voltage level (50% above baseline threshold value) for a duration of 0.5 ms. Records were made with 0.2 Hz nerve stimulation in episodic acquisition setting and analyzed with Clampex software (version 7.0; Axon Instruments). Each n = 1 represents a recording from a different animal. Statistical comparisons were performed using student's t-test or the Mann-Whitney test for non-parametric data.
|
10.1371/journal.ppat.1001082 | Generation of Covalently Closed Circular DNA of Hepatitis B Viruses via Intracellular Recycling Is Regulated in a Virus Specific Manner | Persistence of hepatitis B virus (HBV) infection requires covalently closed circular (ccc)DNA formation and amplification, which can occur via intracellular recycling of the viral polymerase-linked relaxed circular (rc) DNA genomes present in virions. Here we reveal a fundamental difference between HBV and the related duck hepatitis B virus (DHBV) in the recycling mechanism. Direct comparison of HBV and DHBV cccDNA amplification in cross-species transfection experiments showed that, in the same human cell background, DHBV but not HBV rcDNA converts efficiently into cccDNA. By characterizing the distinct forms of HBV and DHBV rcDNA accumulating in the cells we find that nuclear import, complete versus partial release from the capsid and complete versus partial removal of the covalently bound polymerase contribute to limiting HBV cccDNA formation; particularly, we identify genome region-selectively opened nuclear capsids as a putative novel HBV uncoating intermediate. However, the presence in the nucleus of around 40% of completely uncoated rcDNA that lacks most if not all of the covalently bound protein strongly suggests a major block further downstream that operates in the HBV but not DHBV recycling pathway. In summary, our results uncover an unexpected contribution of the virus to cccDNA formation that might help to better understand the persistence of HBV infection. Moreover, efficient DHBV cccDNA formation in human hepatoma cells should greatly facilitate experimental identification, and possibly inhibition, of the human cell factors involved in the process.
| Persistent infection with hepatitis B virus (HBV) causes chronic hepatitis B which frequently progresses to hepatocellular carcinoma, a leading cause of cancer-mediated mortality worldwide. Persistence requires formation and amplification of covalently closed circular (ccc)DNA, an episomal form of the viral genome that is not targeted by current drugs and thus is responsible for the notorious difficulties in therapeutic elimination of infection. Initial generation of cccDNA occurs upon nuclear import of the virion-borne relaxed circular (rc) DNA to which the viral polymerase is covalently linked; amplification occurs via intracellular recycling. The underlying molecular pathway is poorly understood. Because HBV infects only primates, in vivo studies are extremely restricted; in vitro, select hepatoma cell lines transfected with HBV support viral replication, however with little if any cccDNA formation. Here, we compared intracellular recycling of HBV and DHBV, a model hepatitis B virus from ducks, in cross-species transfections. Surprisingly, the major contribution to cccDNA formation comes from the virus rather than the cell as DHBV but not HBV rcDNA converted efficiently into cccDNA in the same human cell background. This unexpected difference might help to better understand persistence of HBV infection; efficient DHBV cccDNA formation in human cells provides a new tool to facilitate identification, and possibly targeting, of the human cell factors involved.
| Currently, more than 350 million people suffer from chronic HBV infection. Chronic hepatitis B frequently progresses to liver cirrhosis and hepatocellular carcinoma, a leading cause of cancer-related morbidity and mortality worldwide [1], [2].
HBV is a small enveloped hepatotropic DNA virus which replicates by reverse transcription of an RNA intermediate, the pregenomic (pg) RNA (for review: [3], [4]), to yield encapsidated, partially double-stranded rcDNA to which the viral polymerase is covalently bound [5]. Upon infection, rcDNA is transported to the host cell nucleus where it is converted into cccDNA (Figure S1). Episomal cccDNA then acts as template for all viral transcripts. These include the subgenomic RNAs encoding the surface proteins, and the pgRNA that serves as mRNA for the polymerase protein and the capsid, or core, protein. Binding of polymerase to the RNA stem-loop structure ε initiates packaging of one pgRNA molecule per newly forming capsid and its reverse transcription. The first product is single-stranded (ss) DNA of minus polarity; due to the unique protein-priming mechanism, its 5′ end is, and remains, covalently linked to the polymerase. The pgRNA is concomitantly degraded, except for its 5′ terminal ∼15–18 nucleotides which serve as primer for plus-strand DNA synthesis, resulting in rcDNA and, as a side-product, some double-stranded linear (dl) DNA. DNA-containing capsids are then enveloped by surface proteins and cellular lipids and secreted as virions. Alternatively, they are redirected to the nucleus to increase cccDNA copy number by a mechanism termed intracellular recycling; many estimates for cccDNA copies per infected hepatocyte are in the range of 5 to 50 [6]. This amplification prevents loss during cell division of the cccDNA which can not be replicated semiconservatively [7]. Thus, cccDNA formation and recycling are central to establish and maintain persistent infection, and they limit the efficacy of antiviral nucleot(s)ides in the treatment of chronic hepatitis B, as these do not directly target cccDNA (for review: [8]).
However, despite its central importance the molecular pathway driving the conversion of HBV rcDNA to cccDNA is poorly understood. In vivo studies face numerous challenges. Liver biopsies from human patients are scarce, affected by the natural history of infection, and they sample only a small volume of the liver; recent estimates for cccDNA content in infected human liver vary from 0.01 to 1.4 [9] to 0.1 to 10 copies per hepatocyte [10], with large patient-to-patient variation. Kinetic studies during acute infection, doable exclusively in chimpanzees, showed peak values of around 10 cccDNA copies per infected cell but much lower numbers before and after [11]. In well controllable experimental settings, on the other hand, such as transfected cell lines [12]–[14] or HBV-transgenic mice [15], [16], HBV produces little if any cccDNA.
DHBV is a related animal virus that is widely used as a model to study HBV infection [17]. Like HBV, DHBV produces cccDNA in cultured duck hepatocytes and in vivo [7], [18], with reported mean values of 2.9 to 8.6 copies per hepatocyte [19] though also with temporal and cell-to-cell fluctuations (from one to >36 copies per cell). Importantly, DHBV transfected into the chicken hepatoma line LMH also generates well detectable amounts of cccDNA. Initially in primary duck hepatocytes, then using that system, Summers and colleagues had first shown that knocking out surface protein expression, and thus virion secretion, dramatically increased cccDNA copy numbers per nucleus to about 200–400 molecules [20], [21]. Recent studies in stably or transiently transfected human cell lines suggest that preventing HBV surface protein expression also stimulates cccDNA formation, but to a much lower degree; instead, a “protein-free” form of rcDNA (pf-rcDNA) accumulated [12], [13]. The term protein-free (which we will adhere to here) was operationally defined by the partitioning of this rcDNA form into the aqueous phase upon phenol extraction without prior proteinase K (PK) treatment (which artificially degrades the protein); polymerase-linked DNA partitions into the organic phase. “Protein-free” does therefore not imply the complete absence of any amino acid from the DNA. Though not finally proven, several lines of indirect evidence suggest that pf-rcDNA is a precursor to cccDNA [12], [13]; not the least, removal of the bound polymerase is a sine qua non for cccDNA formation. Caveats are that nicked cccDNA, generated to some extent during preparation and naturally protein-free, has the same electrophoretic mobility as pf-rcDNA. Furthermore, Southern blots from infected liver nuclei have generally shown only little rcDNA versus cccDNA [22]. However, in such samples one usually looks at an established pool of cccDNA whereas the recent cell culture studies monitored initial cccDNA formation. Potential precursors accumulating under these conditions may not be detectable anymore in in vivo samples.
At any rate, the initial presence of protein-bound rcDNA inside virions and eventually of nuclear cccDNA, associated with histones [23], requires as intermediate steps nuclear transport of the rcDNA genome, its release from the viral capsid and removal of the bound polymerase to allow generation of precisely one genome length equivalents of the plus- and the minus-strand before final ligation into cccDNA (Figure S1B). The order of events is not firmly established. Intact nucleocapsids may deliver the protein-bound rcDNA to the nucleus where its release from the capsid and polymerase removal are mediated by host factors; this view is supported by the minimalistic genomes of hepadnaviruses (only ∼3 kb) and by data from nuclear transport model systems [24], [25]. Alternatively, the nucleocapsid itself may contain corresponding activities such that polymerase removal could precede capsid release, possibly already in the cytoplasm. Some evidence in favor of this view has recently been forwarded [13], [26].
Apart from these mechanistic aspects it appears, in essence, that DHBV in the avian LMH cells produces much more cccDNA than HBV in human hepatoma cells. One conceivable explanation are cell-specific differences. For instance, the routinely used human HepG2 and Huh7 hepatoma cell lines may lack enzymatic activities required for cccDNA formation that are present in the avian cells. Alternatively, the different efficiencies in cccDNA formation may be a feature of the respective viruses.
In order to address this question we took advantage of the principal ability of hepadnaviruses to replicate in hepatoma cell lines of heterologous species origin. After transfection HBV is capable of producing rcDNA in LMH cells and the same holds for DHBV in HepG2 and HuH7 cells [27], [28]. However, cccDNA formation in such cross-species transfections has not yet been addressed. Here we performed such experiments and found that, unexpectedly, the major contribution to cccDNA formation comes from the virus rather than from the cell.
Detection of cccDNA by Southern blotting can severely be hampered by the presence of ssDNA species which have a similar electrophoretic mobility and are often present in excess. A further problem in transient transfections is the highly abundant plasmid DNA. We therefore developed an assay that essentially eliminates ssDNA and plasmid DNA by double-digestion with the restriction enzyme Dpn I and Plasmid safe DNase (PsD). Dpn I requires bacterially methylated DNA to be active and selectively restricts the transfected plasmid. PsD digests single-stranded (ss) and double-stranded linear (dl) but not circular molecules such as cccDNA. Although it has been surmised that rcDNA is a substrate for PsD [9], our own preliminary data suggested this holds only for very immature rcDNA forms. To enhance cccDNA production, we used plasmids coding for surface-deficient HBV and DHBV. Because cccDNA is enriched in the nucleus, we separated nuclei from cytoplasm by treating the cells with the mild detergent NP-40 and subsequent centrifugation. To identify encapsidated DNAs, we incubated the cytoplasmic extracts with micrococcal nuclease (MN) which digests free nucleic acids but not those protected inside capsids. Finally, because protein-bound DNA is neither recovered upon phenol extraction nor using the silica column adsorption (QIAamp) method employed here, all initial DNA preparations included a PK treatment.
The results for DHBV in LMH and HBV in HepG2 cells are shown in Figure 1. Treatment of the cytoplasmic samples with MN revealed the common replicative intermediates, i.e. rcDNA, dlDNA and ssDNA. Digestion with Dpn I only produced a similar pattern, except that additional plasmid-derived bands (Pla) were visible. Nuclear DNA treated with Dpn I alone produced a similar pattern, yet as expected, knock-out of surface protein expression enhanced the cccDNA signal, particularly for DHBV. Additional treatment of nuclear DNA with PsD removed all bands except those at the rcDNA and cccDNA positions; the equal signal intensities before and after PsD treatment demonstrated that mature rcDNA was not appreciably attacked by PsD. For HBV, a band at the cccDNA position was exclusively visible in the surface-deficient background (Figure 1B). However, the nuclear rcDNA signal was much more enhanced, in line with recent reports [12], [13]. Comparable results were obtained in Huh7 cells (Figure S2A). Together, these data demonstrated that the applied procedure enabled the reliable detection of cccDNA and rcDNA, without interference from other virus- or plasmid-derived nucleic acids. Furthermore, they confirmed that DHBV in avian cells produces much more cccDNA than HBV in human cells.
Faint bands with cccDNA-like mobility were also detectable in the cytoplasmic fractions of those samples containing well visible nuclear cccDNA; ethidium bromide staining of the agarose gel used to generate the blot revealed indeed some chromosomal DNA in the cytoplasmic samples, indicating an incomplete separation of two fractions. This prompted us to employ a more efficient separation procedure in later experiments (see below).
Next we performed analogous experiments with DHBV transfected into the human cell lines, and HBV transfected into LMH cells (Figure 2). For DHBV, cytoplasmic extracts treated with MN produced a similar pattern of replicative intermediates in HepG2 (Figure 2A) and HuH7 cells (Figure S2B) as in LMH cells (Figure 1A). HBV in LMH cells, compared to the human cell lines, generated a more complex pattern with a distinct band of an intermediate mobility between rcDNA and ssDNA (Figure 2B). This additional band originates from strongly enhanced splicing of the HBV pgRNA in the chicken cell line (Köck, Nassal, Thoma; unpublished data). The spliced genomes are of linear conformation and therefore accessible to PsD digestion. Most importantly for the current study, the envelope-deficient DHBV produced a strong cccDNA signal in both human cell lines (Figure 2A and Figure S2B), with an intensity equaling that of the nuclear rcDNA as in LMH cells. Conversely, HBV in LMH cells generated a similar pattern as in the human cell lines, with a relatively strong band at the rcDNA yet only a weak band at the cccDNA position (Figure 2B). The HBV specific lack of effect of the envelope knock-out in LMH cells correlated with a much lower abundance of the surface protein-coding mRNAs compared to the human cells (data not shown). Thus wild-type HBV in LMH cells is phenotypically similar to its envelope-deficient counterpart.
Together, these data demonstrated that both HepG2 and HuH7 cells are competent to support cccDNA synthesis, yet with strikingly higher efficiency for DHBV than HBV; conversely, LMH cells did not support more efficient cccDNA formation for HBV. These results were reproduced in about twenty independent experiments. Furthermore, this virus-specific difference was neither changed by increasing pgRNA levels via replacement of the HBV core promoter by the CMV promoter, nor by analyzing the cultures at earlier or later time points post transfection (Figure S3). Thus the virus contributes more profoundly to the efficiency of cccDNA biosynthesis than the cell. More quantitative data on the amounts and intracellular distribution of the different viral DNA were obtained with samples from gradient purified nuclei.
Beyond the markedly different levels of cccDNA, the data shown above also revealed distinct levels of nuclear rcDNA, and of the ratios of cccDNA to rcDNA, between DHBV and HBV. A detailed characterization of the nuclear rcDNA was expected to provide clues on the rate-limiting step of HBV cccDNA synthesis. This, however, required the nuclear preparations to be as free from cytoplasmic contamination as possible. We therefore separated the nuclei from the cytoplasm by sucrose density sedimentation [29]. Western blot analyses of nuclear extracts versus total cell lysates showed that, while nuclear histone H3 was detected in either fraction, the cytoplasmic poly(A) binding protein (PABP) was exclusively present in total cell lysates but not in nuclear extracts (Figure S4A). Co-purification of cytoplasmic capsids with the nuclei was ruled out by the absence of viral DNAs from purified nuclei of non-transfected HepG2 cells that had been mixed with capsid-containing cytoplasmic fractions from HBV- or DHBV-transfected cells and subjected to the same procedure (Figure S4B). Permeability of the nuclei for exogenously added MN, an assay subsequently used to address nuclease sensitivity vs. resistance of the nuclear viral DNAs, was confirmed by dose-dependent generation of chromosomal DNA fragments with sizes of multiples of ∼150 bp, as expected from cleavage between nucleosomes (Figure S4C).
Applying this methodology to HepG2 cells transfected with surface-deficient HBV or DHBV constructs revealed that the purified nuclei contained easily detectable signals at the rcDNA position, and for DHBV a strong and for HBV a weak signal at the cccDNA position; as expected, both converged into a single species with dlDNA mobility upon digestion with Eco RI, and heating converted the rcDNA but not the cccDNA signal into a new band with ssDNA mobility (Figure S5A). However, this assay does not discriminate true rcDNA from randomly nicked cccDNA. For distinction, we exploited the defined discontinuities at the 5′ ends of the complete minus-strand and the 3′ terminally incomplete plus-strand DNA (Figure 3A) in rcDNA and the requirement of most restriction enzymes for a double-stranded substrate structure. The nuclear, and for comparison also the cytoplasmic, HBV DNA preparations were incubated with Nco I whose recognition site (CCATGG; nt positions 2654–2659) is in the 3′ proximal part of the plus-strand but largely double-stranded already in virion DNA (e.g. [30]); Fsp I (TGCGCA; nt positions 3082–3087), recognizing a site in the 5′ proximal plus-strand region that is double-stranded even in DNA with very short plus-strands; and Apa LI (nt positions 2861–2866), with its recognition site (GTGCAC) ending only 5 nt upstream of DR2 (nt positions 2872–2882) where the plus-strand begins. Nicked cccDNA should be linearized by all three enzymes whereas true rcDNA, depending on how far the plus-strand is filled-in, is expected to be partially or completely resistant to Apa LI cleavage. Exactly this was observed, with about 35% of the rcDNA remaining in the Apa LI but not the Nco I and Fsp I treated samples from both the cytoplasm and the nucleus; very similar results were obtained in repeat experiments (42.2±10.7% for the nuclear and 40.0±2% for the cytoplasmic rcDNA), as well as for protein-free nuclear rcDNA (32.0±4.9%; Figure S5B). The differences between different treatments were highly significant (P<0.05 to <0.001) but those between cytoplasmic vs. nuclear rcDNA were not. Activity of Apa LI in the reactions was demonstrated by the disappearance of the Dpn I fragment from the transfected plasmid DNA that harbors the single virus genome-encoded recognition sites for Apa LI and Fsp I but not Nco I. Furthermore, an admixed DHBV plasmid was completely cut by either enzyme (Figure S5B). Hence, in accord with previous reports [12], [13], a substantial fraction of the nuclear HBV rcDNA signal was derived from true rcDNA, confirming that the weak cccDNA signal was not caused by excessive nicking.
For a quantitative estimate of the proportion of nuclear versus cytoplasmic HBV and DHBV rcDNA, we compared the amounts of viral DNA in total cells (nuclei plus cytoplasm) and in gradient-purified nuclei; to account for all DNA species regardless of encapsidation and protein-linkage status, preparations involved PK treatment and subsequent digestion with Dpn I plus PsD, but not MN. Serial dilutions served to improve the accuracy of quantitation by phosphorimaging; one of three independent experiments used for quantitation is shown in Figure 4A. The signals from the four different amounts of total rcDNA (1%, 3%, 10%, and 30% of the whole preparation) and from the 30% portions of nuclear rcDNA were quantitated and corrected for background by phosphorimaging. Accordingly, the ratios for total vs. nuclear rcDNA were 3.6±0.44 : 1 for DHBV, and 6.8±0.26 : 1 for HBV, i.e. about 25–30% of the DHBV and around 15% of the HBV rcDNA were nuclear.
Using the known amounts of HBV and DHBV marker DNAs run on the same gels we also estimated the amounts per sample (one well of a 6-well plate) of total versus nuclear rcDNA and cccDNA, and the copy numbers per transfected cell (see Text S1 for details); accordingly, DHBV generated ∼160 pg (161.6±28.3; 200±35 copies) total rcDNA of which ∼45 pg (44.8±4.5; 56±6 copies) were nuclear, and HBV produced ∼130 pg (130.2±7.6; 162±10 copies) total rcDNA of which ∼20 pg (19.2±1.1; 25±2 copies) were nuclear. Amounts of cccDNA estimated analogously were ∼45 pg (42.3±3.3; 56±5 copies) for DHBV, and ∼2 pg (2.1±0.9; 2.6±1.1 copies) for HBV; cccDNA values for HBV were close to background and therefore difficult to determine more accurately. However, a similar excess of nuclear HBV rcDNA over cccDNA has also been reported for inducible cell lines stably transfected with surface-deficient HBV [12], [13]. Together these data indicated that less efficient nuclear transport of HBV rcDNA contributes to, but cannot solely be responsible for, the much less efficient cccDNA accumulation.
To further investigate the status of the viral DNAs, we treated total lysates and purified nuclei with MN followed by PK, leaving only encapsidated (or otherwise protected) DNA species intact. For DHBV, the results were essentially the same (Figure S6, panel DHBV) as those after Dpn I plus PsD treatment (Figure 4A) but for HBV the signal at the full-length rcDNA position had nearly disappeared, apparently in favor of faster migrating species (Figure S6, panel HBV). For a direct estimate of the fraction of MN sensitive nuclear DNA species, viral DNA was isolated from equally sized aliquots of the nuclei by either the Dpn I plus PsD or the MN procedure and analyzed side-by-side (Figure 4B). For DHBV the total amounts of rcDNA obtained in either way were very similar, indicating that most of the nuclear DHBV DNA was present in largely intact nucleocapsids (as confirmed by anti-capsid immunoprecipitation; see below). The nonetheless strong cccDNA signal suggested that once released from the capsid, DHBV rcDNA gets rapidly converted into cccDNA; alternatively, the stably encapsidated nuclear DHBV rcDNA might be a dead-end product.
For HBV, in contrast, only the cytoplasmic rcDNA signal was largely resistant to MN whereas in the nuclear sample faster migrating species accumulated (Figure 4B and Figure S6); quantitative comparison of the nuclear full-length rcDNA signals with versus without MN treatment from three independent transfections showed that only about 10% (11.1±6.3%) of the full-length rcDNA was resistant. Hence different from DHBV, most of the nuclear HBV rcDNA was sensitive to MN and therefore no more protected by an intact capsid shell.
The nuclease sensitivity of most of the nuclear HBV full-length rcDNA was compatible with its complete release from the capsid; the faster-migrating nuclease resistant species might then represent shorter DNAs that were still fully encapsidated. Alternatively, partial opening of the capsid could have exposed parts of, but not the entire full-length DNA to nuclease attack; finally, partial protection could also have arisen from association with factors other than core protein.
We therefore assessed whether the nuclear viral DNAs could be immunoprecipitated by antibodies against the respective core proteins. Purely osmotic procedures released much less core protein from the gradient-purified nuclei than lysis with 0.5% SDS which, however, disintegrates capsids and prevents analysis of capsid-associated nucleic acids. Instead we treated the nuclei with 0.75× radioimmunoprecipitation (RIPA) buffer which destroys the nuclear envelope but leaves the viral nucleocapsids intact [31]. The cytoplasmic lysates serving as reference source for the immunoprecipitations (IPs) were likewise adjusted to 0.75× RIPA. As a specificity control, HBV samples were incubated with anti-DHBV core antibody and vice versa (mock-IP). Next, DNAs associated with the immunopellets were isolated after prior PK treatment and analyzed by Southern blotting.
For DHBV (Figure 5A), the immunoprecipitated DNA from both the cytoplasm and the nuclei, if treated with only Dpn I, generated a similar pattern as that obtained by direct incubation of the extracts with MN (lanes ø), except it contained some fragmented plasmid DNA; this was found in similar amounts in the mock-IP and represented at most 50 pg, i.e. a tiny fraction of the transfected plasmid DNA. MN treatment of the immunopellets completely removed the residual plasmid DNA, as well as a faint band at the cccDNA position seen only in the nuclear immmunopellets, but left most of the viral rcDNA and dlDNA intact. Hence the non-ccc forms of DHBV DNA in the cytoplasm and nucleus behaved alike: both were immunoprecipitable with anti-DHBV core antibody and both were largely protected from nuclease, consistent with their being stably encapsidated in either compartment.
For HBV (Figure 5B), the picture in the cytoplasmic samples was similar; the anti-HBV core antibody precipitated the same type of DNAs as obtained by direct MN treatment; residual plasmid DNA was completely removed by MN. The nuclear immunopellet, if treated with Dpn I only, generated a similar pattern as that from the cytoplasm; hence at least part of the nuclear full-length rcDNA was still core protein-associated. However, like direct incubation of the nuclei with MN (Figure 4B), MN treatment strongly reduced the full-length rcDNA signal whereas faster migrating DNA species accumulated, suggesting their resistance was indeed due to protection by the capsid. Comparable results were obtained in Huh7 (Figure 5B, left panel) and in HepG2 cells (Figure 5B, right panel; cytoplasmic samples of this experiment are shown in Figure S7). Together these data indicated that at least a fraction of the nuclear HBV DNA including full-length species was still associated with core protein but, different from DHBV, was no more fully protected.
The majority of MN resistant nuclear HBV DNAs migrated reproducibly to a region whose upper boundary would correspond to dlDNA of ∼2.4–2.7 kb (Figure 4B). One explanation was that these molecules represented naturally shorter DNA derived from spliced pgRNA [32], [33] which remained protected in intact nucleocapsids. However, a construct encoding a variant HBV genome in which the major splice acceptor site was mutated generated exactly the same pattern of MN resistant nuclear DNA although splicing was indeed suppressed (data not shown). Next we incubated the MN resistant nuclear, and for comparison also cytoplasmic, HBV DNA with restriction enzymes Nco I and Spe I which cut the HBV genome uniquely at positions 2654 and 1961 (Figure 6A, C). Expectedly, the major effect on the cytoplasmic DNA was conversion of the rcDNA to dlDNA, plus generation of small amounts of fragments that most likely derived from dlDNA; ssDNA was not affected. In the nuclear sample, the little full-length rcDNA was linearized as well. The major faster migrating species disappeared completely upon exposure to either enzyme, indicating they contained both recognition sites in double-stranded form. Intriguingly, within the background smear distinct fragments of about 2.2 kb (Nco I) and 1.2 kb (Spe I) appeared that were absent from the non-restricted sample, suggesting that MN had generated at least one relatively distinct new DNA end at a fixed distance from the restriction sites. One interpretation was that the MN products of rcDNA lacked about 500 bp roughly between position 3000 and 500 (see map in Figure 6C); in that case the second Nco I fragment would comprise only around 300 to 400 bp which, together with some heterogeneity in size, would make it difficult to detect. The 1.2 kb band in the Spe I treated sample, conversely, could consist of two about equally sized products. To corroborate this assumption, we digested the nuclear MN resistant DNAs with three more single cutter enzymes (Figure 6B): Nsi I and Eco RI (recognition sites starting at positions 2346 and 1280, respectively), and Bsp EI (recognition site starting at position 429 in the predicted lacking sequence part). As a further control, intact nuclear rcDNA was isolated by the Dpn I plus PsD procedure and digested with the same enzymes which in all cases led to complete linearization (Figure 6B, left panel). In the MN treated nuclear DNA, the rcDNA signal likewise disappeared completely in favor of a new band at the dlDNA position. Importantly, however, Nsi I and Eco RI produced distinct new bands of about 1.9 kb (marked by *), whereas for Bsp EI the pattern below the dlDNA position was not detectably different from that of the untreated sample (lanes ø). These data are consistent with MN generating from nuclear HBV rcDNA a relatively distinct mixture of double-stranded linear DNAs that lack approximately the region between position 3000 and 500 (Figure 6C). This, in turn, implies that capsid opening occurs at distinct sites relative to the packaged genome.
Together, these data indicated that nuclear HBV genome release from the capsid is efficiently initiated; at most 10% of the full-length rcDNA remained resistant to MN. Furthermore, as shown below, we also found evidence for a fraction of nuclear rcDNA that is completely released from the core protein. Hence poor HBV rcDNA to cccDNA conversion is not caused by a complete block of uncoating dynamics.
To test the polymerase linkage status of the different viral DNAs we prepared cytoplasmic extracts and purified nuclei from DHBV and HBV transfected HepG2 cells. Equal aliquots from each sample were then incubated in buffer containing SDS with or without PK, and subjected to conventional phenol extraction. Transfected plasmid DNAs were digested with Dpn I (Figure 7A). Without PK, only very weak signals were seen for both HBV and DHBV in the cytoplasmic fractions (<10% of the signals with PK), in line with a previous report [12] though not with an other [13]; evaluation of 15 independent HBV samples gave a mean value of 9.2±2.4%.
In the nuclei of the DHBV transfected cells, PK treatment had, expectedly, little effect on cccDNA and the plasmid derived Dpn I fragments (Figure 7A, left panel); but strongly enhanced the rcDNA signal. For nuclear HBV rcDNA, in contrast, the signals obtained without PK were nearly as strong as those from the PK treated aliquot (Figure 7A, right panel). For a semiquantitative estimate we compared the rcDNA signal intensities with versus without PK treatment from two independent experiments; to account for differencies in recovery, these values were normalized for the cccDNA signals (DHBV) and the plasmid fragments (HBV), respectively, in the same lanes. Accordingly, the rcDNA signals from the PK treated samples were 3.78±0.54 fold stronger for DHBV, and 1.17±0.29 fold stronger for HBV, indicating that around 20% of the nuclear DHBV rcDNA and around 70% or more of the nuclear HBV rcDNA were protein-free. Comparing 11 different pairs of PK treated vs. non-treated nuclear HBV samples yielded a mean value of 66.9±13.3% protein-free rcDNA.
That a major fraction of the rcDNA signals originated from true rcDNA was corroborated by subjecting nuclear HBV DNA obtained with or without prior PK treatment to digestion with Nco I, Apa LI, and FspI. As before (Figure 3), Nco I and Fsp I converted the PK treated rcDNA nearly completely into dlDNA, whereas about 40% of the rcDNA signal remained upon Apa LI digestion (Figure S5B); for the protein-free nuclear DNA the mean value was slightly lower (32±5%), suggesting it contained more completely filled-in plus-strand DNA. However, this difference was not statistically significant.
We also assessed the polymerase linkage status of the MN resistant DNAs (Figure 7B). The patterns generated upon prior PK treatment fully matched those previously seen (Figure 4B, C, S6) whereas without PK treatment only faint signals were detectable. For DHBV, the nuclear and cytoplasmic DHBV DNAs were similar in composition and both were largely polymerase-linked. With, but not without PK treatment, nuclear HBV DNA showed the same accumulation of faster migrating species as before (Figure 4B). Hence the MN resistant shorter HBV DNA species as well as the small amounts of resistant full-length rcDNA were mostly protein-linked. This suggested that the relatively high proportion of protein-free species in the total nuclear DNA (Figure 7A) was largely accounted for by MN sensitive rcDNA molecules.
The IP experiments described above (Figure 5B) indicated an association of at least some of the nuclear HBV full-length rcDNA with core protein. Hence failure to completely uncoat the DNA could represent a rate-limiting step in cccDNA formation. We therefore assessed whether the nuclei also contained rcDNA molecules that were no more associated with core protein. To this end, we performed IPs as before yet this time we included the IP supernatants in the analysis. Furthermore, we addressed the polymerase linkage status of such putative species by treating one half of each sample with PK, the other not. Cytoplasmic samples in which the viral DNA is largely present in intact capsids (Figure 5) served as control. Because MN would destroy nonprotected rcDNA, we used Dpn I plus PsD to reduce the background of non-rcDNAs.
The specific IP from the cytoplasm generated a strong rcDNA signal in the immunopellet after PK treatment; the signal from the supernatant was only 3–4% as intense (Figure 8, lane 2 vs. 4). Conversely, the signal from the mock IP pellet had less than 3% the intensity of that from the supernatant (lane 6 vs. 8). These data confirmed that the IP was specific and that the amount of anti-HBc antibody was sufficient to precipitate ≥95% of the cytoplasmic rcDNA. The strongly reduced signals without PK treatment further corroborated that 90% or more of the cytoplasmic rcDNA was protein-linked.
A different picture was seen in the nuclear samples. The anti-HBc supernatant contained more rcDNA than the immunopellet (lane 11 vs. 13); quantification indicated that only about 30% of the rcDNA was precipitated (29.8±3.5% for the PK treated, 34.2±3.2% for the not treated sample; from duplicate determinations of two independent experiments). For the mock IP, only 5.0±3.9% were found in the pellet (lane 15 vs. 17). In further contrast to the cytoplasmic samples, yet in line with the previous data (Figure 7A), omitting the PK treatment only modestly reduced the signal intensities in all nuclear samples (lanes 11 vs. 12; 13 vs. 14; 17 vs. 18), i.e. to 73±16% in the specific immunopellet, to 69±9% in the supernatant, and to 55±7% in the mock IP supernatant. Although the low overall signal intensities precluded a more accurate determination, these data indicated that around two thirds of the nuclear HBV rcDNA were not stably associated with core protein, i.e. probably completely uncoated, and around two thirds of these molecules were no more linked to intact polymerase protein.
The data described above were compatible with at least partial release of the nuclear HBV rcDNA from the capsid, allowing cell factors to engage in polymerase removal. A recently proposed alternative is that polymerase removal might precede capsid opening, possibly already in the cytoplasm and mediated by capsid-intrinsic activities [26]. One line of evidence in favor of this proposal was the reported generation of small amounts of protein-free DNA in detergent-stripped DHB virions subjected to prolonged endogenous polymerase reaction (EPR) conditions; no results for HB virions were reported. In the EPR, exogenously added dNTPs are utilized by the capsid-borne (“endogenous”) polymerase to complete the plus-strand DNA. Because serum virions are secreted from infected hepatocytes, there is very little risk of cross-contamination with nuclear or other cellular factors.
Here we performed analogous experiments with HB virions, and for comparison with the published results [26] also with DHB virions. Nucleocapsids of either virus were obtained from highly viremic sera by sedimentation in Nycodenz gradients containing NP-40 detergent. DHBV nucleocapsids were exposed to EPR conditions for 16 h [26], and the capsid-borne DNAs were isolated via phenol extraction with or without prior PK treatment. The corresponding Southern blot showed ample rcDNA and some dlDNA in the PK treated but not the untreated sample (Figure 9A, left panel), confirming that most of the virion-borne genomes are covalently linked to polymerase. In line with the reported data, a long exposure plus contrast enhancement (Figure 9A, right panel) revealed indeed a faint band of apparently protein-free DNA, however exclusively at the dlDNA, not the rcDNA position. By comparison with the dilution series of the PK treated sample, the protein-free dlDNA accounted for less than 0.3% of the total capsid-borne virus DNA.
Analysis of the HB virion-derived nucleocapsids after PK treatment but prior to the EPR (Figure 9B) showed a smear of bands that migrated faster than mature rcDNA from transfected HepG2 cells, indicating a relatively large gap in the HB virion plus-strand DNA; no signal was visible without PK treatment (Figure 9B). When subjected to EPR (Figure 9C, lanes +dNTP) under the same conditions as used for DHBV, these species were efficiently converted into rcDNA plus some dlDNA, confirming enzymatic activity of the capsid-borne polymerase. The signal was largely stable towards MN, indicating the vast majority of viral genomes remained protected by the capsid shell. Without PK treatment, no signal was visible, not even upon overexposure (Figure 9C, right panel). Thus if plus-strand completion per se caused any release from the capsid or removal of polymerase from viral DNA, its extent for HBV was even less pronounced than for DHBV. That, in contrast, in the nuclei of the transfected cells most of the HBV rcDNA was sensitive to MN and protein-free (Figure 4, 5) strongly supports that HBV capsid opening and deproteinization of rcDNA depend largely on the nuclear environment.
Formation and amplification of cccDNA is essential for the establishment and maintenance of HBV infection. The underlying molecular pathway is poorly understood, not the least because HBV in transfected human hepatoma cell lines, despite efficient replication, produces very little cccDNA even if surface protein expression is prevented. Unexpectedly, our cross species transfection experiments revealed that the very same cell lines support efficient conversion of DHBV rcDNA into cccDNA. Our characterization of the intranuclearly accumulating rcDNA species showed that initiation of nuclear uncoating of the HBV rcDNA was highly efficient whereas complete release from the capsid and complete removal of the covalently linked polymerase contribute to limiting cccDNA formation. However, ∼40% of the nuclear rcDNA were apparently fully capsid-released and polymerase-free, hence a major block lies in the actual rcDNA to cccDNA conversion process; for DHBV, such a block does not appear to exist (Figure 10).
The conclusion of efficient DHBV but poor HBV rcDNA to cccDNA conversion is based on the relative intensities of Southern blot signals at the rcDNA and cccDNA positions in nuclear preparations; a ratio of about 1∶1 for DHBV but 10–20∶1 for HBV was observed in more than 20 independent preparations. The selective partial resistance against Apa LI digestion of nuclear HBV rcDNA (Figure 3, Figure S5) strongly suggests that excessive nicking of HBV (but not DHBV) cccDNA is not a major cause for the strong rcDNA versus weak cccDNA signals, consistent with data from stably transfected cells [12], [13]. Another explanation would be that cccDNA of HBV has a much shorter half-life than that of DHBV. The in vivo half-life of cccDNA may be in the order of weeks to months [11], [34], [35] but the rates of synthesis versus degradation are subject to numerous factors, including cell division and immune responses. Closest to our setting are values derived from inducible cell lines in which rcDNA resynthesis was blocked by nucleoside analogs, with estimates of >48 h for DHBV [36] and >10 to 24 d for HBV [37]. Both values are long compared to the time frame of our experiments and, if anything, HBV cccDNA seems to be more stable than DHBV cccDNA. Hence it appears justified to assume that the weak HBV cccDNA signals truly reflect a slow rcDNA to cccDNA conversion rate.
Formation of cccDNA must involve nuclear transport of the rcDNA, its release from the capsid and removal of the bound polymerase to allow for the subsequent generation of precisely unit-length plus-strand and minus-strand DNA and ligation of their ends.
Regarding transport, the fraction of nuclear versus total rcDNA was about 2-fold higher for DHBV than for HBV (Figure 4); assuming that one cccDNA molecule originates from one imported rcDNA molecule, and given the about equal amounts of both species in DHBV transfected nuclei (Figure 2, 4) this ratio increases to 4∶1. Hence a lower import efficiency of HBV rcDNA can explain only part of the lower cccDNA accumulation.
Irrespective of mechanistic details (see below), the imported rcDNA molecules must eventually be released from the capsid to become accessible for the late repair steps in cccDNA formation. Nearly all of the nuclear, yet little of the cytoplasmic, full-length HBV rcDNA was sensitive to MN (Figure 4, 5, 6), and thus no more protected by an intact capsid shell; hence initiation of uncoating was highly efficient. Curiously though, MN treatment did not induce complete rcDNA degradation; instead, species with intermediate electrophoretic mobility accumulated, suggesting the existence of a distinct uncoating intermediate (see below). Nuclear DHBV rcDNA, in contrast, was as resistant against MN as cytoplasmic rcDNA, implying genome release for DHBV but not for HBV as potentially overall rate-limiting in cccDNA formation.
Nuclear (save cccDNA) and cytoplasmic DNAs of both viruses occurred in core protein associated form (Figure 5). DNA of either virus immunoprecipitated from the cytoplasm was largely resistent against MN, consistent with stable encapsidation. For DHBV, this also held for the nuclear immunopellet whereas drastic differences between cytoplasmic and nuclear fractions were revealed for HBV. The anti-HBc antibody (a monoclonal antibody recognizing a linear epitope exposed on intact capsids yet also on denatured core protein; [38]) precipitated full-length rcDNA from either fraction, yet selectively the nuclear full-length rcDNA was highly sensitive to MN, though not over its entire length. Hence a fraction of the HBV rcDNA in the nucleus was still associated with core protein and this association was likely responsible for partial nuclease protection. Together, these data indicate that the HBV nucleocapsid structure is drastically altered upon nuclear import such that the packaged rcDNA is at least partially exposed; the surprisingly distinct nature of the resulting rcDNA - core protein complexes is discussed below.
Failure of all nuclear HBV rcDNA molecules to be completely released from core protein could have represented the major rate-limiting step in cccDNA formation. However, about two thirds of the nuclear full-length rcDNA could not be immunoprecipitated (Figure 9) although the same amount of antibody precipitated ≥95% of the much more abundant cytoplasmic rcDNA. Hence partial as opposed to complete capsid release contributes to poor HBV rcDNA to cccDNA conversion.
Cytoplasmic DNA of either virus not treated with PK generated signals that were less than 10% as intense as those obtained after artificial deproteinization (Figure 7, 8, S5, S7); even lower values were reported by Gao and Hu [12]. We can not exclude that these apparently protein-free DNAs are present in genuinely cytoplasmic nucleocapsids [26] but they could as well arise from molecules that during work-up have been subject to fortuitous proteolysis or leakage from ruptured nuclei, as supported by the inability of virion-derived nucleocapsids to generate any significant amount of protein-free rcDNA (see below). Very obvious was, in contrast, the high proportion of protein-free HBV rcDNA in the purified nuclei. Consistently about two thirds were recovered without PK treatment from various samples (Figure 7, 8, S5). The low abundance of nuclear HBV rcDNA prevented detection of significant differences in the polymerase-linkage status of core protein associated versus free rcDNA. We therefore currently assume that a similar proportion of rcDNA molecules lacking intact polymerase is present in either fraction (Figure 10B). Even then, about 40% of the nuclear rcDNA were both “protein-free” and completely released from the capsid. If the individual nuclear species are true precursors of one another, a major block in HBV cccDNA formation must occur further downstream. Given the operational definition of “protein-free”, this could involve removal of amino acid remnants from the polymerase from the minus-strand 5′ end, yet also (a) subsequent step(s). Defining the chemical nature of the DNA ends in the protein-free species therefore remains a major objective.
For DHBV, the simple model outlined as pathway b (Figure 10A) may be valid; however, the stable encapsidation and low degree of deproteinization (∼20%) of the nuclear rcDNA (Figure 4B, 5, 7) versus high cccDNA content are also compatible with the existence of two kinds of capsids; one from which rcDNA is rapidly released and efficiently converted into cccDNA such that no stable intermediates accumulate (Figure 10A, pathway b), and another in which uncoating is blocked or slowed down, represented by the nuclease resistant rcDNA (Figure 10A, pathway a). Kinetic studies will be required for a distinction. Remarkably though, already 1990 Summers and coworkers [20] observed an accumulation of apparently protein-free rcDNA in primary duck hepatocytes infected with pseudotyped surface protein-deficient DHBV and suspected this might represent an immediate precursor to cccDNA.
The exact nature of the partially nuclease resistant HBV rcDNA - core protein complexes in the nucleus is not yet clear; the status of the core protein may further be probed using assembly status dependent antibodies (for review: [39]). Conversely, however, our characterization of the DNA in these complexes revealed several unexpected aspects. First, although MN has non-sequence specific endo- and exonuclease activity, MN treatment generated a reproducible pattern (Figure 4, 7) of faster migrating DNAs, with a sharp upper boundary at a position where double-stranded DNA of about 2.7 kb would migrate. These species did not represent stably encapsidated, splicing-derived shorter DNAs [32], [33], [40] but rather double-stranded linear molecules lacking ∼500 bp of viral genome sequence. Most surprisingly, digestion with four different restriction enzymes produced distinct fragments which is only compatible with MN digestion of a defined genome region relative to these restriction sites; accordingly, the MN sensitive region is roughly bordered by the start of the minus-strand DNA and the end of the core protein ORF (Figure 6). Indeed, the fifth restriction enzyme, Bsp EI, whose single recognition site locates to this region, did not detectably alter mobility of the MN treated DNA although it completely linearized rcDNA. These data strongly suggest that nuclear disassembly of the HBV nucleocapsid shell initiates at specific sites defined by their relative position to the packaged genome, and that these partially opened nucleocapsids may represent a novel, transiently stable uncoating intermediate.
Conceivable mechanisms for polymerase removal include nucleolytic cleavage of a piece of rcDNA which carries the protein, or specific cleavage of the Tyr-DNA-phosphodiester linkage; both mechanisms, possibly coupled to proteolysis, have been described for repair of cellular protein-DNA adducts [41]; our unpublished data (C. Königer, M. Nassal, J. Beck) indicate that certain Tyrosyl-DNA phosphodiesterases are indeed able to specifically cleave the polymerase from the DNA in vitro. An alternative model proposes that nucleocapsids themselves harbor an intrinsic ability for polymerase removal, perhaps via capsid-associated cellular factors, such that deproteinization as well as partial capsid opening precede nuclear import [13], [26]. Evidences were the reported presence of substantial amounts of protein-free and partially nuclease sensitive rcDNA in the cytoplasm, and the formation of small amounts of protein-free DNA in DHBV nucleocapsids derived from serum virions upon prolonged EPR conditions.
Our results do not support this view. First, in line with results from others [12] we found only a small fraction of the cytoplasmic (but most of the nuclear) HBV rcDNA in protein-free form. Second, in our hands the cytoplasmic HBV rcDNA was largely nuclease resistant, whereas most of the nuclear rcDNA was nuclease sensitive (Figure 4, 5). Third, we were unable to detect protein-free rcDNA in serum-derived HBV nucleocapsids upon prolonged EPR although the endogenous polymerase was clearly active (Figure 9). For DHBV, the same conditions led indeed to small amounts of protein-free DNA, as reported [26]. However, protein removal was extremely inefficient (<0.3%), and it concerned exclusively dlDNA, not rcDNA (Figure 9); notably, this was also observed by the authors when they analyzed cytoplasmic rather than virion-derived nucleocapsids [26]. While dlDNA can be circularized by nonhomologous end joining [42], most of the resulting molecules carry insertions or deletions, preventing generation of functional progeny virus. Thus capsid-autonomous deproteinization of dlDNA may exist but does not appear to reflect a major pathway in cccDNA synthesis.
The exclusive presence and strong enrichment, respectivley, of partially nuclease sensitive and protein-free HBV rcDNA in the nucleus in our experiments (Figure 4, 5) strongly favors that uncoating and polymerase removal are dependent on nuclear import. Furthermore, the polymerase linkage of the core protein associated MN products (Figure 7) yet overall large fraction of protein-free rcDNA in the nucleus (Figure 7, 8) suggests that uncoating precedes polymerase removal. Altogether, our data are much more compatible with a nuclear import-dependent, cell factor-mediated generation of protein-free rcDNA (Figure 10) than with a capsid-autonomous process in the cytoplasm. This would be in line with nuclear important-mediated alterations in capsid properties seen in digitonin-permeabilized cells [25]. Moreover, continued inhibition of viral polymerase activity during HBV infection of primary tupaia hepatocytes [30] and of HepaRG cells [43], the as yet only HBV-infectable human cell line [44], did not prevent cccDNA formation, further supporting a dominant role for cellular factors.
The reasons for the partially discordant results are not obvious; cell type or cell clone specific differences are not excluded. However, while we were able to prepare nuclei essentially free from cytoplasmic contaminants, the reverse was not true. A more substantial nuclear contamination of the cytoplasmic fractions employed by Guo et al. would as well account for many of the seeming discrepancies.
The efficient formation of DHBV cccDNA in HepG2 and Huh7 cells demonstrates that also in these cells the underlying pathways are principally operating; this may also hold for human cells of non-hepatic origin, as discussed in a previous publication [12]. Given the similar structures of HBV and DHBV rcDNA it is not obvious as to why conversion into cccDNA would be so much more efficient for DHBV. One option was that the plus-strand DNA in HBV capsids is less extended in comparison to DHBV capsids; however, HBV rcDNA in transfection-derived capsids from HepG2 cells was as mature as DHBV rcDNA (Figure 9). Another option is that the cellular repair activities operate in a sequence-specific manner because the two viruses share less than 40% sequence identity on the genome level [45]. Such sequence-dependence remains to be explored.
Perhaps most conceivable is that viral proteins differentially affect cccDNA formation rates, either directly or in concert with cellular factors. Absence of the surface proteins favored nuclear import for both viruses, yet with accumulation of cccDNA for DHBV, and of rcDNA for HBV. Hence more complex consequences of surface protein - core protein interactions, or an effect of the surface proteins on cellular events affecting cccDNA formation are not excluded. Nuclear import of rcDNA is mediated by the viral core proteins [24]; although the DHBV genome is even smaller than that of HBV its core protein is much larger, including a surface-exposed domain of unknown function and a differently organized C terminal nucleic acid binding domain [46]. Obviously, the rates of nuclear transport and the fates of the intranuclear capsids differed between the two viruses. Even after uncoating and polymerase removal, both proteins, or fragments thereof, in the nucleus could affect further rcDNA processing. Phosphorylation - dephosphorylation events in the core protein nucleic acid binding domains are important during replication of both viruses [33], [47], [48] but potentially differing consequences for cccDNA have not yet been addressed. Not the least, mammalian but not avian hepatitis B viruses encode an additional gene product, HBx, whose function is still elusive [49]; however, the ortholog from woodchuck hepatitis virus (WHV) appears to be crucial for in vivo infection in the natural host [50], [51]. If HBx was involved in cccDNA formation, its improper expression or functioning in the human hepatoma cells might limit HBV cccDNA formation.
In a preliminary attempt to reveal potential regulatory activities of gene products from one virus on cccDNA production by the other we co-transfected HepG2 cells with the surface-deficient variants of both viruses. HBV had little effect on nuclear DHBV rcDNA and cccDNA formation; vice versa, HBV replication appeared slightly reduced overall, with no enhancement of cccDNA accumulation. In view of the multiple options for positive and negative regulation this may not be too surprising. A more revealing approach will therefore be coexpression of one virus with defined individual gene products from the other.
Our study identified several steps downstream of initiation of nuclear uncoating that slow down (but do not completely prevent) HBV rcDNA to cccDNA conversion in human hepatoma cell lines. Because the same cells supported DHBV cccDNA formation very well, they can not completely lack the necessary molecular means. One option for the seeming discrepancy to efficient cccDNA formation during HBV infection in vivo is that cellular activities promoting the conversion are limited in these cell lines, another that inhibitory factors are overexpressed; this seems unlikely, though, because HBV cccDNA accumulation was not boosted in the avian LMH cells. Still another option is that rcDNA to cccDNA conversion is intrinsically slower for HBV than for DHBV. Determining cccDNA copy numbers in HBV infected human liver suffers from various experimental restrictions (see Introduction). Hence more telling than the final yields of cccDNA molecules per hepatocyte may be a comparison of the rates of cccDNA accumulation in HBV versus DHBV infection. In experimentally HBV infected chimpanzees, cccDNA levels during the early phase increased exponentially with a doubling time of around 4 days, expectedly paralleled by a corresponding increase in infected hepatocytes [11]; in ducks, the doubling time was only around 16 h [52]. These in vivo data would be compatible with DHBV cccDNA formation occurring at a higher rate yet at this time, a direct correlation with our findings remains speculative. Practically, however, the accumulation in the hepatoma cells of distinct nuclear HBV rcDNA forms that are likely intermediates in the rcDNA to cccDNA pathway will allow to much more thoroughly dissect the underlying molecular mechanisms. Efficient DHBV cccDNA formation in the same cells, on the other hand, provides a unique new tool to facilitate identification, e.g. by RNAi technology, of the human cellular factors involved in cccDNA biosynthesis and the screening for specific inhibitors that directly address the form of viral genome which ensures establishment and persistence of HBV infection.
Detailed descriptions of the plasmid constructs and procedures employed are reported in Text S1.
Virus vectors contained 1.5× genome length sequences of HBV (genotype D, subtype ayw; Genbank accession no.: V01460) or DHBV (DHBV16; accession no.: K01834). Numbering for HBV starts with the first nucleotide of the core open reading frame [53], for DHBV with the last nucleotide of the unique Eco RI site [54]. Plasmid pCH-9/3091 [55] contains a 1.05× genome of the same HBV isolate under control of the cytomegalovirus immediate early (CMV-IE) promoter. Surface protein deficient HBV carried mutations at positions 1399 and 1438, introducing a stop codon in the preS2 ORF and a Met>Thr exchange at the S ORF start. In surface deficient DHBV, a G>A exchange at position 1165 creates a stop codon in the S ORF [56]. In the splicing-deficient HBV construct, A1769 in the major splice acceptor consensus site CAG|G (A1769 underlined) was changed to C.
HepG2, Huh7 and LMH cells were cultured essentially as previously described [18], [33]. Transfections were performed using TransIT-LT1 reagent as recommended by the manufacturer (Mirus). Cells were harvested 3 days post transfection, unless indicated otherwise.
Briefly, cells were lysed NP40 lysis buffer, and the nuclei were separated by low speed centrifugation [33], [57]. Alternatively, nuclei were purified by sucrose gradient sedimentation (see below). Total cell extracts were obtained by lysis in 0.5% SDS. After different treatments, including or not including incubation with proteinase K (PK) and/or micrococcus nuclease (MN), viral DNAs were isolated using QIAamp silica columns (Qiagen), or by conventional phenol extraction as indicated. Dpn I (NEB) and Plasmid-Safe DNase (Epicentre Biotechnologies) were applied as suggested by the manufacturers. Viral DNAs on Southern blots were visualized using 32P labeled full-length genome DNA probes.
Cells were subjected to detergent lysis and subsequent sucrose step gradient centrifugation, essentially as described [29].
Absence of cytoplasmic contamination from the purified nuclei was assessed by comparing the presence of cytoplasmic poly-A binding protein (PABP) and nuclear histone H3 in total versus nuclear extracts. Nonspecific association of viral capsids with nuclei was addressed by mixing capsid-containing cytoplasmic extracts from transfected cells with total lysates from untransfected cells, followed by sucrose gradient purification of the nuclei. Permeability of the isolated nuclei for exogenously added MN was confirmed by fragmentation of the chromosomal DNA.
Identity of cccDNA was confirmed by linearization upon incubation with single-cutter restriction enzymes and resistance against heat denaturation. RC-DNA and nicked cccDNA were distinguished by the defined versus randomly positioned discontinuites in the DNA strands. The unique recognition site for Apa LI is located immediately upstream of the plus-strand 5′ end in RC-DNA; plus-strands not extended through this region prevent cleavage of RC-DNA whereas randomly nicked cccDNA is not affected. Cleavability at sites further upstream (Nco I) or downstream (Fsp I) served as control.
For immunoprecipitation (IP) of core protein associated DNAs monoclonal antibody 312 against HBV core protein [38] and a polyclonal antiserum raised against recombinant DHBV capsids [58] were used. For IPs from purified nuclei, the nuclei were incubated in 0.75× radioimmunoprecipitation (RIPA) buffer (1× RIPA buffer is 20 mM Tris (pH 7.2), 1% sodium deoxycholate, 1% Triton X-100, 0.1% sodium dodecyl sulfate, 150 mM NaCl), and briefly sonicated; cytoplasmic extracts were adjusted to 0.75× RIPA as well. IPs were then performed as previously described [58], with the specific antibody immobilized to protein A or protein G Sepharose (GE Healthcare). Associated DNAs were isolated as described above.
Virion-derived nucleocapsids were obtained from highly viremic sera by sedimentation in Nycodenz gradients containing 0.5% (v/v) NP-40, and subsequently incubated in EPR buffer containing 1 mM each of the four dNTPs for 16 h, as described [26].
Southern blot signal intensities were determined by phosphorimaging, using AIDA (Fuji) or ImageQuant software (GE Healthcare). Background corrections were performed by subtracting from the signal of interest the value obtained for an equally sized area in the same lane. Quantitations from each blot were performed in duplicate. In several cases, dilution series were used to more accurately determine relative amounts of a given species in different compartments or after different treatments. Standard deviations were derived from at least two and mostly more independent experiments and calculated using Microsoft Excel software. Statistical significance was evaluated using Graphpad Prism 5 for Mac software.
|
10.1371/journal.pgen.1006847 | Flipping chromosomes in deep-sea archaea | One of the major mechanisms driving the evolution of all organisms is genomic rearrangement. In hyperthermophilic Archaea of the order Thermococcales, large chromosomal inversions occur so frequently that even closely related genomes are difficult to align. Clearly not resulting from the native homologous recombination machinery, the causative agent of these inversions has remained elusive. We present a model in which genomic inversions are catalyzed by the integrase enzyme encoded by a family of mobile genetic elements. We characterized the integrase from Thermococcus nautili plasmid pTN3 and showed that besides canonical site-specific reactions, it catalyzes low sequence specificity recombination reactions with the same outcome as homologous recombination events on DNA segments as short as 104bp both in vitro and in vivo, in contrast to other known tyrosine recombinases. Through serial culturing, we showed that the integrase-mediated divergence of T. nautili strains occurs at an astonishing rate, with at least four large-scale genomic inversions appearing within 60 generations. Our results and the ubiquitous distribution of pTN3-like integrated elements suggest that a major mechanism of evolution of an entire order of Archaea results from the activity of a selfish mobile genetic element.
| Mobile elements (MEs) such as viruses, plasmids and transposons infect most living organisms and often encode recombinases promoting their insertion into cellular genomes. These insertions alter the genome of their host according to two main mechanisms. First, MEs provide new functions to the cell by integrating their own genetic information into the DNA of the host, at one or more locations. Secondly, cellular homologous recombination will act upon multiple integrated copies and produce a variety of large-scale chromosomal rearrangements. If such modifications are advantageous, they will spread into the population by natural selection. Typically, enzymes involved in cellular homologous recombination and the integration of MEs are distinct. We describe here a novel plasmid-encoded archaeal integrase which in addition to site-specific recombination can catalyze low sequence specificity recombination reactions akin to homologous recombination.
| Large-scale genomic rearrangements allow organisms to evolve much more rapidly than through random mutation alone. Rearrangements can result in the movement of genes within genomes, changes in coding strand use, loss of nonessential functions and the incorporation of foreign DNA. As a result, the organization, content and processing of genetic information can be deeply altered. In all three domains of life, chromosomal reorganization is mainly promoted by recombination between homologous sequences, for example between redundant ribosomal operons [1,2] or integrated copies of mobile elements (ME) such as prophages [3,4], transposons [5,6] and insertion sequences (IS) [7]. Such recombination can result in the DNA inversions readily observed in closely related genomes [8,9]. In addition to homologous recombination, chromosomes can undergo rearrangement through retrotransposon-associated non-homologous recombination [10]. Other elements like integrons confer rapid adaptation to bacteria in changing environments by shuffling cassette arrays encoding a variety of functions, a process involving a site-specific recombinase and two types of attachment sites [11]. Further genomic rearrangement/reorganization can occur through the acquisition of new genetic material, predominantly by lateral gene transfer. Such gene transfer occurs in all organisms through infection by mobile elements such as viruses or plasmids, or through the uptake of free or encapsulated DNA from the environment [12,13]. Genomes can acquire novel genes in a fashion ranging from transient to permanent depending on the type of element and the physiological conditions of the host. When ME succeed in stably inserting their genome, the inserted DNA is then replicated as part of the host chromosome. The transactions between ME DNA and host genome are catalyzed by recombinases typically encoded by the elements themselves. These recombinases rank in different classes based on their enzymatic activity and the specificity of their DNA targets. The smallest ME are insertion sequences (IS) composed of a short DNA segment encoding only the enzymes involved in their transposition which can occur at many different genomic locations [14]. The related transposons are larger DNA segments which can be transposed by two flanking IS and frequently carry additional genes such as antibiotic resistance determinants [15]. The most frequent IS recombinases are DDE transposases which do not form covalent transposase-DNA intermediates during transposition [16]. Other and typically larger ME such as plasmids and viruses encode recombinases promoting DNA transactions with a stronger DNA sequence specificity. Such site-specific recombination is not only used for mobile element integration and excision in bacteria but also in the spread of antibiotic resistance by transposable elements, the control of plasmid copy number, regulation of gene expression and the resolution of concatenated chromosomes [17]. Site-specific recombinases can be categorized into the serine recombinases and tyrosine recombinases (Y-recombinases); which, in contrast to DDE transposases, form covalent enzyme-DNA intermediates during recombination, albeit with markedly different mechanisms of action. Before religation of the two recombining DNA strands, serine recombinases generate breaks in all strands while Y-recombinases produce two sequential single-strand breaks [17]. As a rule, site-specific integration/excision reactions promoted by Y-recombinases occur via a synaptic complex composed of two DNA duplexes carrying the specific sites bound by four recombinase protomers [17]. The two-recombinase pairs are activated sequentially, allowing one strand from each duplex to be exchanged at a time via two consecutive and symmetrical Holliday junctions. A notable exception is Vibrio cholerae phage CTX. Not only does this phage integrate into its host genome in single stranded form where two sites fold into a hairpin structure, mimicking a recombination target for the cellular XerCD chromosome resolvase; but also only requires XerC for integration [18].
One of the best-studied Y-recombinases is the integrase of phage λ. The primary function of this enzyme is the integration of phage DNA into the chromosome of its bacterial host (and its excision). This function is achieved by promoting site-specific recombination between the phage attachment site attP and its chromosomal counterpart attB [19]. Under particular circumstances, the integrase of the lambdoid phage HK022 is capable of generating inversions between attP and a secondary attachment site in the HK022 left operon [3]. Similarly, the primary function of the yeast FLP protein is the control of the 2μ plasmid copy number [20] by DNA inversion between two divergent 34bp FRT sites located on the plasmid [21]. FLP recombinase activity has also been successfully used for integration and excision of synthetic DNA in mammalian genomes [22]. The recombination activities of both λ integrase and FLP recombinase are summarized as shown in S1 Fig. Historically, this reciprocal and conservative recombination between two stringently defined double-stranded DNA sequences in each chromosome was denominated the Campbell model [23].
The sequences of a considerable number of Y-recombinases have been compared to reveal the position of conserved residues and infer the location of the catalytic active site [24]. They share in their C-terminal moiety a rather well conserved region of ~120 amino acids containing up to six nearly invariant amino acids R..K..HxxR..[W/H]..Y forming the active site [25,26]. A small number of Y-recombinases have been characterized biochemically in Archaea, for example the XerA recombinase of the hyperthermophilic euryarchaeon Pyrococcus abyssi which exhibits a perfect active site consensus [27]. Sequence alignments have revealed that other archaeal active sites diverge slightly from the bacterial consensus R..HxxR..Y [28]. The integrases of viruses SSV1 isolated from the hyperthermophilic crenarchaeon Sulfolobus shibatae [29] and SSV2 from Sulfolobus islandicus [30] share the consensus R..KxxR..Y while the plasmidic integrase of Sulfolobus sp. NOB8H2 displays R..YxxR..Y [28].
Mobile elements therefore contribute to genome evolution through both site-specific and homologous recombination, which usually operate by distinct mechanisms and enzymatic activities. Homologous recombination is also known to occur frequently between multiple IS copies resulting in large scale archaeal genomic rearrangements, as observed in both Crenarchaeota e.g. Sulfolobus islandicus [31] and Euryarchaota e.g. Pyrococcus abyssi [32]. The distribution of archaeal ISs is patchy not only at the phylum level but also at genus level [9]. Interestingly, genome shuffling occurs in Thermococcus [33] even if ISs are seldom found in this genus suggesting that alternative recombination mechanisms are capable of producing large-scale genomic rearrangements.
If site-specific recombination only requires specific nucleotide sequences targeted by a dedicated recombinase, homologous recombination on the other hand is a much more complex process. In all organisms, homologous recombination constitutes one of several pathways to repair double-strand breaks. In addition to DNA synthesis, it requires dedicated recombinases and their accessory factors which act on stretches of near-sequence-identical DNA. In eukaryotic and bacterial cells, the enzymes and pathways involved in homologous recombination have been extensively studied (see [34,35] for reviews), whereas archaeal homologous recombination is still an active field of investigation. It is known that the initial resectioning step after double-strand break involves the Rad50–Mre11–HerA–NurA complex to generate 3’ single-strand substrates [36,37]. The RecA paralog RadA and its accessory functions associate with this ssDNA to constitute the presynaptic filament, which will scan and pair with homologous sequences [38]. In the archaeon Thermococcus kodakarensis, homologous recombination has been detected experimentally between stretches of identical DNA sequences equal to or greater than 500bp [39].
To our knowledge, a direct overlap between site-specific and homologous recombination processes has not been described so far. In the present work, we report the discovery and characterization of a new integrase from the hyperthermophilic archaeon Thermococcus nautili [40,41] capable of catalyzing both site-specific recombination and low sequence specificity recombination reactions mimicking homologous recombination. The wide distribution of this particular Y-recombinase among the Thermococcus genus provides a valid rationale for the observed genomic rearrangements in these Archaea.
We compared the chromosomes of the 13 completely sequenced Thermococcus species available to date by dotplot analysis and observed high levels of genome scrambling as shown in Fig 1A. Strikingly, comparison of T. onnurineus and T. sp. 4557 chromosomes by this approach revealed only two large inversions of 139/143Kb and 102/74Kb respectively (Fig 1B & 1C). This relatively small number of inversions facilitated the investigation of the synteny breakpoints bordering both inversions. Using the SyntTax web tool [42], a composite representation was obtained as shown in Fig 1C. Gene order is conserved immediately upstream and downstream of each inversion border and was used to identify the synteny breakpoints. For each inversion, the breakpoints are located within tRNA gene pairs, transcribed in opposite orientations. Interestingly, T. nautili plasmid pTN3 integrates in the tRNALeu gene BD01_0018 [41,43] (S2 Fig) and this gene displays over 97% sequence identity with tRNALeu (GQS_t10759), which borders a large chromosomal inversion between T. onnurineus and T. sp. 4557 (Fig 1B). The concordance between the chromosomal attachment site of the pTN3 integrase (IntpTN3) and the recombination targets bordering each inversion (in opposite orientations) led us to define a working model to explain the formation of genomic inversions observed in the Thermococcus genus. We hypothesize that the frequent genomic inversions observed in the evolution of the Thermococcales order are a result of enzymatic activity of the integrase encoded by horizontally mobile elements, such as pTN3.
The integrase of pTN3 shares significant sequence similarity with canonical Y-recombinases and its predicted active site can be defined as R..K..AxxR..Y which only slightly diverges from the consensus (S3A Fig). In addition, IntpTN3 displays a high degree of conservation with two biochemically characterized hyperthermophilic Y-recombinases, the archaeal IntSSV1 [44] and IntSSV2 [30] (S3B Fig). Thus, it seemed worthwhile to compare the enzymatic activities of IntpTN3 to those of other enzymes of the same family such as phage λ integrase and Saccharomyces cerevisiae 2μ plasmid FLP protein and to validate them against the canonical Y-recombinase model.
In order to characterize the activities of IntpTN3, it was necessary to over-produce and purify the enzyme (S4 Fig) and to construct DNA substrates carrying appropriate attachment sites (as determined by sequential deletions (S5 Fig). An integrase variant (IntpTN3Y428A) in which the catalytic tyrosine is substituted with an alanine was constructed, purified and tested (S6 Fig). We used these proteins and DNA components in a series of in vitro and in vivo experiments, detailed below, to ascertain the properties of IntpTN3.
The large-scale genomic inversions observed between T. sp. 4557 and T. onnurineus display minor gene order rearrangements near the recombination endpoints indicating that these events are not recent and might have undergone remodeling (Fig 1C). In order to identify more recent rearrangements, we investigated whether large-scale genomic inversions could occur spontaneously under laboratory conditions. T. nautili carrying its natural plasmids was sub-cultured in two independent experiments for 60 and 66 generations (therefore termed T. nautili 60G and 66G) in rich liquid medium with intermittent storage at 4°C and the metagenomes of the resulting populations were completely re-sequenced. We observed in both T. nautili 60G and 66G sub-cultures a high proportion of a novel rearranged genome exhibiting four new large-scale chromosomal inversions when compared to the original published T. nautili genome (GenBank accession NZ_CP007264) [41] (Fig 6A). By mapping the frequency of the Illumina reads around the four inversion sites, we measured the incidence of the rearranged genome in the T. nautili 66G population, which was found in most cases to exceed that of the original genome (S3 Table). Both T. nautili 60G and 66G rearranged chromosomes were remarkably similar when compared by dotplot analysis (S7 Fig). Additionally, plasmid pTN3 was largely underrepresented in the T. nautili 66G sub-culture (S3 Table), whereas the smaller pTN1 and pTN2 were conserved. The chromosomally-integrated pTN3 copy carrying the disrupted integrase gene was also retained. The chain of nested inversion events leading to these new recombined genomes could be reconstructed (Fig 6C) and allowed us to analyze and precisely map the recombination endpoints. Each of the four genomic inversions occurred between paralogous gene pairs: between tRNAGly genes BD01_1557 and BD01_1976, between methyl accepting chemotaxis genes BD01_1166 and BD01_1584, between transposase genes BD01_1317 and BD01_1763 and finally between UDP-glucose-6 dehydrogenase genes BD01_1333 and BD01_1481. For each pair of paralogous genes, the inversion events always occurred between two inverted segments of DNA sharing extensive sequence identity (S8 Fig). However, we could not detect significant similarity between inverted DNA segments corresponding to different pairs of paralogous genes using BLAST (e-value ≥ 0.075). Furthermore, none of these sequences could be aligned with the original pTN3 attachment site, tRNALeu (e-value ≥ 10). In a control experiment, in contrast to T. nautili, the genome of a closely related organism, the plasmid-less Thermococcus sp. 5–4 (GenBank accession CP021848) remained stable when sub-cultured for 36 or 66 generations in two separate experiments (Fig 6B and S7 Fig).
The remarkable differences in the outcome of T. nautili and T. sp 5–4 sub-culturing experiments and the observation that tRNAGly genes could recombine in these conditions suggested a causal link between IntpTN3 and genome shuffling. To ascertain if the new recombinations in T. nautili 60G and 66G could have been indeed generated by IntpTN3, we decided to test whether this integrase was able to catalyze in vitro inversions using the sequences detected at the borders of these recombination events. New inversion templates pCB548 and pCB552 were thus constructed respectively carrying sequences encompassing tRNAGly genes BD01_1557 and BD01_1976 or sequence fragments from chemotaxis genes BD01_1166 and BD01_1584 (S8 Fig). To limit the number and size of generated fragments, an in vitro inversion assay was conducted on linear fragments originating from these plasmids and compared to a linear fragment carrying inverted attP sites derived from pCB524. Inversions could be detected with all three templates albeit with significantly longer incubation times or higher IntpTN3 concentrations for pCB548 and pCB552-derived templates as compared to pCB524 (Fig 7). To confirm this recombination event, one of the products of the pCB548 template inversion reaction was further characterized by DNA sequencing and corresponded to a bona fide cross-over between BD01_1557 and BD01_1976 (S9 Fig). We conclude that IntpTN3 is able to catalyze low sequence specificity recombination reactions between sites that differ in sequence from its cognate att site, with the same outcome as homologous recombination events. It is to be noted that IntpTN3 catalyzes these two types of reactions with a different efficiency. Site-specific recombination reactions reach the equilibrium within 30 minutes whereas several hours and higher enzyme concentrations are required to detect all low sequence specificity recombinations.
The absence of inter-pair DNA similarity observed in T. nautili 60G and 66G chromosomal inversions prompted us to test whether IntpTN3 could catalyze recombination between homologous non-archaeal sequences. The simplest experiment consisted of the incubation of cloning vector pBR322 DNA with the integrase in the same conditions as described above. This recombination reaction promoted by IntpTN3 yielded a ladder of plasmid multimers produced by sequential integration, which could be readily observed by eletrophoretic migration whereas no homologous integration reaction was detected with the mutated IntpTN3Y428A (Fig 8A). Surprisingly, IntpTN3 generated also a double-strand cut at the pBR322 ColE1 origin of replication for which we have no explanation at this stage (S10 Fig). This cleavage does not constitute an intermediate step in the recombination reaction since none of IntpTN3 linear substrates shown in Fig 7 carries the ColE1 origin. In addition to the homologous integration reaction, we investigated the capacity of IntpTN3 to promote inversions between homologous sequences of bacterial origin. Short DNA segments of decreasing length (250, 175 and 100bp, see S11 Fig) originating from the E. coli lacZ gene were cloned in opposite orientations respective to the lacZα gene of pUC18 to generate plasmids pCB574, pCB571 and pCB558, respectively. These templates were linearized, incubated with IntpTN3 and tested by subsequent restriction analysis. In each case, IntpTN3 generated additional bands consistent with homologous inversion reactions displaying efficiencies proportional to the extent of DNA identity (Fig 8B).
The major mechanism producing chromosomal rearrangements is recombinational exchange between homologous sequences [46]. These rearrangements often consist of DNA inversions between IS elements [9,46,47]. The observation that, in the Thermococcus genus, large chromosomal inversions occur even in the absence of IS elements prompted us to investigate the molecular mechanism behind these rearrangements. The presence of tRNA genes at recombination endpoints in genomes as diverse as plant chloroplasts [48,49] and Thermococcales [9], combined with the fact that integrases often target tRNA genes [50], lead us to propose a precise molecular model involving IntpTN3 to explain large-scale genomic rearrangements. Using a combination of comparative genomics, in vitro analyses, and serial culturing experiments, we uncovered a mechanism and enzymatic activity responsible for the shuffling-driven chromosomal evolution in Thermococcales. By means of deep comparative genomic analyses, we were able to correlate genome scrambling with the presence of a mobile element. This mobile element has been identified as plasmid pTN3, naturally present in T. nautili both as an episome and integrated in the genome [41,43]. Plasmid pTN3 encodes the IntpTN3 integrase of the Y-recombinase superfamily capable of promoting its site-specific plasmid integration at a tRNALeu gene of its host. Due to perfect DNA conservation between attB and attP attachment sites (S2B Fig), an intact and presumably expressed tRNALeu is reconstituted upon pTN3 chromosomal integration. We successfully reproduced, with high efficiency in a purified in vitro system, the canonical DNA reactions of integration and excision expected from a bona fide integrase. Site-specific mutation of the active site tyrosine to alanine abolished these activities. A positive excision reaction was also obtained in vivo by expressing wild-type IntpTN3 and the catalytic tyrosine mutant IntpTN3Y428A in T. kodakarensis KOD1 cells. The genome of this strain carries the integrated episome TKV4 [45] which is remarkably similar to pTN3 (Fig 9). Surprisingly, both wild-type and mutant forms of the integrase excised TKV4 in circular form. This suggests that a truncated C-terminal IntTKV4, presumably impaired in DNA-binding but carrying the catalytic tyrosine, can complement IntpTN3Y428A. A plausible explanation invokes the participation of integrase dimers in the recombination reaction. In this case, only the heterodimeric form would possess an active catalytic site where Tyr428 is provided by the first monomer while the second monomer contributes the remaining conserved residues. This cleavage in trans was initially reported for the FLP recombinase [51,52]. Similarly, the complementation of activity between a DNA-binding impaired mutant and a catalytic tyrosine residue mutant has been described for another archaeal integrase, IntSSV1 [44].
The peculiar location of tRNALeu GQS_t10759 at the exact border of a large DNA inversion observed between the genomes of T. onnurineus and T. sp. 4557 suggested that this inversion could have occurred by the recombinase activity of IntpTN3. In our purified system, we could obtain highly efficient DNA inversions between two inverted copies of GQS_t10759. Paradoxically, we were unable to promote inversion between tRNALeu GQS_t10759 and tRNAThr GQS_t10745 contrary to what the genomic comparisons between T. onnurineus and T. sp. 4557 suggested. An experiment of prolonged T. nautili cultivation was instrumental in elucidating the large-scale inversion mechanism in Thermococcus. The strain carrying its natural plasmids was cultivated during 60 or 66 generations; total DNA was extracted from this population and sequenced in a manner similar to a metagenome. We observed the high incidence in the resulting populations of a particular recombined genome with four large chromosomal inversions and a very low copy number of plasmid pTN3 encoding active IntpTN3 (< 2/chromosome) (S3 Table). This plasmid loss could have contributed to the higher fitness and spread of a particular clone in the population. The four large-scale inversions occurred between four pairs of naturally occurring paralogous genes sharing at least 104bp of sequence identity in inverted orientation (S8 Fig). No significant sequence conservation could be detected between the four pairs. We did not observe chromosomal rearrangements after prolonged incubation of Thermococcus sp. 5–4, which does not carry plasmids. The potential causal link between pTN3 and a number of unrelated sequence pairs involved in large scale genomic shuffling in T. nautili was difficult to conciliate with the classical site-specific recombination properties we described for IntpTN3. Remarkably, by in vitro assays with this integrase, we succeeded in producing inversions between several pairs of inverted paralogous genes detected in our T. nautili sub-culturing experiments. These results suggested that the recombination properties of IntpTN3 could be extended to virtually any homologous pair of DNA sequences. Using exogenous pBR322 plasmid DNA or genes segments from bacterial origin, we demonstrated in vitro that IntpTN3 actively promotes low sequence specificity reactions mimicking homologous integration and inversion of any sequence pair as short as 100bp. The catalytic site mutation in variant IntpTN3Y428A abolishes this particular recombination reaction as well. Interestingly, cellular homologous recombination in Archaea operates according to a different pathway with dedicated enzymes [36,37] and in Thermococcus kodakarensis has only been reported between DNA segments of 500bp or more [39].
These reactions unveiled a specific IntpTN3-generated double-strand cut at the ColE1 origin of replication carried by pBR322 and its derivatives (S10 Fig). At this moment, we do not have a precise rationale to explain this observation other than a potential distant secondary structure similarity between the small RNAI and RNAII encoded by the ColE1 origin and the tRNALeu encoded by IntpTN3 attB substrate. Biological interactions between tRNAs and ColE1 RNAs have been reported [53]. Clearly, this double-strand cleavage does not participate in any recombination reaction since we demonstrated all in vitro IntpTN3 inversions on linear DNA segments devoid of ColE1 origin.
The positive in vitro IntpTN3-promoted low sequence specificity recombination results explain the failure of this enzyme to promote inversion between tRNALeu GQS_t10759 and tRNAThr GQS_t10745. These sites were initially thought to constitute inversion endpoints between the genomes of T. onnurineus and T. sp. 4557 but do not share sufficient sequence similarity to be efficiently recombined in vitro. The particular positioning of these sequences in opposite orientations could have occurred through previous overlapping inversions between a different set of paralogs or by less frequent native homologous recombination. We observed a similar situation in the sequence of the T. nautili 60G and 66G populations. In several cases, homologous segments were in direct orientation in the original genome but became opposed due to a previous overlapping inversion therefore indicating that T. nautili 60G and 66G inversions occurred sequentially.
In order to investigate whether pTN3 could account for large-scale rearrangements in the Thermococcus genus, we examined by synteny analysis the distribution of pTN3-like integrated element among completely sequenced Thermococcales. Out of 17 sequenced Thermococcus, and in addition to the previously reported T. kodakarensis TKV4 element [45], five isolates were found to harbor a pTN3-related element (Fig 9). The natural competence for DNA uptake of some Thermococcales such as T. kodakarensis [39] and the capacity of pTN3 to be transferred between cells using membrane vesicles [43] could explain the ubiquitous presence of this mobile element.
Protein sequence and structural comparisons between IntpTN3 and other hyperthermophilic archaeal integrases such as that of crenarchaeal virus SSV1 indicate that these proteins are clearly related. However, IntpTN3 possesses several additional interspersed domains relative to SSV1 (S2 and S12 Figs). We surmise that these additional domains contribute to the low sequence specificity recombination reactions akin to homologous recombination events that we have observed.
By summing up all direct and indirect evidence reported here, it is very likely that the integrase encoded by pTN3-like plasmids can account for the genomic shuffling observed in the Thermococcus genus. Plasmids of the pTN3 class are genetically closely related to viruses as they encode a capsid protein and a DNA packaging ATPase [43] but pTN3 virions have not be observed to date. It is not clear at this stage whether plasmids or viruses equipped with an IntpTN3-like integrase have a better fitness either due to provirus maintenance or by virion spreading. An integrase mimicking homologous recombination could promote viral integration into the host genome only if both viral and cellular chromosomes share significant DNA similarity. This enzyme however, could facilitate integration of a virus into the genome of a closely related provirus.
The question arises whether an enzyme promoting genome shuffling using very short repeated segments as substrates, would be beneficial for a cellular organism. On one hand, ‘wrongly’ recombined genomes would result in suboptimal gene expression programs and cells carrying scrambled genomes would display a reduced fitness and clearly be counter-selected in the population. Interestingly, the presence of a pTN3-specific spacer in a T. nautili CRISPR locus strongly suggests that the presence of this plasmid is deleterious [41]. On the other hand, it is also possible to envision situations where high-level genome shuffling by inversion could be advantageous. Alternate gene expression patterns could increase, for instance, adaptation to rapid environmental changes. In addition, for organisms such as Thermococcales where highly-expressed essential housekeeping genes maintain invariable positions [33], genome scrambling could be beneficial by relocating “less desirable” integrated elements to chromosomal areas of reduced gene expression, therefore minimizing their impact on cellular physiology.
Escherichia coli strain XL1-Blue was used for cloning, plasmid amplification and site-directed mutagenesis. Overexpression of recombinant wild-type or mutant IntpTN3 was carried out in strain BL21 (DE3) (Novagen). All E. coli strains were grown in Luria-Bertani medium supplemented with 100μg/mL ampicillin or/and 50μg/mL kanamycin when necessary. T. kodakarensis KUW1 (ΔpyrF ΔtrpE) was grown anaerobically in ASW-YT medium [54] at 85°C. Long term Thermococcus sub-culturing experiments were carried out in the same conditions by sequential 50x dilutions of stationary phase cultures into fresh media. The number of generations was assessed statistically at each dilution step using a Thoma cell counting chamber under 400x magnification. The plasmids used or constructed in this work are listed in S1 Table. Transformation with pRC524 and pRC526 plasmids (see below) was performed following standard protocols [55]. Plasmid-containing KUW1 strains were grown in ASW-CH medium [54] supplemented with uracil (10 μg/mL). T. nautili sp. 30–1 (CP007264) was grown anaerobically at 85°C in Zillig’s broth [56].
Genomic sequences were compared and aligned by dotplot analysis using Gepard [57]. Conservation of gene order was assessed by synteny analysis using Absynte [58] and SyntTax [42]. The original genome of Thermococcus 5–4 JCM31817 (GenBank accession CP021848) and the genomes of sub-cultured T. nautili 60G and 66G and T. sp. 5–4 36G and 66G were sequenced by Genoscope (Centre National de Séquençage, France), using Illumina MiSeq. Reads were assembled with Newbler (release 2.9) and gap closure was performed by PCR, Sanger sequencing and Oxford Nanopore MinION. The primary genomic sequences of rearranged T. nautili 60G, 66G and T. 5–4 36G, 66G are available in S1, S2, S3 and S4 Datasets, respectively. These genomic sequences are compared by dotplot analysis in S7 Fig.
Genomic regions corresponding to ~2000bp upstream and downstream of inversion break-points were extracted from both the ancestral T. nautili sequence, and the sub-cultured T. nautili 66G sequence. Illumina sequencing reads were mapped to the ancestral sequence, and the pool of unmapped reads were mapped to the 66G sequence (Geneious 6.1.8). Two positions close to the break-point which differ in base composition between ancestral and 66G sequences were chosen to classify reads as resulting from original or inverted genome sequences. Bases were enumerated at these positions, and the percentage of reads corresponding to original sequences or inversions were calculated. The prevalence of pTN3 in the population was determined by comparing read depth across the entire T. nautili 66G genome (excluding the integrated pTN3 region) to that of pTN3 (S3 Table).
The gene encoding the integrase of the plasmid pTN3 of T. nautili 30–1, (gene ID: 17125032) was codon-optimized for expression in E. coli and synthesized by GenScript. The synthetic gene contained a Strep-Tag at the 5’ end and was cloned into pET26b+ expression vector (Novagen) to yield pJO344. Plasmid pJO496 carrying the mutated IntpTN3Y428A was obtained by site directed mutagenesis of pJO344 with primers Int_A and Int_B (S2 Table) using the Agilent QuikChange Lightning Site-Directed Mutagenesis Kit. Wild-type IntpTN3 and mutated IntpTN3Y428A were purified from E. coli BL21 (DE3) strain (Novagen) harboring respectively pJO344 or pJO496 by affinity chromatography and gel filtration (S4 Fig). All integrase enzymatic assays were conducted with strep-tagged protein derivatives.
Plasmids used for the integrase dimerization assays were constructed as follows. EcoRI and BamHI restriction sites were added respectively at the 5’ and 3’ end of the various oligonucleotides shown in S5 Fig. Each oligonucleotide (Sigma-Aldrich) was annealed to its complementary sequence and the resulting double-stranded segments were cloned between the corresponding restrictions sites of pUC18. To generate plasmid pMC451, the Leu2-88 fragment was cloned in pBR322 instead of pUC18. Plasmids pMC477 and pMC479 used respectively for att integration/excision and inversion assays were constructed using pMC451 as backbone. The insertion fragment was amplified with primers Leu43scaI_fw and Leu43scaI_rev using pMC449 plasmid DNA as template. It contains tRNALeu gene (2-44bp) and lacZa gene for blue-white screening. This amplified region was cloned in pMC451 in both possible orientation using ScaI and NruI blunt sites. Plasmid pCB538 was obtained by amplifying with primers LacZ100-Sac1-For and KanR-Xba1-Rev (S2 Table) a 1364bp fragment from pUC4K and subsequent cloning between the XbaI-SacI sites of pUC18. The other plasmids: pCB548, pCB552, pCB572 and pCB574 used for non-att inversion assays were generated by Gibson Assembly [59]. Briefly, for pCB548, the genomic region corresponding to -80 to +245 of BD01_1557 (T. nautili) was amplified by PCR (Phusion Polymerase, ThermoScientific) using primers 1557_fwd and 1557_rev (S2 Table); the region from –80 to +245 of BD01_1976 was amplified using primers 1976_fwd and 1976_rev. The KmR gene was amplified from plasmid pUC4K using primers KanR_fwd and KanR_rev. Fragments were assembled into EcoRI + SalI digested pUC18 using the NEBuilder HiFi DNA Assembly Master Mix (New England Biolabs) following the manufacturer’s protocols. Similarly, for pCB552, part of the genes BD01_1166 and BD01_1584 (S8 Fig) were amplified by PCR and assembled into EcoRI + SalI digested pUC18 with the KmR gene sequence. To construct pCB538, a fragment containing KmR and the beginning of the lacZ gene (lac100) was PCR-amplified from pUC4K with the primers LacZ100-Sac1-For and KanR-Xba1-Rev containing the restriction sites for SacI and XbaI, respectively, at the 5’ end. The adequately digested fragment was then ligated into a SacI-XbaI digested pUC18. For plasmids pCB572 and pCB574, part of the lacZ gene was amplified from pUC18 and the KmR gene sequence was amplified from plasmid pUC4K. The two fragments were then assembled into the EcoRI digested pUC18. Purified plasmids pCB548, pCB552, were digested using ScaI and EcoRI and plasmids pCB572 and pCB574 were digested using ScaI. The fragments containing the non att-sites were then gel purified using the kit NucleoSpin Gel and PCR Clean-up (Macherey Nagel). All plasmid constructs were confirmed by DNA sequencing (Beckman Coulter Genomics).
Standard in vitro integrase assays were performed as follows: 165ng (8.25ng/μL, 3.1pmol) purified IntpTN3 and 0.5μg (25ng/μL, 10pmol) supercoiled plasmid substrates were incubated 30 min at 65°C in a reaction buffer containing 300mM KCl, 27 mM Tris HCl pH8, 0.17mM DTT and 1mM MgSO4. Depending on the size of the plasmid substrate, the DNA/integrase molar ratio varied from 30 to 60. For substrates with non-att sites, the integrase concentration was increased up to 50pmol. To assay dimer formation, the reaction products were separated by gel electrophoresis and visualized with ethidium bromide. For the excision and inversion assays, reaction products were purified with the NucleoSpin Gel and PCR Clean-up kit (Macherey-Nagel) and digested with appropriate restriction enzymes (Thermo Scientific) prior to eletrophoretic separation. In vitro circularization of TKV4 was performed in a standard integrase assay with genomic DNA of T. kodakarensis isolated as described previously [60]. The reaction products were purified using NucleoSpin Gel and PCR Clean-up kit (Macherey-Nagel). Recircularized products were scored by amplifying a reconstituted full-length TKV4 integrase gene. PCR was performed using Phusion Polymerase (ThermoScientific) and primers TKV4_FW and TKV4_REV (S2 Table) in conditions recommended by the supplier. In vivo circularization of TKV4 was obtained using total DNA from T. kodakarensis KUW1 transformed with plasmid pRC524 or pRC526. These plasmids express constitutively wild type integrase and mutated IntpTN3Y428A from the PhmtB promoter present in parental pLC70. DNA extraction and PCR reactions was performed as per the in vitro assay described above. To generate plasmids pRC524 and pRC526, the IntpTN3 integrase gene was amplified by PCR with primers int_fwd and int_rev (S2 Table), using total T. nautili genomic DNA as a template. The amplification product was cloned into pJET1.2 using the CloneJET PCR Cloning Kit (Thermo Fischer Scientific). The Y428A mutation was introduced into the integrase gene using the QuickChange II Site Directed Mutagenesis Kit (Agilent Technologies) with primer intY428A_fwd and its reverse complement. Both the wild-type and Y428A alleles were digested from pJET1.2 using SalI and NotI and cloned into the corresponding sites of pLC70. All in vitro and in vivo recombination junctions and plasmid constructs were confirmed by DNA sequencing (Beckman Coulter Genomics).
|
10.1371/journal.pntd.0006639 | Development of NanoLuc-PEST expressing Leishmania mexicana as a new drug discovery tool for axenic- and intramacrophage-based assays | The protozoan parasite Leishmania causes leishmaniasis; a spectrum of diseases of which there are an estimated 1 million new cases each year. Current treatments are toxic, expensive, difficult to administer, and resistance to them is emerging. New therapeutics are urgently needed, however, screening the infective amastigote form of the parasite is challenging. Only certain species can be differentiated into axenic amastigotes, and compound activity against these does not always correlate with efficacy against the parasite in its intracellular niche. Methods used to assess compound efficacy on intracellular amastigotes often rely on microscopy-based assays. These are laborious, require specialist equipment and can only determine parasite burden, not parasite viability. We have addressed this clear need in the anti-leishmanial drug discovery process by producing a transgenic L. mexicana cell line that expresses the luciferase NanoLuc-PEST. We tested the sensitivity and versatility of this transgenic strain, in comparison with strains expressing NanoLuc and the red-shifted firefly luciferase. We then compared the NanoLuc-PEST luciferase to the current methods in both axenic and intramacrophage amastigotes following treatment with a supralethal dose of Amphotericin B. NanoLuc-PEST was a more dynamic indicator of cell viability due to its high turnover rate and high signal:background ratio. This, coupled with its sensitivity in the intramacrophage assay, led us to validate the NanoLuc-PEST expressing cell line using the MMV Pathogen Box in a two-step process: i) identify hits against axenic amastigotes, ii) screen these hits using our bioluminescence-based intramacrophage assay. The data obtained from this highlights the potential of compounds active against M. tuberculosis to be re-purposed for use against Leishmania. Our transgenic L. mexicana cell line is therefore a highly sensitive and dynamic system suitable for Leishmania drug discovery in axenic and intramacrophage amastigote models.
| The protozoan parasite Leishmania causes a spectrum of diseases collectively known as leishmaniasis. The parasite is transmitted to humans by the bite of its vector, the sand fly, following which the parasite invades host white blood cells, particularly macrophages. Leishmaniasis is classified as a neglected tropical disease, and is endemic in 97 countries. Symptoms of the disease depend on the species of Leishmania. These include skin lesions, destruction of the mucosal membranes, and the visceral form which is usually fatal if untreated. Current therapeutic options for leishmaniasis have a number of associated problems that include toxicity, the development of drug resistance and poor patient compliance due to lengthy and painful treatment regimens. New therapeutics are therefore urgently needed. The ability to screen potential drug candidates requires robust screening assays. Currently, screening the intracellular parasite relies on microscopy-based techniques that require expensive equipment, are time consuming and only detect parasite burden, not viability. By using a transgenic cell line that expresses the NanoLuc-PEST luciferase, we show that we have a parasite-specific viability marker that can be used to measure the efficacy of compounds against the intracellular parasite. We validate the potential of this cell line by screening the MMV Pathogen Box.
| The leishmaniases are a spectrum of diseases caused by infection with protozoan pathogens of the Leishmania genus, with an estimated 1 million new cases per annum. [1] Leishmania parasites are transmitted to a mammalian host via the bite of an infected sand fly. Highly motile metacyclic promastigotes invade host macrophages and differentiate into the amastigote form, which is highly adapted for intracellular survival. [2] Current treatments for leishmaniasis (reviewed in [3, 4]) are unsatisfactory due to high associated toxicity, cost, complex administration and the emergence of resistant strains. Efforts have greatly increased over the last decade to identify novel compounds with anti-leishmanial properties, or to repurpose existing drugs to widen the therapeutic options for this disease. The Drugs for Neglected Diseases initiative (DNDi) was set up to identify potential lead compounds. This has yielded success with the identification of three new chemical series that display considerable anti-leishmanial potential. [5]
Efficient compound screening requires robust, sensitive and reproducible assays that are suitable for high throughput application. These assays primarily fall into two categories: target directed screening, and phenotypic screening. Recent advances have been made to bridge this gap using yeast-based systems, which express leishmanial target proteins for screening. [6–8] The main argument for phenotypic assays is that they directly measure compound activity against the target cell. This tends to be determined using either a colourmetric or fluorescent regent that measures metabolism (e.g. MTT [9] or AlamarBlue [10]), or using a fluorescent reporter molecule such as GFP [11–13] or mCherry. [14] The metabolism-based reagents are useful for studying the parasite alone in either its promastigote or axenic amastigote form, but they cannot distinguish between the parasite and host cells in an in vitro cell infection model. Parasite-specific fluorescent reporter molecules show the presence and localisation of parasites; however the assays used tend to look for the presence of the fluorescent parasite (either by microscopy or flow cytometry), not parasite viability.
Screening for novel anti-leishmanial compounds using an intramacrophage model is more relevant, and is likely to provide a better translation of hits. This is because the model incorporates both the multiple membranes that the compound must traverse to reach the amastigote, and the environment within the parasitophorous vacuole. The current methods for assessing efficacy against the amastigote in its intramacrophage niche involve either flow cytometric or microscopy-based techniques. These can be scaled up to a high content screening system, [15] but this requires the use of specialist and expensive equipment. In addition, these methods can only detect parasite burden, not viability. One technology that can overcome this hurdle is the use of bioluminescence. Bioluminescence offers a dynamic method for determining both the viability and location of the tagged cell, and has been applied to a number of infectious disease models. [16–18] This technology utilises transgenic pathogens that express one or more luciferases; a diverse group of enzymes that have the ability to generate light in the presence of a specific substrate. In Plasmodium falciparum, luciferases have proven to be robust and sensitive reporters in drug screens. [16, 19–22] For kinetoplastid research, luciferases from the North American firefly (Photinus pyralis) and the sea pansy (Renilla reniformis) have been used for drug screening [23–25] and in vivo studies. [26–29] A red-shifted variant of the firefly luciferase (PRE9) shows significantly improved sensitivity in Trypanosoma brucei infection in animal models, and has proven to be a powerful technique for detecting low numbers of parasites in the brain. [30]
Whilst luciferases derived from P. pyralis and R. reniformis are the most commonly used, other luciferases with differing properties are now being explored. A luciferase isolated from the deep sea shrimp (Oplophorus gracilirostris), known as NanoLuc, is a relatively small (19 kDa, compared to 61 kDa and 36 kDa for P. pyralis and R. reniformis respectively) and very stable enzyme that produces a high intensity, glow-type bioluminescence. [31] A modified form of the enzyme, NanoLuc-PEST (23 kDa), retains high enzymatic activity but has a reduced intracellular half-life due to fusion of a PEST sequence, which marks the molecule for rapid degradation. [31] NanoLuc has been successfully expressed in Plasmodium falciparum [32] but there are no reports to date describing expression of this reporter (or the PEST-fusion derivative) in kinetoplastids.
New molecular tools, such as the NanoLuc-PEST enzyme, are being used to improve compound screening to aid drug development. [33] This study focuses on the use of transgenic parasites expressing this enzyme, in both axenic- and intramacrophage-based assays. This new method, which specifically detects amastigote viability within the macrophage, will help bridge the gap between axenic and in vivo testing, in a manner that is time efficient and scalable to high-throughput systems.
L. mexicana strain MNYC/BZ/62/M379 was maintained in vitro in the procyclic promastigote stage by culture at 26°C in Schneider’s medium (Gibco) pH 7.0 containing 10% FBS (Gibco), 100 U/mL penicillin (Lonza) and 100 μg/mL streptomycin (Lonza). Differentiation to the axenic amastigote stage was performed as described previously. [34] Briefly, axenic amastigotes were cultivated at 32°C in Schneider’s medium pH 5.5 supplemented with 10% FBS, 100 U/mL penicillin and 100 μg/mL streptomycin (complete Schneider’s media pH 5.5). The human monocyte cell line THP-1 [35] was maintained in vitro by culturing at 37°C with 5% CO2 in Dutch modified RPMI-1640 (Gibco) containing 10% FBS and 2 mM L-glutamine (Gibco) (complete RPMI media). Differentiation of THP-1 cells into macrophages was performed by seeding 2.5 x105 cells/mL in complete RPMI media, supplemented with 20 ng/mL phorbol 12-myristate 13-acetate (PMA). [36] Cells were incubated at 37°C with 5% CO2 for 24 hours.
NanoLuc, NanoLuc-PEST and red-shifted firefly luciferase (PRE9) open reading frames were amplified by PCR from plasmid DNA templates: pNL1.1, pNL1.2 (Promega) and pCMV-Red Firefly Luc (Thermofisher) respectively. All oligonucleotide sequences are provided in S1 Table. Amplified genes were digested with BamHI and KpnI and ligated into pSSU-Neo [37] to produce the constructs pSSU-NanoLuc, pSSU-NanoLuc-PEST and pSSU-PRE9, for constitutive expression in L. mexicana. The pSSU expression vector contains flanking regions for integration into the rDNA locus of the parasite genome, and has been used in a number of Leishmania strains. [38–41] The three constructs (PacI/MssI digested) were transfected into mid-log L. mexicana procyclic promastigotes by nucleofection using a 4b Nucleofector system (Lonza), as described previously. [42] Transfectants were transferred to M199 agar plates containing 40 μg/ml Geneticin (Life Technologies) and clonal cell lines were established. Integration of the construct into the genome was assessed by PCR amplification of 50–100 ng total parasite DNA, using the oligonucleotide primers pSSU-F (region of the 18S gene) and pSSU-R (splice acceptor site in the pSSU vector). Total parasite DNA was purified from mid-log promastigote cells using the DNeasy Blood and Tissue Kit (Qiagen).
Parasite growth was counted at 24 hour intervals for 120 hours using a haemocytometer. Statistical significance was determined using a two way ANOVA, with the Dunnett Post-Hoc test for multiple comparisons (GraphPad Prism 6.0).
Detection of NanoLuc and NanoLuc-PEST enzymes was performed using the lytic Nano-Glo Assay (Promega), according to the manufacturer’s instructions. Detection of PRE9 was performed using either the Firefly Luciferase Glow Assay (Pierce), or the Bright-Glow Assay (Promega). Bioluminescence was measured using the Promega GloMax Multi Detection System. All data was normalised by subtracting bioluminescence values for the untransformed, parental line, and all assays done in triplicate.
Cycloheximide assays were performed over 8 hours. Cells were seeded to a cell density of 5 x105 cells/well (100 μL/well). Cycloheximide (Sigma) was added to each well to a final concentration of 100 μM and incubated with cells for a range of time points between 0 and 8 hours. Bioluminescence was measured as described above, and the half-life of each luciferase was calculated by one phase decay non-linear regression (GraphPad Prism 6.0). Proteosome assays were performed using the method above with the addition of 5 μM MG-132 (Sigma) for 6 hours. [43] Statistical significance was determined using a paired, two-tailed T-test (GraphPad Prism 6.0).
Toxicity assays were performed over 72 hours. THP-1 cells were differentiated into adherent macrophages, as described above. Cells were washed once in PBS to remove non-adherent monocytes, and compounds were added at varying dilutions. Cells were incubated with the compounds at 37°C with 5% CO2 for 72 hours. PrestoBlue (ThermoFisher) was added at a dilution of 1:10 per well. Plates were incubated in the dark at 37°C with 5% CO2 for 6 hours before reading on a Promega GloMax Multi Detection System (λex/λem = 525/580-640 nm).
For infection assays, the THP-1 cells were differentiated into adherent macrophages, as described above. Stationary phase L. mexicana metacyclic promastigotes were added at a ratio of 10:1 (parasites:macrophages) in complete RPMI media, and incubated at 32°C with 5% CO2 for 24 hours. Adherent macrophages were washed three times with PBS to remove extracellular parasites. Cells were then treated with 0.8 μM Amphotericin B (Fungizone, ThermoFisher) for 72 hours. Parasite load was determined using two methods; bioluminescence and indirect immunofluorescence. Bioluminescence was measured using the protocols described above. For indirect immunofluorescence, cells were fixed with 4% (v/v) formaldehyde (ThermoFisher) for 10 minutes. Cells were permeabilised with 0.1% (v/v) Triton-X 100 (Sigma) for 5 minutes, then blocked for 30 minutes with Image iT FX Signal Enhancer (Life Technologies). Cells were incubated with anti-HASPB [44] (1:250) for 1 hour, washed three times with PBS, then incubated with Alexa Fluor 594 conjugated goat-anti-rabbit IgG (Invitrogen) for 1 hour. Slides were washed three times with PBS, then mounted using Prolong Diamond Antifade Mountant (Life Technologies). Images were acquired using the EVOS FL Cell Imaging System (ThermoFisher), and parasite load assessed for a minimum of 100 macrophages per sample using the following equation:
InfectionIndex=Infectedmacrophages(%)*NumberofamastigotesTotalnumberofmacrophages
The initial MMV Pathogen Box screen was performed on axenic amastigotes. Axenic amastigotes were seeded at a density of 1 x 105/ml in duplicate (50 μl/well) in complete Schneider’s media pH 5.5. MMV Pathogen Box screening was performed using two concentrations of each compound: 2 and 10 μM. Hits were defined as compounds that, at a concentration of 2 μM, decreased the relative bioluminescence signal to 5% or less of that produced by the transgenic parasites incubated in solvent (DMSO) only. Both positive (Amphotericin B) and negative (equivalent volume of DMSO) controls were included on all plates. Assays were carried out as technical replicates with two independent assays performed (n = 4). Cells were incubated at 32°C with 5% CO2 for 72 hours, prior to viability measurements using either fluorescence or bioluminescence-based assays. For the fluorescence-based assays, AlamarBlue (ThermoFisher) was added at a dilution of 1:10 per well. Plates were incubated in the dark at 32°C with 5% CO2 for 6 hours before measuring the fluorescence on a Promega GloMax Multi Detection System (λex/λem = 525/580-640 nm). For bioluminescence, 20 μl of treated axenic amastigote culture was transferred in duplicate to a white 96-multiwell plate (Greiner, UK) and 20 μl of luciferase reagent (Nano-Glo Luciferase Assay buffer and Nano-Glo Luciferase Assay substrate, 200:1) was added to each well. After 3 minutes, the bioluminescence signal was measured using a Promega Glomax Multi Detection System. Results were analysed using GraphPad Prism 6.
For all assays, the percentage viability was calculated using the following equation:
Viability(%)=100x[μ(s)−μ(−)][μ(+)−μ(−)]
Where μ(s) = mean value for the sample, μ(+) = mean of the DMSO only control, and μ(-) = mean of the positive control (2 μM Amphotericin B). For each screening assay, the assay quality parameters Zʹ score and signal:background ratios were calculated as described previously. [45]
A panel of transgenic L. mexicana lines was generated that constitutively express either NanoLuc, NanoLuc-PEST or PRE9. Correct integration of the luciferase genes into the L. mexicana genome was assessed by PCR on total parasite DNA using primers complementary to the rRNA locus and the exogenous DNA (S1 Fig). A product of the expected size (900 bp), consistent with integration in the rRNA locus, was amplified from transgenic lines expressing NanoLuc-PEST and PRE9, but not from the NanoLuc cell line (S1A Fig). Amplification of the respective luciferase open reading frame was successful for all three lines (S1B Fig), suggesting that the introduced DNA in the NanoLuc line is episomal rather than integrated. The generation of an integrated NanoLuc cell line was not pursued as we found that the NanoLuc-PEST luciferase was a more dynamic indicator of cell viability (see below). Growth of the transgenic parasite lines was assessed by monitoring procyclic cells over a 5 day time period, and compared to the parental strain (Fig 1A, S2A Fig). There were comparable growth rates in all parasite lines over this period, showing no gross defects in fitness in the transgenic lines.
Expression of the luciferase enzymes was analysed by measuring bioluminescence in lysed cells. There was a significant linear correlation between cell number and bioluminescence over a wide range (5–500,000 cells per assay) in all luciferase-expressing cell lines, in both procyclic promastigotes and axenic amastigotes (Fig 1B and 1C respectively, S2B Fig). Luciferase activity (above background levels) was detected at less than 10 cells per well for all three luciferase enzymes, highlighting the sensitivity of the bioluminescence-based approach. Cells expressing NanoLuc showed considerably higher bioluminescence than those expressing the other two luciferases, with levels over 100-fold higher than the equivalent cells expressing PRE9 (S2 Fig). In comparison, the bioluminescence produced by NanoLuc-PEST was over 10 fold brighter than PRE9 in axenic amastigotes, whilst the signal intensity was comparable between the two luciferases in promastigotes (Fig 1B and 1C).
The half-life of each luciferase was determined by treatment with a supralethal dose of cycloheximide (100 μM), a known eukaryotic protein translation inhibitor. [46] Following an 8 hour incubation with cycloheximide, the bioluminescent signal from NanoLuc-expressing cells in both life cycle stages remained relatively constant, showing that this reporter is extremely stable (S2C Fig). In contrast, there was a 10-fold decrease in the bioluminescent signal after 1 hour in promastigotes and axenic amastigotes from the NanoLuc-PEST-expressing line (Fig 1D and 1E). DMSO, the solvent for cycloheximide, has no effect on bioluminescence levels at the volume used in this assay (S3 Fig). The calculated half-life for each of the luciferase enzymes was calculated to be >8 hours for NanoLuc (both parasite forms), 16 and 9 minutes for NanoLuc-PEST (axenic amastigotes and promastigotes, respectively) and 163 and 261 minutes for PRE9 (axenic amastigotes and promastigotes, respectively). Proteasome targeting of the NanoLuc-PEST enzyme was assessed in the presence of the proteasome inhibitor MG-132 (S4 Fig). [43] Addition of 5 μM MG-132 produced a statistically significant increase in bioluminescence relative to the DMSO control (S4A Fig; p = 0.0272 and 0.0007 in axenic amastigotes and promastigotes respectively). In the presence of both MG-132 and cycloheximide, bioluminescence values did decrease, but were still significantly higher than in the presence of cycloheximide alone (S4B Fig; p = 0.0219 and 0.0010 in axenic amastigotes and promastigotes respectively).
In order to evaluate the NanoLuc-PEST enzyme as a dynamic reporter of anti-leishmanial activity, we compared this transgenic cell line against a standard resazurin-based fluorescent viability assay using Amphotericin B and Miltefosine (Table 1, S5 Fig). This was tested in axenic amastigotes. The robustness of the assays was assessed by calculating the Zʹ factor and signal:background (S:B) ratio. The EC50 values obtained from the parental and transgenic line using both bioluminescence- and fluorescence-based assays was similar (0.20–0.27 μM; Table 1, S5 Fig), and comparable to the EC50 value of 0.30 ± 0.02 μM previously described for L. mexicana axenic amastigotes against Amphotericin B. [47] All assays had a calculated Zʹ factor value of ≥ 0.64, demonstrating the robustness of each assay (defined as a Zʹ value greater than 0.5). [48] However, there were marked differences in the S:B ratio. The bioluminescence-based assay displayed a S:B ratio between 50- and 100-fold higher than the standard fluorescence-based assay on the same cell line (Table 1).
We then evaluated the potential of the NanoLuc-PEST cell line to determine parasite viability in an intramacrophage assay, and compared this to the standard microscopy-based counting assay. We observed a correlation between the bioluminescence- and microscopy-based methods (Fig 2); specifically, the infection index and the bioluminescent signal decreased in the presence of a supralethal dose of Amphotericin B by ≥99% (Fig 2).
Our findings indicate that the NanoLuc-PEST transgenic cell line is the most dynamic reporter of those evaluated here. In addition, it is sensitive and robust in both the axenic and intramacrophage assay formats. Consequently, we evaluated this system for compound screening. We used a two-step process to do this: i) identify hits against axenic amastigotes, ii) screen these hits using our bioluminescence-based intramacrophage assay. Prior to this, the assay was optimised to reduce the volume and cell concentration while maintaining a linear readout in a 96-well microplate format (S6 Fig). The MMV Pathogen Box resource, comprising 400 diverse drug-like molecules, was tested at two concentrations (10 μM and 2 μM) on axenic amastigotes (S7 Fig), and are summarised in Fig 3. Hits were defined as compounds that, at a concentration of 2 μM, decreased the relative bioluminescence signal to 5% or less of that produced by the transgenic parasites incubated in solvent (DMSO) only (Fig 3B). The complete dataset is depicted graphically in S7 Fig, and in S2 Table. A total of 23 hits were identified (Fig 3B), 3 of which were the reference compounds Buparvaquone, Mebendazole and Auranofin. Of these 23 identified hits, 52% originated from Mycobacterium tuberculosis screening programmes.
All 23 hits were analysed to determine their EC50 (S3 Table), except for Mebendazole (MMV003152), which could not be resolved. Eight of these compounds displayed an EC50 value less that that observed for Amphotericin B (0.2 μM). Of these 8 compounds, 2 were reference compounds (Buparaquone and Auranofin). We picked the most potent reference compound (Buparaquone) and the 6 test compounds to screen using the intramacrophage assay in parallel with Amphotericin B and Miltefosine (Table 2, Fig 4). There was an increase in the EC50 values from the intramacrophage assay for all 7 compounds when compared to the EC50 values from the axenic amastigote screen (Table 2, S3 Table and Fig 4). EC90 values from the intramacrophage assay are also included (Table 2). Of these 7 compounds, 5 are known to be active against kinetoplastids, with 4 of these active against Leishmania spp (Table 2). Our results for MMV690102 correlate well with the existing data. [49–51] However, our EC50 results for MMV689480 (Buparvaquone) in the intramacrophage assay is at least three times lower than the previously reported values against L. mexicana. [52] Our results for MMV688262 (Delamanid) and MMV595321 were higher than previously reported values against L. donovani (Table 2) [49, 53, 54].
Compounds supplied in the MMV Pathogen Box have been tested for cytotoxicity against human cell lines, and this information is available online (https://www.pathogenbox.org/about-pathogen-box/supporting-information). We have summarised this existing data alongside results from cytotoxicity screens against the THP-1 cell line for the compounds that we had sufficient available material (S4 Table). Tested compounds displayed an EC50 > 50 μM.
A key issue with the anti-leishmanial drug discovery process is the step from axenic amastigotes to in vivo models. The compromise–using an in vitro intramacrophage assay–involves the use of laborious and time consuming microscopy-based techniques that assess parasite burden, but cannot determine parasite viability. This paper describes the use of a tractable bioluminescent marker (NanoLuc-PEST) that correlates specifically with parasite viability. This method uses a simple and robust bioluminescence assay that decreases the time required to screen compounds against intramacrophage amastigotes in vitro, and provides an environment that should be more similar to the in vivo models.
All three of the luciferases tested enabled the detection of less than 10 cells per well (Fig 1B and 1C, S2 Fig), demonstrating the highly sensitive nature of the bioluminescence approach. Luciferase activity was not significantly greater in axenic amastigotes relative to promastigotes for NanoLuc and NanoLuc-PEST lines, although the inserted genes were fused to the CPS intergenic region, which has previously resulted in the upregulation of GFP in the amastigote stage. [59] However, the NanoLuc and NanoLuc-PEST luciferases displayed higher bioluminescence signals in the axenic amastigote form when compared to the PRE9 alternative (Fig 1C). There are several factors that may be involved in this. Specifically, the NanoLuc luciferase was the only one of the three tested that was not genomically integrated (S1A Fig), therefore it may have higher copy numbers compared to the NanoLuc-PEST and PRE9 enzymes. Differences in bioluminescence may also stem from the use of different substrates, specifically furimazine (for the NanoLuc and PEST derivative) and beetle luciferin (for PRE9). However, it is the sensitivity and brightness produced by the NanoLuc enzyme, and its PEST variant, that makes them highly attractive for screening purposes.
Signal intensity is not the only factor that should be taken into consideration when assessing a reporter molecule. The signal must also correlate with cell viability. The NanoLuc version of the enzyme is very stable, with a half-life greater than 8 hours (S2 Fig). The addition of the PEST domain, which is comprised of a proline, glutamic acid, serine and threonine rich sequence, [60] targets the enzyme for degradation by the 26S proteasome (S4 Fig). [61] This makes the NanoLuc-PEST variant a more dynamic reporter for cell viability. Whilst the Zʹ values for both the bioluminescence and fluorescence based assays were above 0.5 (the lower limit for an acceptable screening assay [48]), the S:B ratios for the NanoLuc-PEST transgenic cell line was 50- to 100-fold higher than the standard fluorescence-based assay (Table 1). The higher S:B ratio of NanoLuc-PEST reflects the relatively high enzymatic activity and short half-life of this protein, contributing to the greater dynamic range achieved using this modified reporter. Consequently, it was the clonal, genomically integrated NanoLuc-PEST cell line that we felt was most appropriate for further testing.
The main advantage with this bioluminescent technique is its potential for use with intramacrophage assays. The current gold standards for intramacrophage compound screening are microscopy-based assays. The microscopy techniques use nuclear staining, parasite-specific antibodies or stable reporter molecules. [11–15, 36, 62] These rely on the use of fluorescence, which can be affected by potential autofluorescence of test compounds. Furthermore, these methods rely solely on detecting the presence of the parasite, not determining parasite viability. We directly compared the potential of the NanoLuc-PEST cell line using the bioluminescence- and counting-based assays on paired samples. Our results from the bioluminescence-based assay mirrored that of the microscopy-based assay (Fig 2); specifically the level of infection decreased by ≥99%. Our bioluminescent intramacrophage assay is not only more sensitive, but has the potential to be automated, and would therefore allow large-scale screening of compound libraries using a more relevant in vitro assay. The NanoLuc-PEST expressing cell line thus provides a useful tool to assess compound efficacy against intracellular parasites without the need for arduous, less sensitive, microscopy-based assays. Expressing the NanoLuc-PEST protein in other Leishmania species for screening purposes is also possible, as the pSSU-Int expression vector has been used successfully in L. donovanni, [39] L. major, [37, 40] and L. infantum [37, 38, 41].
One limitation of the bioluminescent assay is that it requires cell lysis. This means that the cells, once measured, cannot then be used in additional, downstream applications. However, the volumes required for lysis is small (<100 μL per technical replicate), so an aliquot can be taken and tested from a larger sample, leaving the remaining material for further analysis. A second limitation is that any extracellular promastigotes present within the well will produce a signal when lysed. Whilst every care was taken to ensure the macrophages were washed carefully, we cannot guarantee that all of the extracellular promastigotes were removed. However, incubation at 32°C should induce differentiation into axenic amastigotes, which are less likely to be retained on the macrophage surface during subsequent wash steps.
We then tested the NanoLuc-PEST cell line against the MMV Pathogen Box. The EC50 values obtained from the intramacrophage assay were higher than those values obtained from the screen of the axenic amastigotes alone (Table 2). This is perhaps expected, as the compound must traverse two additional membranes before it reaches the Leishmania amastigote. However, all seven compounds displayed an EC50 < 6 μM, which is below the 10 μM limit detailed for hits against intracellular L. donovani. [63] Of these seven compounds, five are known to be active against kinetoplastids, with four active against Leishmania spp (Table 2). Our results for MMV690102 correlate well with the existing data, despite this previous data being gathered against L. donovani. However, our EC50 results for MMV689480 (Buparvaquone) in the intramacrophage assay is at least three times lower than the previously reported values against L. mexicana. This may indicate the increased sensitivity of our intramacrophage assay, as only viable parasites within the macrophage are detected. Our results for MMV688262 (Delamanid) and MMV595321 were higher than previously reported values (Table 2), which may be due to species variation.
In conclusion, we have demonstrated the use of transgenic L. mexicana expressing a novel luciferase as a tractable, rapid and sensitive system for compound screening, using the MMV Pathogen Box as proof of principle. This system has the added advantage of allowing detection of parasite viability in an in vitro infected macrophage model, a method that has previously been shown to be advantageous for studying Plasmodium falciparum and Mycobacterium tuberculosis in an intracellular environment. [16, 17] This allows screening programmes to assess compound activity against the intracellular parasite without the time burden, requirement for specialist equipment and post-assay processing associated with the current, microscopy-based, techniques.
|
10.1371/journal.pgen.1005820 | Post-transcriptional Control of Tumor Cell Autonomous Metastatic Potential by CCR4-NOT Deadenylase CNOT7 | Accumulating evidence supports the role of an aberrant transcriptome as a driver of metastatic potential. Deadenylation is a general regulatory node for post-transcriptional control by microRNAs and other determinants of RNA stability. Previously, we demonstrated that the CCR4-NOT scaffold component Cnot2 is an inherited metastasis susceptibility gene. In this study, using orthotopic metastasis assays and genetically engineered mouse models, we show that one of the enzymatic subunits of the CCR4-NOT complex, Cnot7, is also a metastasis modifying gene. We demonstrate that higher expression of Cnot7 drives tumor cell autonomous metastatic potential, which requires its deadenylase activity. Furthermore, metastasis promotion by CNOT7 is dependent on interaction with CNOT1 and TOB1. CNOT7 ribonucleoprotein-immunoprecipitation (RIP) and integrated transcriptome wide analyses reveal that CNOT7-regulated transcripts are enriched for a tripartite 3’UTR motif bound by RNA-binding proteins known to complex with CNOT7, TOB1, and CNOT1. Collectively, our data support a model of CNOT7, TOB1, CNOT1, and RNA-binding proteins collectively exerting post-transcriptional control on a metastasis suppressive transcriptional program to drive tumor cell metastasis.
| The majority of human cancer related death is due to the effects of metastasis, the process of cancer dissemination to and growth in distant organs. Primarily due to its complexity, the metastatic process remains incompletely understood. This complexity stems from the tumor cell’s dependence on multiple cellular and molecular systems for the successful colonization of distant sites. In this study, we demonstrate that one of the factors that contributes to metastatic progression is the control of tumor cell RNA stability. Previously, we demonstrated that a structural component of the CCR4-NOT transcription regulatory complex was an inherited metastasis susceptibility gene. Here we demonstrate that one of the enzymatic components of the CCR-NOT complex, Cnot7, is also a metastasis-associated gene, and that enzymatic activity of Cnot7 is required for its promotion of metastatic disease. These results suggest that large-scale control of RNA abundance may be modulating specific metastasis-related transcriptional programs, and that inhibition of specific RNA deadenylases may be a viable avenue in the development of anti-metastatic therapeutics.
| Metastasis is a complex process in which tumor cells disseminate from the primary tumor site to form life-threatening lesions at distant sites. To successfully complete the metastatic process tumor cells must activate a series of molecular functions. These include motility and invasion to escape the primary site and penetrate the parenchyma at the secondary organ, anti-apoptotic programs to survive anoikis during transit through the lymphatic or hematic vasculature, and proliferative programs to establish clinically relevant macroscopic lesions [1]. Each of these programs requires the action of multiple genes in a coordinated fashion. As a result, control of transcriptional programs in the metastatic cascade has been the focus of many studies over the past decade. For example, activation of embryonic programs through up-regulation of transcription factors is thought to be important in the migratory and invasive steps of the metastatic cascade [2, 3]. Post-transcriptional control of metastasis-associated genes by microRNAs has also been the subject of a variety of studies [e.g. [4, 5]]. Activation or suppression of these pleiotropic factors, through mutation, amplification, or deletion, therefore plays critical roles in tumor evolution and progression.
In addition to activation or suppression of whole transcriptional programs, factors that significantly alter transcriptional units might also alter the ability of a tumor cell to complete one or more of the steps of the metastatic cascade. Studies in recent years have demonstrated an important role for inherited polymorphism in gene expression programs [6, 7] suggesting that inherited factors can significantly influence tumor phenotypes. Our laboratory previously demonstrated that inherited polymorphisms significantly influence metastatic outcome [8] and that inheritance plays a role in the establishment of transcriptional profiles that discriminate patient outcome [9]. More recently we have integrated gene expression analysis and susceptibility genetics studies to identify co-expressed transcriptional networks associated with metastatic disease. One such network was centered on Cnot2, a scaffolding component of the CCR4-NOT RNA deadenylase complex [10]. In vivo validation studies demonstrated that varying CNOT2 levels significantly influenced tumor metastatic capacity and implicated the CCR4-NOT complex as a novel determinant of tumor cell metastatic potential [10].
The CCR4-NOT complex is a modular, multifunctional protein complex highly conserved in eukarya [11]. Components of CCR4-NOT are found both in the nucleus and cytoplasm, and mediate transcriptional and post-transcriptional regulatory functions [12, 13]. In mammalian cells, CCR4-NOT has reported roles in epigenetically mediated transcriptional regulation [14], nuclear hormone receptor-mediated transcription [15], and initiation of transcript decay by deadenylation [16–18]. These observations suggest that the CCR4-NOT complex is a pleiotropic regulator of transcript abundance.
Cnot2 depletion has been shown to disrupt CCR4-NOT deadenylase activity [19] which may be expected to alter metastasis-associated transcriptional programs. The absence of CNOT2 catalytic activity led us to hypothesize that other CCR4-NOT effector functions may drive metastasis. Moreover, the co-expression network analyses that identified Cnot2 also implicated the CCR4-NOT deadenylase Cnot8 and its binding partner Tob1 as candidate metastasis driving genes [10], suggesting that CCR4-NOT deadenylase function may be an important determinant for metastatic progression.
Deadenylation, the progressive 3’-to-5’ shortening of the polyA tail, is a rate-limiting step of transcript destabilization [20]. Translational inhibition and transcript decay mediated by microRNAs [21], AU-rich elements [22], RNA binding proteins [23–25], and nonsense-mediated decay [26] occur through the recruitment of deadenylase complexes. The CCR4-NOT complex therefore plays an important role in maintaining mRNA equilibrium and coordinated control of transcriptional programs. We previously demonstrated that modulation of transcriptional elongation, mediated by Brd4 [27], significantly altered the metastatic capacity of mammary tumor cells. In this study we extend these results by demonstrating that transcriptional decay, mediated by the deadenylation activity of Cnot7 is also an important determinant in tumor progression. Our findings are consistent with the existence of post-transcriptional regulatory deadenylase complexes that promote metastasis by destabilizing metastasis suppressive transcriptional programs. Importantly, our work identifies CNOT7 deadenylase activity as a novel therapeutic target for anti-metastatic therapy.
Metastasis assays by orthotopic implantation were performed to test if perturbation of Cnot7 or Cnot8 expression alters metastatic capacity. Cnot7 was knocked down in three independent murine mammary tumor cell lines [28, 29] by stable transduction of short hairpin RNAs (shRNAs), resulting in reductions of protein and transcript abundance (Fig 1). Cnot8 was knocked down with the same method in 6DT1 and Mvt1 cells. After selection, transduced cells were implanted into the fourth mammary fat pad of syngeneic mice. At assay endpoint (t = 30 days), Cnot8 knockdown did not produce consistent results between the tested cell lines (S1 Fig) and therefore was not included in further studies. Cnot7 knockdown consistently reduced pulmonary metastasis without significant effect on primary tumor mass in vivo (Fig 1).
To validate these results in an independent experimental system we next assessed the effect of diminished Cnot7 expression in an autochthonous transgene-driven metastasis model. The MMTV-PyMT transgenic mammary tumor model [30] was bred to Cnot7 hemizygous knockout [31] mice to produce PyMT+ Cnot7+/- and PyMT+ Cnot7 wild type mice (Fig 2A). Quantitative real time polymerase chain reaction analysis confirmed that Cnot7+/- mice expressed Cnot7 transcript approximately two-fold lower than wildtype mice in the spleen and tumor (Fig 2B & 2C). Consistent with observations in the orthotopic metastasis model, deletion of one copy of Cnot7 significantly reduced the metastatic incidence and burden with no effect on primary tumor mass (Fig 2D–2F).
To test if reduction of Cnot7 expression in the stroma influenced metastasis, we then crossed C57BL/6J Cnot7+/- to FVB/NJ or Balb/cJ mice to generate Cnot7+/- and Cnot7+/+ mice that are immune-tolerant to the FVB- or BALB-derived tumor cells, respectively (Fig 3A). Wild type tumor cells were then orthotopically implanted into Cnot7+/- and Cnot7+/+ mice and animals were aged for 28 days prior to necropsy. No consistent difference in tumor mass or metastasis was observed (Fig 3B–3J) suggesting the primary role of Cnot7 in metastatic progression is tumor cell-autonomous.
Tumor cell colonization of distant organs is thought to be the rate-limiting step of the metastatic cascade [32, 33]. Lung colonization assays were therefore performed by intravenous injection of Cnot7-depleted tumor cells into the tail vein of mice (Fig 4A). Three of four Cnot7 knockdown conditions resulted in significant suppression of lung colonization (Fig 4B–4D). Cross-sectional area of metastatic lesions was subsequently measured in hematoxylin and eosin (H&E) stained lung sections to determine if differences in colonization occurred secondary to proliferative differences. No difference in metastatic size between control and Cnot7-depleted cells was observed for either 6DT1 or 4T1 cells (Fig 4E & 4F). These results are consistent with a role of Cnot7 promoting early stages of lung colonization in a proliferation-independent manner.
To investigate the potential cellular mechanisms underlying the suppression of metastatic capacity upon Cnot7 depletion proliferation, motility, and colony formation in low attachment conditions were assessed. Cnot7 depletion resulted in a consistent reduction of cellular proliferation in all three cell lines tested (S2 Fig) although, as noted above, this proliferative suppression was not observed in the in vivo orthotopic implantation assays.
Motility assays were performed on Cnot7 depleted 6DT1 and Mvt1 cells. Cnot7 knockdown in Mvt1 cells resulted in a reduced motility in a wound healing assay. In contrast, Cnot7 knockdown in 6DT1 using the same shRNA constructs showed no consistent difference in motility as measured by wound healing assay (S2 Fig). Similar discrepancies across cell lines were observed in soft agar assays, where Cnot7 depletion in 6DT1 resulted in significant reduction of colony formation while no difference was observed in Mvt1 cells (S2 Fig). Due to the lack of consistency among the in vitro assays, wound healing and soft agar assays were not performed for 4T1 Cnot7 depleted cell lines. Overall the ambiguous results of the in vitro assays suggested that further investigations into the mechanisms of the role of Cnot7 in metastatic disease would be best examined in vivo. Further efforts therefore focused on metastatic capacity based on orthotopic transplant assays.
The CCR4-NOT complex has been implicated in multiple cellular functions, including RNA deadenylation and degradation as well as transcriptional control [12]. To determine whether CNOT7-mediated metastasis promotion was deadenylase-dependent, we stably expressed the wild type and a deadenylase-inactive point mutant (D40A) [34–36] in mammary tumor cell lines. Transduced cultures were selected for approximately equal expression of CNOT7 and D40A protein (Fig 5A–5C), and used for in vivo orthotopic transplant assays.
Ectopic expression of CNOT7 or the D40A mutant showed no differences in primary tumor mass in Mvt1 or 6DT1 cells (Fig 5D & 5F). However, CNOT7 but not D40A significantly promoted metastatic burden for 6DT1 and Mvt1 (Fig 5G, 5I, 5J and 5L). In 4T1 cells, a statistically significant increase in tumor mass was observed for CNOT7 but not for D40A expressing cells (Fig 5E). Normalization of metastatic burden by tumor mass to account for difference in primary tumor growth still resulted in a significant difference in metastatic capacity for CNOT7 expressing 6DT1 or 4T1 cells and borderline significance in Mvt1 cells. Furthermore, although the 4T1 D40A expressing cells exhibited increased metastasis compared to control, a significant reduction of metastatic capacity was observed compared to CNOT7-wild type expressing cells (Fig 5H and 5K). Overall these results are consistent with a major role of the deadenylase function of CNOT7 in modulating metastatic capacity of mammary tumor cell lines.
CNOT7 is a non-specific RNA-binding deadenylase protein [35]. Specificity for transcripts is mediated by sequence specific RNA binding proteins that interact with the CCR4-NOT complex. TOB1 is an adaptor protein that recruits CNOT7 to specific RNA-binding proteins, while CNOT1 is a scaffolding protein for the CCR4-NOT complex [37, 38] (Fig 6A and 6B). Previous work from our laboratory found that Tob1 expression was correlated with metastasis in the [PyMT x AKXDn]F1 mice [10] (Fig 6C). In addition, human breast cancer datasets–available through the Gene expression-based Outcome for Breast cancer Online (GOBO) database, an expression array-based meta-analysis data set of 1,881 breast cancer patients [39]–also showed that high expression of either TOB1 or CNOT1 correlated with poor survival (Fig 6D & 6E). Furthermore TOB1 has previously been associated with poor distant metastasis free survival in breast cancer patients [40].
Orthotopic metastasis assays were conducted to test the role of the CCR4-NOT adaptor proteins in metastatic disease. Attempts to generate stable Cnot1 knockdown cells were unsuccessful and therefore orthotopic assays were not performed. Tob1 knockdown in 6DT1 cells (Fig 6F) showed diminished metastasis with no effect on primary tumor mass (Fig 6G–6J). Experimental metastasis assays showed that Tob1 knockdown suppressed lung colonization of 6DT1 cells (Fig 6K), consistent with Tob1 acting at the same stage of the invasion-metastasis cascade as Cnot7. In 4T1 cells, Tob1 knockdown suppressed tumor mass and pulmonary metastasis (S3 Fig). Significant suppression of metastasis was observed after normalizing by tumor mass, consistent with an effect other than just on tumor growth (S3 Fig).
To test if CNOT7-mediated metastasis promotion was dependent on a complex with TOB1 or CNOT1, we constructed expression vectors of CNOT7 mutants that disrupted interaction with TOB1 or CNOT1. CNOT7 E247A/Y260A mutations have previously been shown to disrupt interaction with BTG/TOB family proteins but maintain interaction to CNOT1 [36]. Conversely, CNOT7 M141R substitution abolishes complex formation between CNOT7 and CNOT1 [34], but interaction with TOB1 has not previously been assessed. Co-immunoprecipitation (IP) experiments were therefore performed to address this question. As expected, IP of CNOT7 E247A/Y260A but CNOT7 M141R co-precipitated CNOT1. In contrast, the CNOT7 M141R mutant retained the ability to co-precipitate TOB1 but no longer interacted with CNOT1 (Fig 7A & 7B), indicating that the two mutant constructs specifically disrupted interaction with the TOB1 or CNOT1 adaptor proteins.
CNOT7 and the mutant constructs were then expressed in 4T1 cells to achieve equal levels of CNOT7, CNOT7 E247A/Y260A, and CNOT7 M141R protein (Fig 7C) and the cells were then implanted orthotopically into mice. No significant changes in the rate of tumor growth or primary tumor mass at endpoint were observed (Fig 7D). Consistent with previous observations, overexpression of CNOT7 promoted tumor cell metastatic potential but expression of CNOT7 E247A/Y260A or CNOT7 M141R showed no change in metastasis compared to control (Fig 7E and 7F). These results indicate that CNOT7-mediated metastasis promotion depends on contact with both TOB1 and CNOT1.
The above results suggest that Cnot7 mediates its effect on metastasis by modulating the RNA equilibrium. Therefore, to gain a better idea of the global gene expression program affected by Cnot7, we identified mRNAs that exhibited an inverse relationship with Cnot7 expression in 4T1 cells in which Cnot7 was knocked down or over-expressed. Array-based transcriptome analysis yielded 842 significantly dysregulated transcripts (p<0.01, S1 Table). Of these, 514 transcripts were upregulated upon Cnot7 knockdown and down-regulated upon CNOT7 overexpression (t<0, S1 Table; Cnot7-anticorrelated transcripts). 3’ untranslated regions (3’UTRs) were then interrogated for known consensus RNA-binding protein (RBP) sequence motifs enriched in the Cnot7 inversely-correlated transcripts (S4 Fig). The inversely correlated transcripts showed enrichment for the cytoplasmic polyadenylation element (CPE) [41], Pumilio binding element (PUM) [42], Nanos response elements (NRE) [43], and cleavage and polyadenylation stimulation factor binding element (CPSF) [44]. In contrast neither permissive (AUUUA) nor stringent (UUAUUUAUU) AU-rich elements (ARE) [45] were found to be enriched indicating that, in 4T1 cells, Cnot7 preferentially mediates the degradation of a specific subset of mRNAs (S4 Fig).
To identify transcripts directly regulated by CNOT7, RNA-immunoprecipitation was performed using the anti-FLAG (M2) antibody in 4T1 cells overexpressing FLAG-CNOT7. Co-precipitated RNA was subjected to high throughput sequencing (RIP-seq). 149 transcripts showed enrichment relative to input and control (M2 RIP in 4T1 empty vector cells not expressing FLAG-CNOT7) and showed an inverse correlation with Cnot7 expression (Fig 8A). These transcripts showed 3’UTR enrichment for CPE, CPSF, and NRE sites (Fig 8B). PUM and ARE sites were not considered since fewer than 10 of the 149 genes contained these binding motifs. Seventy-one of 149 transcripts (48%) possessed CPE, CPSF, and NRE sites (Fig 8C, S2 Table) suggesting that CPEB, CPSF, and Nanos family proteins may collectively constitute specificity factors that cooperatively drive metastasis by targeting CNOT7 to metastasis-associated transcripts.
We next tested if the 71 transcripts that shared the tripartite motif were prognostic in human breast cancer data sets. Forty-six (65%) human orthologs of the CPE/CPSF/NRE containing genes were present in GOBO [39] (S3 Table). These 46 genes were applied as a gene signature, weighted by their inverse correlation with CNOT7 expression, to determine whether they could discriminate patient outcome. Consistent with the possibility that CNOT7 drives progression by degrading metastasis suppressing mRNAs, high expression of the signature was correlated with favorable distant metastasis free survival (DMFS, Fig 8D).
Since CNOT7-bound and inversely regulated transcripts were reproducibly enriched for the CPE/CPSF/NRE motifs, we speculated that this tripartite motif may specify CNOT7 target transcripts, which are enriched for metastasis-associated genes. We thus interrogated the entire genome for transcripts that possessed CPE, CPSF, and NRE 3’UTR elements, and filtered this set for those transcripts expressed in 4T1 cells. This analysis identified 3424 genes (12.5% of the mouse genome, Fig 8E). Filtering this set for those inversely-correlated with Cnot7 expression yielded 293 genes (S4 Table). Human orthologs for 217 (74%) of the 293 genes were available in GOBO, and were able to significantly discriminate DMFS (Fig 8F). High expression of this signature predicted favorable survival, suggesting that this set of tripartite motif containing, Cnot7-anticorrelated transcripts constituted a metastasis suppressive post-transcriptional program.
Subjecting this 293-gene set to Ingenuity Pathway Analysis identified cancer as the most highly represented disease annotation, with 247 (84%) transcripts previously annotated as cancer-associated. The top five represented canonical pathways associated with extravasation signaling, and cancer-associated HER2 signaling, HGF signaling, and MAPK-signaling. The top five upstream regulators in this network included Kras, Erbb2, and Tgfb1 signaling (S5 Table). Gene sets regulated by each of these upstream regulators predicted distant metastasis free survival (S5 Fig). Consistent with our model, Erbb2 and Kras are previously reported upstream signaling components that regulate the activity of CNOT7 adaptor and deadenylation cofactor TOB1, transducer of ERBB2 [46, 47].
CCR4-NOT is a highly conserved protein complex that has been implicated in diverse functions associated with gene regulation [12, 13]. It exists in both cytoplasmic and nuclear forms and is thought to play different roles depending on its subcellular localization and subunit composition [12]. In the nucleus the CCR4-NOT complex has been implicated in numerous activities, including chromatin modification, transcriptional elongation, RNA export, nuclear RNA surveillance and transcription-coupled DNA repair [14, 15]. In the cytoplasm the CCR4-NOT complex is thought to be the main RNA deadenylase, initiating both mRNA decay and translational repression by polyA tail shortening [12, 13]. The CCR4-NOT complex also participates in miRNA-mediated gene silencing through interactions with the GW182/Argonaute complex [21]. This large, multi-functional protein complex therefore has the potential to play a variety of important roles in establishing and maintaining cellular function and response to extracellular cues.
Previously our laboratory implicated the CCR4-NOT complex as an important determinant for metastatic mammary cancer. Generation of co-expressed gene maps from mouse strains with differing inherited sensitivity for pulmonary metastasis identified a network module centered on the CCR4-NOT component Cnot2 that was capable of discriminating breast cancer patient outcome. Furthermore, in vivo modeling demonstrated that suppression or over-expression of CNOT2 within tumor cells resulted in enhanced or reduced pulmonary metastases, respectively, indicating that Cnot2 has metastasis suppressing activities [10]. CNOT2 however does not have known enzymatic activities [48]. CNOT2 coordinates the interaction of CNOT3 with the core CCR4-NOT complex, as well as additional regulatory molecules including HDAC3 [14]. Despite lacking catalytic function, CNOT2 has been shown to be an important positive regulator of CCR4-NOT deadenylase activity [19], cellular apoptosis, and mouse embryonic stem cell pluripotency [49]. The contribution of Cnot2 to metastatic capacity through the CCR4-NOT complex could therefore occur through a variety of CCR4-NOT molecular functions.
In this study we have begun to dissect the role of CCR4-NOT in mammary tumor metastasis by investigating the role of RNA deadenylation in tumor progression. The integrated genetics and gene expression analysis that initially identified Cnot2 also implicated other genes associated with RNA deadenylation (Cnot8, Angel2, Tob1) as potential modulators of metastatic disease, suggesting that this function of CCR4-NOT might be a critical determinant [10]. Bioinformatics analysis indicated that CNOT8 and TOB1 were associated with distant metastasis free survival in human patients. We therefore selected Cnot8 and its highly conserved paralog Cnot7 to determine whether the deadenylation function of CCR4-NOT plays a critical role in tumor progression.
Cnot7 and Cnot8 are members of the DEDD superfamily of deadenylases [35]. Both genes are expressed ubiquitously in tissues of adult animals [50] and are thought to have overlapping functions [36]. Biochemical studies suggest that only one of the two proteins exist in the CCR4-NOT complex at a time, suggesting unique functions for the mutually exclusive CNOT7- or CNOT8-containing complexes, in addition to redundant functions [34, 51]. This interpretation is consistent with the differing results observed for the shRNA knockdowns of the two genes in our studies. Cnot7 knockdown had little or no effect on primary tumor growth, indicating that its role in tumor progression is related to the metastatic process. In contrast Cnot8 suppressed both primary tumor growth and metastatic disease, suggesting a more general role in regulating tumor cell proliferation. Further elucidation of the commonalities and differences in molecular pathways controlled by these two deadenylases would likely provide interesting insights into tumor growth and progression.
Importantly, due to the multifunctional nature of the CCR4-NOT complex, the ability of Cnot7 to promote metastatic disease was dependent on deadenylase activity. Point mutations eliminating enzymatic activity or that disrupted interactions mediating recruitment of RNA binding proteins to the CCR4-NOT complex suppressed the pro-metastatic activity of CNOT7. Knockdown of the adaptor protein TOB1, which acts as a bridge between CNOT7 and RNA binding proteins CPEB3 and CPEB4, which recruit RNAs to the complex for deadenylation, had similar effects. Attempts to knockdown CNOT1, which is responsible for recruitment of the NANOS1 and PUM2 RNA binding proteins, was unsuccessful. However, due to the central role of CNOT1 in the CCR4-NOT complex and increased apoptosis in CNOT1-depleted cells [52–54], this result was not unexpected. Taken together however, the point mutant and Tob1 depletion results suggest that the majority of the effect on metastasis was likely due to the influence of CCR4-NOT on RNA equilibrium or translational efficiency, rather than on the other many functions ascribed to the complex.
The CCR4-NOT complex is thought to be one of two general deadenylase complexes in mammalian cells [18]. However, the difference in phenotypes for Cnot7 and Cnot8 knockdowns suggest that the different complexes likely target overlapping subsets of RNAs within the cell [36]. Alternatively, the two paralogs may be differentially expressed and regulate different post-transcriptional programs in different cell types. To gain a better understanding of what subset Cnot7-containing CCR4-NOT complexes target global gene expression analysis was performed in two independent experiments. We focused specifically on genes that were inversely correlated with Cnot7 levels to enrich for those that were likely direct targets rather than those dysregulated due to secondary effects on the transcriptome. Since transcript recruitment to the CCR4-NOT complex is specified by RNA binding proteins a screen for RNA binding proteins was performed [23–25, 38, 55][56]. This screen revealed an enrichment of some but not all of the known RNA binding protein motifs, suggesting that the Cnot7 metastatic suppressive program is mediated by specific RNA binding partners. Further investigation of these RNA binding proteins and their RNA targets will likely provide additional insights into the molecular pathways important for metastatic progression.
Encouragingly enrichment of three of the four RNA binding protein motifs (CPE, CPSF and NRE) was replicated in the Cnot7 RIP-seq experiment, providing more direct evidence of the interaction of the CNOT7-containing CCR4-NOT complex with the Cpsf, Nanos and Cpeb families of RNA binding proteins. Furthermore, almost half (71/149) of the RNAs identified by the RNA-immunoprecipitation contained all three motifs. When used as a weighted gene signature, this set of transcripts was capable of discriminating metastasis outcome in human breast cancer. Global transcriptome analysis in 4T1 mammary tumor cells revealed only 293 expressed genes bearing all three RNA binding protein motifs, consistent with previous findings of only limited numbers of genes changing after Cnot7 knockdown [36]. Like the 71 genes identified by RIP-Seq, the 293 triple motif containing genes effectively discriminate outcome in breast cancer patients, suggesting enrichment of genes and molecular functions associated with breast cancer patients. This interpretation was further supported by pathway analysis, which identified previously known metastasis-associated functions such as extravasation and Tgfb1 signaling as enriched in the 293-gene set.
Overall this study supports the hypothesis that in addition to initiation of specific transcriptional programs, such as epithelial-to-mesenchymal transition, degradation of specific RNAs may play an important role in the establishment of metastatic capacity. Further investigations of the RNA binding proteins that recruit transcripts for deadenylation and studies into possible roles of the other CCR4-NOT deadenylase subunits Cnot6 and Cnot6l in metastatic progression may reveal additional important insights into tumor autonomous metastatic mechanisms. Moreover, these results also suggest that targeting Cnot7 deadenylase activity may be useful for anti-metastatic therapy. Cnot7 knockout animals are viable, with limited known phenotypes, indicating that pharmaceutical suppression of Cnot7 deadenylase activity may not be unacceptably toxic in the clinic. If true, this would provide a novel class of therapeutic agents to suppress colonization or the emergence of disseminated but dormant tumor cells, ultimately leading to a reduction in the morbidity and mortality associated with metastatic disease.
The research described in this study was performed under the Animal Study Protocol LCBG-004, approved by the NCI Bethesda Animal Use and Care Committee. Animal euthanasia was performed by cervical dislocation after anesthesia by Avertin.
The mouse mammary carcinoma cell lines 4T1, 6DT1, Mvt-1 [28] (provided by Dr. Lalage Wakefield) and human embryonic kidney HEK293 cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM) (Gibco) supplemented with L-Glutamate (Gibco), 9% fetal bovine serum (FBS) (Gemini BioProducts), and 1% Penicillin and Streptomycin (P/S) (Gemini BioProducts).
Two milliliter suspensions of 105 cells were incubated at 37°C in 5% CO2 overnight. Cells were then infected with lentivirus suspension, and selected 30 hours post-infection with 5mg/mL blasticidin for over-expression (Invitrogen) constructs or 10ug/mL (shRNA) puromycin for shRNA constructs.
Co-immunoprecipitation was conducted as described in [57] using mouse origin anti-FLAG and Protein G Dynabeads magnetic beads (Invitrogen).
Proliferation and wound healing assays were performed on the Incucyte ZOOM (Essen BioScience) system following the previously described protocols [58]. For soft agar assays, 5,000 trypsinized cells were seeded in triplicate in 0.4% low-melting-point agarose (Sigma) on top of a 1% agarose layer and colonies enumerated 21 days later.
Array-based transcriptome profiling of Cnot7-knockdown and CNOT7-overexpressing 4T1 tumor cells was performed on Affymetrix GeneChip Mouse Gene 1.0 ST arrays by the Microarray Core in the NCI Laboratory of Molecular Technology.
Library preparation was performed using the NEBNext Ultra Directional RNA Library Prep Kit for Illumina with NEBNext multiplexing oligos using manufacturer’s protocol. RNA sequencing was conducted on the Illumina HiSeq 2500.
RNA was isolated from tumors and cell lines using RNeasy kit (Qiagen) or TriPure (Roche) and reverse transcribed using iScript (Bio-Rad). Real-Time PCR was conducted using VeriQuest SYBR Green qPCR Master Mix (Affymetrix).
cDNA sequences of human FLAG-CNOT7, FLAG-CNOT7-D40A, and FLAG-CNOT7-E247A/Y260A were describe previously [36]. Lentiviral expression vectors were produced with Multisite Gateway recombination. An entry clone using the murine Pol2 promoter was recombined with the cDNA entry clone and N-terminal entry clone encoding the MYC (EQKLISEEDL) or FLAG (DYKDDDDK) epitope tag into a Gateway destination vector pDest-658. pDest-658 is a modified version of the pFUGW lentiviral vector which contains the enhanced polypurine tract (PPT) and woodchuck regulatory element (WRE) to provide higher titer virus. It also contains an antibiotic resistance gene for blasticidin resistance. Entry clones were subcloned by Gateway Multisite LR recombination using the manufacturer’s protocols (Invitrogen). Expression clones were transformed into E. coli STBL3 cells to minimize unwanted LTR repeat recombination, and verified by agarose gel electrophoresis and restriction digest. Transfection-ready DNA for the final clones was prepared using the GenElute XP Maxiprep kit (Sigma). A control vector (8166-M24-658) was generated by standard Gateway LR recombination of a stuffer fragment made up of a non-coding DNA into the pLenti6-V5-DEST vector (Invitrogen). CNOT7 and CNOT7-D40A lentivector constructs were generated by the Protein Expression Laboratory and the Viral Technology Group in NCI, Frederick, MD.
Site directed mutagenesis was employed to generate CNOT7-M141R mutant using the primers listed in S6 Table. The cDNA segment containing these mutations was subcloned into the wild type CNOT7 lentivector by restriction digest and ligation.
4T1 cells expressing 8166-M24-658 control vector or FLAG-CNOT7 constructs were grown to ~85% confluence in ten 15cm culture plates. Cells were then rinsed with 15mL ice cold sterile PBS, scraped off, pelleted (~1mL pellet), and snap frozen in liquid nitrogen. Cells were thawed on ice and lysed in 3.5mL lysis buffer (100mM NaCl, 5mM MgCl2, 10mM HEPES (pH 7.3), 0.5% NP40, 200 units RNasin (Promega), Protease inhibitor cocktail (Roche)). The resulting 4mL of cell lysate was spun down twice at 21,000*g for 20 minutes at 4°C. 40uL of lysate was saved 1% input to confirm immunoprecipitation. 200uL of lysate was saved for 5% RNA-seq input and was purified by TriPure (Roche) RNA extraction. 25uL beads per 1mL lysate of Protein G Dynabeads were blocked with 1mL 0.5% BSA in PBS at room temperature for 20 minutes then washed twice with 1mL NT2 wash buffer (50mM Tris-HCl (pH 7.4), 150mM NaCl, 1mM MgCl2, 0.05% NP40). 24ug antibody was added to each sample and incubated at 4°C overnight then beads were added to each sample and rotated for 30 minutes at room temperature. Beads were washed four times in 1mL NT2 buffer. Eighty five percent of beads were subjected to RNA extraction with TriPure. Protein from the remaining 15% of beads was eluted with Laemli buffer to confirm immunoprecipitation.
Protein was extracted with Pierce lysis buffer, vigorously homogenized, and incubated on ice for twenty minutes. 20ug lysate per sample in NuPage LDS Sample Buffer and NuPage Reducing Agent (Invitrogen) was used for western blotting. PVDF membrane (Millipore) containing transferred proteins was incubated overnight in solution of 5% milk protein, tris-buffered saline supplemented with 0.05% Tween-20, and primary antibody. The membrane was then incubated with horse-radish peroxidase linked anti-mouse (GE Healthcare), anti-rat, or anti-rabbit (Santa Cruz Biotechnology) IgG secondary antibodies. Immunoblot was visualized using Amersham ECL Prime Western Blotting Detection System and Amersham Hyperfilm ECL (GE Healthcare). Rabbit origin anti-CNOT7 was a generous provided by G. Sebastiaan Winkler. Commercial antibodies used in this study include rabbit origin anti-TOB1 (GeneTex), rabbit origin anti-CNOT1 (Protein Tech), rat origin anti-HA (Roche), mouse origin monoclonal anti-FLAG (M2, Sigma).
Female FVB/NJ or Balb/cJ mice from Jackson Laboratories were injected at 6–8 weeks of age. Two days prior to orthotopic injections, cells were placed in non-selective media. On the day of injection, 1x105 cells were injected orthotopically into the fourth mammary fat pad of age-matched virgin females. After 30 days the mice were euthanized by intraperitoneal injection of 1mL Tribromoethanol with subsequent cervical dislocation. Primary tumors were resected, weighed, and snap frozen in liquid nitrogen. Lungs were resected, surface metastases were counted; lungs were inflated with 10% nitrate-buffered formalin and sent for sectioning and staining. For tail vein injection, 105 were injected into the lateral tail vein, mice were euthanized 22 days post-injection. All procedures were performed under the Animal Safety Proposal (LCBG-004) and approved by the NCI-Bethesda Animal Care and Use Committee.
Statistical analysis comparing two samples were conducted using the Mann-Whitney test on Prism Version 5.03 (GraphPad Software, La Jolla, CA). Multiple-comparison data was analyzed by Kruskal-Wallis test with post-hoc Conover-Inman correction for multiple analyses by R-script. Survival data was conducted with the Mantel-Cox test on Prism.
RBP motif sites were mapped on the 3’UTR regions of genome genes (27305 mRNA FASTA format) by using a Perl script. RBP motif site enrichment analysis was performed by the random sampling the genes from genome in the same number (the final overlapping gene number) and calculate the numbers of the gene with the RBP motif site and of the RBP motif site and repeat the sampling 1000 times. The p-value was estimated by one tail t-test. Differential gene expression analyses were done by t-test using the software package R.
Array-based gene expression and RIP-seq studies from this study have been submitted to the NCBI Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/ under the accession numbers GSE73296 (array) and GSE73366 (RIP-seq).
|
10.1371/journal.pcbi.1006382 | Determinants of early afterdepolarization properties in ventricular myocyte models | Early afterdepolarizations (EADs) are spontaneous depolarizations during the repolarization phase of an action potential in cardiac myocytes. It is widely known that EADs are promoted by increasing inward currents and/or decreasing outward currents, a condition called reduced repolarization reserve. Recent studies based on bifurcation theories show that EADs are caused by a dual Hopf-homoclinic bifurcation, bringing in further mechanistic insights into the genesis and dynamics of EADs. In this study, we investigated the EAD properties, such as the EAD amplitude, the inter-EAD interval, and the latency of the first EAD, and their major determinants. We first made predictions based on the bifurcation theory and then validated them in physiologically more detailed action potential models. These properties were investigated by varying one parameter at a time or using parameter sets randomly drawn from assigned intervals. The theoretical and simulation results were compared with experimental data from the literature. Our major findings are that the EAD amplitude and takeoff potential exhibit a negative linear correlation; the inter-EAD interval is insensitive to the maximum ionic current conductance but mainly determined by the kinetics of ICa,L and the dual Hopf-homoclinic bifurcation; and both inter-EAD interval and latency vary largely from model to model. Most of the model results generally agree with experimental observations in isolated ventricular myocytes. However, a major discrepancy between modeling results and experimental observations is that the inter-EAD intervals observed in experiments are mainly between 200 and 500 ms, irrespective of species, while those of the mathematical models exhibit a much wider range with some models exhibiting inter-EAD intervals less than 100 ms. Our simulations show that the cause of this discrepancy is likely due to the difference in ICa,L recovery properties in different mathematical models, which needs to be addressed in future action potential model development.
| Early afterdepolarizations (EADs) are abnormal depolarizations during the plateau phase of action potential in cardiac myocytes, arising from a dual Hopf-homoclinic bifurcation. The same bifurcations are also responsible for certain types of bursting behaviors in other cell types, such as beta cells and neuronal cells. EADs are known to play important role in the genesis of lethal arrhythmias and have been widely studied in both experiments and computer models. However, a detailed comparison between the properties of EADs observed in experiments and those from mathematical models have not been carried out. In this study, we performed theoretical analyses and computer simulations of different ventricular action potential models as well as different species to investigate the properties of EADs and compared these properties to those observed in experiments. While the EAD properties in the action potential models capture many of the EAD properties seen in experiments, the inter-EAD intervals in the computer models differ a lot from model to model, and some of them show very large discrepancy with those observed in experiments. This discrepancy needs to be addressed in future cardiac action potential model development.
| Under diseased conditions or influence of drugs, cardiac myocytes can exhibit early afterdepolarizations (EADs) [1–3]. EADs are depolarization events during the repolarizing phase of an action potential (AP), which are known to be arrhythmogenic [4–7]. Many experimental and computational studies have been carried out, which have greatly improved our understanding of the causes and mechanisms of the genesis of EADs. It is well known that EADs can occur in an AP when inward currents are increased and/or outward currents are reduced, a condition called reduced repolarization reserve [8]. Under this condition, L-type calcium (Ca2+) current (ICa,L) can be reactivated to cause depolarizations in the repolarization phase of the AP to manifest as EADs. The importance of ICa,L reactivation for EAD genesis has been widely demonstrated in experiments [1,9] and computer simulations [10, 11]. Recent studies [12–15] using bifurcation theories have brought in additional mechanistic insights into the genesis of EADs, which show that EADs are oscillations originating via a supercritical or subcritical Hopf bifurcation and terminating via a homoclinic bifurcation, or via an unstable manifold of a saddle focus fixed point in the full AP dynamics [14]. In other words, irrespective to the specific ionic causes, EADs are oscillations caused by a Hopf bifurcation [16, 17] after which the quasi-equilibrium state becomes unstable and the system oscillates around the unstable equilibrium. As the slow outward currents grow gradually, the oscillation amplitude increases until a new bifurcation point—the homoclinic bifurcation at which the oscillation stops. A detailed discussion on the links between the ionic causes and nonlinear dynamics for the genesis of EADs was presented in our previous review article [18].
Despite the wide experimental and computational studies on the ionic causes and dynamical mechanisms of EAD genesis, less attention has been paid on EAD properties, such as the EAD latency (the time from the upstroke of the AP to the upstroke of the first EAD), the inter-EAD interval, and the EAD amplitude. Although it is well-known that reactivation of ICa,L is required for EAD genesis, since EADs are a collective behavior arising from the interactions of many ionic currents, it is unclear how these ionic currents affect the EAD properties and what are the major determinants. For example, since ICa,L plays a critical role in the genesis of EADs, one would intuitively expect that increasing the maximum conduction of ICa,L might increase the amplitude of EADs, but as we show in this study that this is not the case. On the other hand, understanding the EAD properties and their determinants is important for understanding the mechanisms of EAD-related arrhythmogenesis. For example, in an early experimental study [19], Damiano and Rosen showed that phase-2 EADs cannot propagate as premature ventricular complexes (PVCs) while phase-3 EADs can propagate as PVCs. This was also shown in our simulation studies [20–22]. Therefore, understanding what determine the EAD amplitude and takeoff potential may provide insights into EAD propagation to produce PVCs. If a PVC is a direct consequence of EAD propagation, then the EAD latency may provide information for the coupling interval between a sinus beat and the following PVC. EADs are also thought to be responsible for focal arrhythmias in the heart, and if this is true, then the oscillation frequency of EADs should be the same as the excitation frequency of ventricular arrhythmias. Furthermore, understanding the EAD properties and their determinants can also be important for the development of robust mathematical AP models. For example, we observed a discrepancy in inter-EAD interval between those from some widely used AP models and the experimental data. Experimental measurements in ventricular myocytes isolated from animal and human hearts almost exclusively show that the inter-EAD intervals are greater than 200 ms with few exceptions (Table 1). However, many ventricular myocyte AP models show inter-EAD intervals much shorter than 200 ms [11, 20, 23–27], raising a question on what ionic current properties have been missed in these models.
The previous computational studies mainly focused on the ionic causes and dynamical mechanisms of the genesis of EADs, and to our knowledge, no studies have been carried out to investigate the EAD properties and their determinants. Furthermore, a close comparison of the EAD properties from mathematical models with those from experimental recordings has not been done. The objective of this study is to use bifurcation theories and computer simulations to systematically investigate the EAD properties and their major determinants. We first made theoretical predictions of the EAD properties using the 1991 Luo and Rudy (LR1) model [28] based on our previous bifurcation theory of EADs [12]. We then carried out computer simulations using physiologically more detailed ventricular AP models [20, 22, 26, 29–31] to verify the theoretical predictions. In computer simulations, we also used parameter sets randomly drawn from assigned intervals so that large parameter ranges are explored to ensure generality of the simulation results. Theoretical and simulation results were compared with experimental results, and potential caveats of the current AP models were discussed.
Computer simulations were carried out in single ventricular myocytes. The governing equation of the transmembrane voltage (V) for the single cell is
CmdVdt=−Iion+Isti
(1)
where Iion is the total ionic current density and Isti the stimulus current density. Cm is the membrane capacitance which was set as Cm = 1 μF/cm2. We simulated six ventricular AP models: the 1991 Luo and Rudy (LR1) guinea pig model [28]; the 1994 Luo and Rudy (LRd) guinea pig model in a modified version [29]; the UCLA (HUCLA) rabbit model with modifications by Huang et al [22]; the 2004 ten Tusscher et al (TP04) human model [30]; the Grandi et al (GB) human model [26]; and the O’Hara et al (ORd) human model [31].
The differential equations were numerically solved using a first-order Euler method and the Rush and Larson method [32] for the gating variables with a fixed time step Δt = 0.01 ms.
For each model, a set of control parameters was used. The control parameter set is not the parameter set of the original model but a set we used for the AP to exhibit EADs. The major changes of parameters from the original models are either by increasing the maximum conductance of both ICa,L (was called slow inward current in the LR1 model, denoted as Isi) and the slow component of the delayed rectifier potassium current (IKs) or by increasing the maximum conductance of ICa,L but decreasing IKs or IKr (the rapid component of the delayed rectifier potassium current). The former corresponds to a normal myocyte (the original model) under isoproterenol while the later corresponds to the condition of long QT syndrome with isoproterenol. The specific changes of each model are detailed in S1 Text and the control APs exhibiting EADs are shown in Fig A in S1 Text.
To explore the effects of ionic current conductance on EAD properties in a wide parameter range, we varied the parameters in two ways: 1) We varied one parameter incrementally at a time but kept other parameters in their control values. The fold change of a specific parameter p is then defined as
α=p/pc
(2)
where pc is the control value of p; and 2) We randomly selected parameter sets, with each parameter drawn randomly from a uniform distribution in the interval (0.4pc, 1.6pc). The rationale for choosing such a parameter interval is that this interval can cover the range from no EAD to many EADs in an AP for most of the parameters in the models simulated in this study (see the figures in S1 Text).
We investigated three EAD properties in the AP models—amplitude, inter-EAD interval, and latency. As illustrated in Fig 1A, the EAD amplitude (AEAD) is defined as the difference between the takeoff voltage (Vtakeoff) and the peak voltage (Vpeak) of an EAD. The inter-EAD interval (TEAD) is defined as the time interval between the peaks of two consecutive EADs. The latency (LEAD) is defined as the time interval between the AP upstroke and the time when the 1st EAD takes off.
Table 1 summarizes the EAD properties observed experimentally in isolated ventricular myocytes from literature survey [33–53], which includes Vtakeoff, AEAD, TEAD, and LEAD. Based on this literature survey, we found that the Vtakeoff is always above -50 mV except some of the mouse [53] and guinea pig [48, 49] experiments. However, the EADs in the guinea pig experiments [48, 49] were induced by injection of a constant inward current, which may behave differently from the ones occurring intrinsically. Therefore, the observed EADs in isolated ventricular myocytes are mainly phase-2 EADs. The maximum amplitude of the phase-2 EADs can be as large as 70 mV. It has also been observed in experiments that Vpeak (thus AEAD) exhibits a negative linear correlation with Vtakeoff (Fig 1B). As shown in Table 1, TEAD ranges from 200 ms to 500 ms except some of the mouse experiments [53]. LEAD varies in a wide range, from 30 ms to 3 seconds.
To further assess the theoretical predictions and simulation results from the LR1 model, we carried out simulations using physiologically more detailed AP models. Fig 5 shows AEAD versus Vtakeoff for the five physiologically detailed models we simulated. The negative linear correlation holds roughly for all AP models while the slopes vary from -2 to -5 (corresponding to slopes ranging from -1 to -4 in plots of Vpeak versus Vtakeoff). Shifting the steady-state activation and inactivation curves of ICa,L results in roughly the same shift in the AEAD and Vtakeoff relationship, indicating that the ICa,L reactivation is responsible for EADs in all the models. The maximum EAD amplitude and the lowest detectable takeoff potential also vary largely from model to model. The maximum EAD amplitude (without the shifts in ICa,L kinetics) of the TP04 model is the smallest (<35 mV) while that of the ORd model is the largest (~ 90 mV) among the five models. The lowest Vtakeoff of the TP04 model is the highest (~ -20 mV) while that of the HUCLA model is the lowest (~-45 mV) among the five models. Although the EAD amplitude properties vary largely from model to model, they generally agree with the experimental data shown in Fig 1B and Table 1.
We investigated the effects of the maximum conductance of the major ionic currents on AEAD for all five models, including ICa,L (Fig B in S1 Text), IKs (Fig C in S1 Text), IKr (Fig D in S1 Text), IK1 (Fig E in S1 Text), INCX (Fig F in S1 Text), INaK (Fig G in S1 Text), INaL (Fig H in S1 Text), and Ito (Fig I in S1 Text). The effects of the inactivation time constant of ICa,L were also shown (Fig J in S1 Text). The general observations are the same as those from the LR1 model, i.e., increasing an inward current increases the number of EADs in the AP, decreases AEAD until a new EAD suddenly appears in the AP at which AEAD becomes maximum. Increasing an outward current decreases the number of EADs in the AP, and increases AEAD and AEAD of the last EAD in the AP reaches maximum before it suddenly disappears from the AP. However, there are some exceptions. For example, increasing the maximum conductance of Ito can either promote or suppress EADs (Fig I in S1 Text). In both the LRd model and the ORd model, increasing Ito promotes EADs (more number of EADs in the AP), but suppresses EADs in the TP04 model and GB model. For the HUCLA model, increasing Ito,s promotes EADs, while increasing Ito,f first promotes EADs but then suppresses EADs. Since Ito is an outward current, it is generally known that it suppresses EADs [54]. However, recent studies have demonstrated that Ito can also promote EADs [45, 55]. Whether Ito promotes or suppresses EADs depends on its magnitude, speed of inactivation, and its pedestal component. Slow inactivation or large pedestal current tends to suppress EADs [55]. Another exception is INCX. Increasing INCX promotes EADs in all the models except in the TP04 model in which EADs may also be suppressed by increasing INCX (see Fig F in S1 Text).
As shown in Table 1, TEAD in isolated ventricular myocytes is typically from 200 ms to 500 ms. In this section, we investigate TEAD and its determinants in the AP models. We first used the LR1 model for theoretical treatments and then used the physiologically more detailed models to verify the theory.
The simulation results from the physiologically more detailed models show similar characteristics of TEAD dependence on different parameters (Fig 7). Increasing an inward current decreases TEAD while increasing an outward current (except Ito) increases TEAD. In the LRd and HUCLA models, increasing Ito conductance decreases TEAD. This is because Ito promotes EADs in these models. Similar to the LR1 model, the change in TEAD is not very sensitive to a change in the maximum conductance of an ionic current until it is close to the transition from two EADs to one EAD (or from one EAD to two EADs) in the AP at which TEAD changes rapidly. The predominant TEAD for the 5 models are roughly: LRd—90 ms; HUCLA—85 ms; TP04–270 ms; ORd—275 ms; and GB—125 ms. As shown in Fig 7, TEAD in the LRd, HUCLA, and GB models are shorter than 200 ms (between 50 ms to 200 ms), while those in the TP04 and ORd models range from 250 ms to 500 ms. In all the models, τf exhibits a stronger effect on TEAD than the other parameters we explored.
In Fig 7, we only show the first inter-EAD interval versus a specific parameter. To explore wider ranges of TEAD in these models, we used random parameter sets (see Methods) and measured all TEAD in an AP. Fig 8 shows the TEAD distributions for the AP models. The TEAD ranges are: LRd—from 75 ms to 125 ms (peak ~90 ms); HUCLA—from 50 ms to 150 ms (peak ~85 ms); TP04—from 225 ms to 350 ms (peak ~250 ms); ORd—from 200ms to 500 ms (peak ~250 ms); and GB—from 100 ms to 200 ms (peak ~125 ms). Therefore, the TEAD of the TP04 and ORd model is in the experimentally observed range, while those of the GB, LRd, and HUCLA models are too short comparing to the experimental recordings. Note that the TEAD range using the random parameter sets is similar to the range seen in Fig 7 for each model, indicating that the TEAD range of an AP model is not sensitive to ionic current conductance but mainly determined by the period variation in the dual Hopf-homoclinic bifurcation.
Based on the theoretical predictions that τf is the main determinant of the inter-EAD interval, we changed τf functions in the LRd and HUCLA models from upward bell-shaped functions to downward bell-shaped functions (see Fig 8C) to lengthen τf in the plateau phase, we can effectively shift the TEAD distributions toward the longer periods (red histograms in Fig 8A and 8B). The inactivation time constants of ICa,L in the plateau voltage for the TP04 model and the ORd model are much longer, and thus inter-EAD intervals of these two models are also much longer. The GB model has a much shorter inactivation time constant of ICa,L in the plateau, similar to those of LRd and HUCLA, and thus the inter-EAD interval is also short.
As shown in Table 1, LEAD varied in a large range, from less than 100 ms to a couple of seconds. Based on the bifurcation theory of EADs [12, 18], for EADs to occur, besides the instability leading to oscillations, the voltage needs to decay into the window voltage range of ICa,L activation and the LCCs need to be recovered by a certain amount so that there are enough LCCs available for re-opening.
To reveal how LEAD is determined by the ion channel properties, we started with simulations of the LR1 model (Fig 9A). We varied four parameters: Gsi, GK, GK1, and τf. Increasing Gsi first decreases LEAD quickly and then increases LEAD slowly (red curve in Fig 9A). The longest LEAD occurs when the first EAD appears in the AP (green arrow in Fig 9A). When a new EAD first appears in an AP, its takeoff potential is the lowest which is close to the potential of the homoclinic bifurcation point (see Fig 2A), and it takes a longer time for the EAD depolarization to occur (the same reason that the TEAD is the longest and AEAD is the largest when a new EAD appears). As Gsi increases, the takeoff potential is higher and thus LEAD shortens. However, increasing Gsi also slows the decay of voltage into the window range of LCC reactivation, and thus lengthens LEAD. Increasing GK does the opposite (blue curve in Fig 9A) for the same reasons mentioned for Gsi. Changing GK1 has no effect since IK1 is almost negligible in early phase-2 of the AP. τf has a big effect on LEAD, which can vary LEAD in a much wider range than the conductance. τf affects LEAD by two ways: 1) slowing τf causes a slower inactivation of Isi which delays the voltage decay to the window range; 2) slowing τf delays recovery of LCCs, which then delays the depolarization of the first EAD.
Similar behaviors occur in all other models (Fig 9B–9F): changing a conductance usually has a small effect on LEAD until it causes the only EAD in the AP to disappear, at which LEAD changes steeply; and the inactivation time constant of ICa,L is the most sensitive parameter for LEAD. The LEAD varies largely from model to model. In Fig 10, we show LEAD distributions from random parameter sets for all models, and the LEAD ranges are: LR1—from 600 ms to 1000 ms (peak ~750 ms); LRd—from 240 ms to 550 ms (peak ~300 ms); HUCLA—from 135 ms to 180 ms (peak ~150 ms); TP04—from 640 ms to 720 ms (peak~650 ms); ORd—from 360 ms to 600 ms (peak ~400 ms); GB—from 240 ms to 500 ms (peak~280 ms).
In this study, we used theoretical analyses and computer simulations to investigate EAD properties and their major determinants in AP models of ventricular myocytes. Our major observations and conclusions are summarized and discussed below.
In all the models we simulated, the EAD takeoff potentials are usually above -40 mV (see the 0 mV shift cases in Fig 5), which is in agreement with experimental data from isolated ventricular myocytes (Table 1). A negative linear correlation between EAD amplitude and takeoff potential, which has been shown in experimental recordings [1, 2, 9], can be implied from the dual Hopf-homoclinic bifurcation and is shown in simulations of all models. The slopes of the negative linear correlations, the lowest takeoff potentials, and the maximum EAD amplitudes vary substantially from model to model. The ORd model exhibits the largest maximum EAD amplitude and the steepest slope of the linear correlation between EAD amplitude and takeoff potential.
Although EADs are promoted by increasing inward currents or decreasing outward currents, once EADs occur in an AP, increasing the maximum conductance of an inward current or decreasing that of an outward current does not increase the amplitudes of the EADs. Increasing the maximum conductance of an inward current causes more EADs in the AP, and the maximum EAD amplitude occurs when a new EAD appears in the AP. The amplitude of this new EAD then decreases as the conductance increases. Increasing the maximum conductance of an outward current decreases the number of EADs in the AP but increases the EAD amplitude. The maximum EAD amplitude occurs before an EAD disappears from the AP. Based on the bifurcation analysis [12–15], the EAD amplitude grows as the system evolves from the Hopf bifurcation point to the homoclinic bifurcation point (Fig 2A and Fig 4). This behavior should hold for both supercritical [12, 15] and subcritical [14, 15] Hopf bifurcation. At the homoclinic bifurcation point, the takeoff potential is the lowest and the EAD amplitude becomes the maximum. Therefore, the amplitude of the last EAD in an AP depends on how far away the takeoff potential is from the homoclinic bifurcation point. Born of a new EAD or death of an existing EAD always occurs when the EAD takes off near the homoclinic bifurcation point.
Note that in our simulations of assessing EAD amplitude and takeoff potential, we used random parameter sets to explore a wide range of parameters for each model. In these simulations, only phase-2 EADs in the ventricular myocyte models were observed. A recent simulation study [15] using the ORd model and the Kurata et al model [56] also showed that only phase-2 EADs could be observed. This indicates that varying maximum ionic current conductance may not produce phase-3 EADs. This agrees with the experimental data that EADs observed in isolated ventricular myocytes are mainly phase-2 EADs (Table 1), while phase-3 EADs are rarely observed in isolated cells, except under Ca2+ overload [57] or by external current injection [49]. On the other hand, phase-3 EADs are observed in Purkinje fibers [19] and cardiac tissue [58–60]. Previous computer simulations showed that phase-3 EADs could occur in single cells with a strong InsCa under elevated intracellular Ca2+ concentration [20, 21, 61] or in tissue with repolarization heterogeneity induced dynamical instabilities [22, 58]. Therefore, phase-3 EADs can be caused either by strong Ca2+ overload in isolated myocytes or by repolarization heterogeneities in tissue with reduced repolarization reserve, while phase-2 EADs are mainly due to reduced repolarization reserve and reactivation of ICa,L [9, 10, 18, 62, 63].
Since phase-2 EADs cannot propagate into PVCs in tissue [19–22], this raises question on how are EADs linked to arrhythmias under LQTS and many other diseased conditions where Ca2+ may not be overloaded. In recent studies [22, 64], we demonstrated how phase-2 EADs and tissue-scale dynamical instabilities interact to result in PVCs and arrhythmias under LQTS, linking mechanistically phase-2 EADs to arrhythmogenesis.
Based on the bifurcation analysis, the inter-EAD interval is governed by the period of the limit cycle oscillation between the Hopf bifurcation and the homoclinic bifurcation. Similar to AEAD, the inter-EAD interval increases as the system evolves from the Hopf bifurcation point to the homoclinic bifurcation point. This behavior has been demonstrated in experimental recordings previously [65]. Our theoretical analysis and simulation of the AP models showed that the inter-EAD intervals (except the last one in an AP), which are mainly determined by the period of the limit cycle oscillation at the Hopf bifurcation, is insensitive to the change of maximum ionic current conductance but more sensitive to the recovery of LCCs (Figs 7 and 8, and Eq 7). The TEAD range of an AP model is determined by the oscillation period between the Hopf bifurcation and the homoclinic bifurcation. However, the inter-EAD interval from different models exhibits different ranges, differing several folds. On the other hand, the inter-EAD intervals observed in isolated ventricular myocyte experiments mostly are in the range from 200 ms to 500 ms, irrespective of species (Table 1). In the AP models simulated in this study, the inter-EAD intervals of LR1, TP04, and ORd are in the same range as observed experimentally, but other models exhibit much faster inter-EAD intervals, indicating that caveats may exist in these models. Our simulation indicates that the major caveat may lie in the formulation of ICa,L, in particular the recovery time of ICa,L during the plateau phase (see more detailed discussion below).
EAD latency is determined by many factors. In term of biophysics, since EADs are caused by reactivation of LCCs, the voltage needs to decay into the window range for reactivation of LCCs, which depends on the speed of activation of the outward currents (namely, IKs, IKr, and Ito) and inactivation of the inward currents (namely ICa,L). Then there are enough LCCs recovered for re-opening, which depends on how fast the LCCs recovers. In more general term of nonlinear dynamics as indicated by the bifurcation theory, the voltage and other variables need to enter the basin or the vicinity of the basin of attraction of the limit cycle. This requires not only the voltage but also the other state variables to reach their proper values. For example, in the LR1 model, the X-gating variable needs to grow to a certain value to engage the Hopf bifurcation as shown in Fig 2A. If X grows too slowly, the system may stay at the quasi-equilibrium state for a long time with no oscillations until reaching the Hopf bifurcation point, such as the cases shown in Song et al [51]. Note that transient oscillations around a stable focus can occur before the Hopf bifurcation, and thus EADs can occur before the Hopf bifurcation (see Fig 2A, Fig 4C and 4D in this study, and Figs 2 and 4 in the study by Kügler [14]). Therefore, the EAD latency can be very variable, explaining the experimental observation that EAD latency varies in a wide range, from less than 100 ms to several seconds (Table 1).
It is obvious that understanding the EAD properties and nonlinear dynamics is of great importance for understanding arrhythmogenesis in cardiac diseases [66], but it also provides important information for cardiac AP modeling. Previous AP modeling has mainly considered AP morphology, APD, as well as APD restitution, but not the EAD properties. For example, the inter-EAD intervals of the guinea pig ventricular myocyte models [11, 23, 24] are much shorter than what have been observed in isolated guinea pig ventricular myocytes (Table 1). This is also true for the rabbit ventricular myocyte models. As shown in Fig 8A and 8B, we can effectively increase the inter-EAD interval by increasing the inactivation time constant τf of ICa,L. However, for both the guinea pig model [67] and the rabbit model [68], the original inactivation time constants were based on experimental measurements. Since in the Hodgkin-Huxley formulation of ICa,L, the f-gate is a voltage-dependent inactivation gate, but it also governs the recovery of ICa,L. Therefore, one would conclude that experimentally-based τf might be a correct constant but the recovery properties of ICa,L in these models are incorrect, which gives rise to the discrepancy in inter-EAD intervals between mathematical models and experimental measurements. On the other hand, τf is large in the LR1, TP04, and ORd models, which gives rise to the right recovery times to result in inter-EAD intervals in the ranges as observed in experiments. That also does not mean that ICa,L models are completely correct in these AP models since we know that the one for the LR1 model gives rise to a too slow inactivation of ICa,L. Therefore, the EAD properties provide additional important information for AP modeling, which need to be considered in future AP model development.
A major limitation is reliable experimental data curation. First, most of the values in Table 1 were estimated from the published figures, which is difficult to be accurate and unbiased. Second, experimental data of EADs recorded from isolated ventricular is not abundant. Moreover, to calculate TEAD, we have to select APs with two or more EADs, which further limited our data sources. Third, most of the experimental plots do not have coordinates but indicated by scale bars. Sometimes, these scale bars may not be correctly labeled due to different reasons. For example, we confirmed with Dr. Li that the time scale bar in Fig 2C of Ref. [33] is 300 ms instead of 150 ms. In computer simulations, we only explored the maximum conductance of the ion currents and the voltage-dependent inactivation time constant of ICa,L, but it is obvious any parameter that affects repolarization will have an effect on EAD genesis and EAD properties. For example, the time constants of ion channel activation and inactivation, the Ca2+-dependent inaction of ICa,L [69], the intracellular Ca2+ transient, mitochondrial metabolism and Ca2+ cycling [61, 70], as well as spatial distribution of the ion channels will impact the EAD behaviors, which need to be investigated in future studies. Another limitation of the simulations is that our conclusions may depend on the setting of control parameter and the assigned intervals for random parameters. However, as we show in this study, despite certain distinct difference between models, such as the inter-EAD interval, the general conclusions are not model dependent, and thus, not likely to be affected by the choice of control parameter sets and the assigned intervals. We only investigated the effects of voltage-dependent inactivation of ICa,L on EAD properties. It is shown that Ca2+-dependent inactivation also play important roles in EAD genesis [69], one would anticipate that the Ca2+-dependent inactivation might have a large effect on EAD properties. However, the Ca2+-dependent inactivation is modeled very differently in different models, we do not have a unique way to alter a parameter to study the effects in the models. For example, in most models, Ca2+-dependent inactivation was modeled by an instantaneous function of Ca2+, and thus it is not clear how to change the time constant of Ca2+-dependent inactivation in these models. One major caveat of the current study is that in the models we simulated, the EADs are caused by reactivation of ICa,L, however, experimental studies showed EADs can also be caused by spontaneous Ca2+ release [57, 71, 72]. In a recent modeling study by Wilson et al [73], the authors showed that spontaneous Ca2+ oscillation can lead to EADs. We performed the same analyses of this model as we did for other models and showed the results in Fig K in S1 Text. The EAD behaviors and their dependence on the ionic conductances are similar to other models, but the model indeed exhibits some differences from the other models. For example, TEAD exhibits a much wider range (from 100 ms to 1200 ms) than those of other models and the TEAD histogram shows characteristic distributions indicating that there are two mechanisms of EADs involved. However, further investigation is needed to pinpoint the underlying mechanisms of EADs in this model and compare model results with experimental data in future studies.
|
10.1371/journal.pbio.1001931 | Adaptive Evolution and Environmental Durability Jointly Structure Phylodynamic Patterns in Avian Influenza Viruses | Avian influenza viruses (AIVs) have been pivotal to the origination of human pandemic strains. Despite their scientific and public health significance, however, there remains much to be understood about the ecology and evolution of AIVs in wild birds, where major pools of genetic diversity are generated and maintained. Here, we present comparative phylodynamic analyses of human and AIVs in North America, demonstrating (i) significantly higher standing genetic diversity and (ii) phylogenetic trees with a weaker signature of immune escape in AIVs than in human viruses. To explain these differences, we performed statistical analyses to quantify the relative contribution of several potential explanations. We found that HA genetic diversity in avian viruses is determined by a combination of factors, predominantly subtype-specific differences in host immune selective pressure and the ecology of transmission (in particular, the durability of subtypes in aquatic environments). Extending this analysis using a computational model demonstrated that virus durability may lead to long-term, indirect chains of transmission that, when coupled with a short host lifespan, can generate and maintain the observed high levels of genetic diversity. Further evidence in support of this novel finding was found by demonstrating an association between subtype-specific environmental durability and predicted phylogenetic signatures: genetic diversity, variation in phylogenetic tree branch lengths, and tree height. The conclusion that environmental transmission plays an important role in the evolutionary biology of avian influenza viruses—a manifestation of the “storage effect”—highlights the potentially unpredictable impact of wildlife reservoirs for future human pandemics and the need for improved understanding of the natural ecology of these viruses.
| Human populations have experienced several pandemics involving new subtypes of influenza virus over the past century. All of these pandemic strains contained gene segments that originated in wild birds, a host pool that supports a very large and genetically diverse array of influenza viruses. However, once an avian strain establishes itself within the human population, the genetic diversity of the resulting human subtypes is typically quite low compared to that of their avian counterparts. Here we compare the evolutionary dynamics of human and avian influenza viruses in North America and test different hypotheses that might explain these two contrasting evolutionary patterns. Our analysis shows that a combination of characteristics of the host (especially demography) and virus (such as durability in water and mutability) explains the diversity observed. Using a theoretical model, we show that the combination of the short lifespan of wild birds, and greater durability of viruses in aquatic environments, is key to maintaining the high levels of influenza diversity observed in wild birds.
| Seasonal epidemics of influenza viruses are responsible for significant human morbidity and mortality [1]. Owing to their RNA makeup, evolution of influenza A viruses occurs rapidly [2],[3] and is an important driver of their epidemiology [4],[5]. Over the past decade, there has been an extensive effort to understand the concurrent epidemiology and evolutionary trajectory of human influenza viruses [4]–[7], an approach termed “phylodynamics” [8]. Surprisingly, parallel analyses in wild birds, the natural reservoir of influenza viruses [9],[10], are lacking. Such analysis is particularly timely because of the recent recognition of H5N1 and H7N9 avian influenza viruses (AIVs) as pandemic threats [11]–[14].
The epidemiological and evolutionary histories of human and AIVs in North America from 1976–2001 are summarized in Figure 1. In humans, seasonal influenza outbreaks exhibit substantial annual variation (Figure 1A), which is also reflected in shifting dominance of co-circulating subtypes (Figure 1C). Human influenza viruses exhibit very limited subtype diversity (Figure 1C), as defined by the number of serologically distinct hemagglutinin (H or HA) glycoprotein types [9], where only H1 and H3 subtypes of influenza A viruses have significantly circulated since 1968 [15]. In addition to this paucity of subtypes, genetic diversity is also limited within H1 (Figure 1E) and H3 (Figure 1F) subtypes, as reflected in the slender trunk of the consensus phylogenetic tree (Figure 1I and 1J).
These patterns in human influenza are consistent with “immune escape,” a phenomenon that has been suggested to be common in directly transmitted, immunizing pathogens with a short infectious period, in which antigenic evolution results in partial cross-immunity between strains [8]. In contrast, influenza A viruses in avian populations exhibit a rich array of subtypes, with fully 13 of the known 18 HA subtypes isolated from North American birds over this time span (Figure 1D). This pattern of higher subtype diversity through time is further enriched by higher genetic diversity within subtypes, for instance in H1 (Figure 1G) and H3 (Figure 1H). Indeed, AIVs typically exhibit a scaled effective population size (, which measures the phylogenetic diversity of the virus population [16]) that is an order of magnitude greater than for their human counterparts: We estimated to be 7.0 y and 1.5 y, respectively, for H1 and H3 in humans and 38.7 y and 77.5 y in birds (Figure 1E and 1F; as we show below, other avian subtypes also exhibit higher diversity than commonly observed in H1 and H3 human subtypes). Thus, although phylogenetic trees of H1 and H3 AIVs show some evidence of selection (immune escape) [17], they also document broad viral coexistence (Figure 1K and 1L).
The mechanistic origins of these differences remain unclear. Here, we propose the following nonmutually exclusive hypotheses as possible explanations: (1) The immunological hypothesis holds that more rapid loss of immunity and/or weaker heterologous cross-protection in birds than humans reduces competition among strains, leading to higher diversity; (2) the ecological hypothesis suggests that associations between virus lineages and avian host species diversity allow contemporaneous evolution within multiple bird species, sustaining an enriched gene pool; (3) the geographic hypothesis supposes that greater geographic isolation in birds than in humans leads to allopatric evolution; (4) the genetic hypothesis posits that mutation rate differences between avian and human viruses explains the disparity in viral diversity; (5) the demographic hypothesis focuses on the higher fecundity and shorter lifespan of birds compared to humans, which may mitigate the selective pressure of herd immunity via substantial recruitment of immunologically naïve individuals propagating the pathogen; and finally, (6) the epidemiological hypothesis predicts that there exists a long-lived environmental reservoir for avian strains, but not for human strains, facilitating coexistence of a broad spectrum of genetically, immunologically, and ecologically similar viruses.
To date, there have been no attempts to synthesize the available evidence for or against these different explanations. In this study, we address these hypotheses through a combination of statistical analysis of empirical covariates and epidemiological modeling to identify the most parsimonious explanation for the observed differences of HA genetic diversity between human and AIVs.
We analyzed available AIV sequence data from GenBank for 11 HA subtypes from 1976–2013 (see Materials and Methods as well as Text S1, section S2) and developed a set of covariates reflecting specific predictions of the competing hypotheses. We sought to place these covariates on an equal footing and simultaneously to assess the contribution of each to HA diversity using a statistical model. Ordinary linear regression with a large number of covariates (of the same order as the number of observations) results in variance inflation, low statistical power, and issues of statistical identifiability [18]. We therefore adopted a regularized regression method, known as elastic-net regression [19], that solves this problem using a shrinkage estimator to trade off a small amount of bias for substantial reductions in the variance of estimated parameters. As a side effect of this estimation scheme, the resulting coefficients may be interpreted as evidence for or against the inclusion of a covariate (if the coefficient shrinks either to nonzero or zero, respectively). Furthermore, by normalizing covariates we can straightforwardly quantify and compare the relative size of each effect.
This regression analysis yielded the following conclusions (Text S1, sections S4.1–S4.3). As shown in Table 1, among AIV subtypes, each hypothesis we considered made a contribution towards HA genetic diversity, though the magnitude of effects varied considerably (Table S2 for covariates values). Indeed, the strongest covariates were immune selective pressure (Hypothesis 1, quantified via the amino acid substitution rate) and the environmental durability of virions (Hypothesis 6, inferred by experimental incubation data on viral persistence), whose respective impacts were at least twice as big as the effects of nucleotide mutation rate (Hypothesis 4, quantified by multiplication of nucleotide substitution rate by the substitution rate at third position sites), geographic structure (Hypothesis 3, inferred by FST that measures population differentiation through space), and host diversity (Hypothesis 2, characterized by Shannon Index of host species sampled).
Some of these results are not surprising. For instance, our finding concerning the modest contribution of host species diversity to HA genetic variation had previously been observed by Chen and Holmes [20] and probably arises from frequent interspecies transmission. Similary, the impact of geographic structuring on gene flow among North American AIVs has been elegantly demonstrated elsewhere [20],[21]. The important novel result to emerge from our statistical analyses is the substantial contribution of virus durability to HA genetic diversity. This is a component of virus biology not previously considered in the study of AIV evolution. Thus, to better dissect how the durability of AIVs in the environment affects transmission dynamics and subsequently HA diversity and simultaneously to explore Hypothesis 5 (host demography), we constructed a mechanistic phylodynamic model.
Our model is stochastic, seasonally forced, and agent-based [22] and incorporates a one-dimensional antigenic space, where nonneutral mutations change antigenic phenotype from neighbor to neighbor, thus decreasing cross-immunity (Text S1, section S3) [23]. Crucially, our model allows the tracking of virus antigenic diversity and hence reconstruction of within-subtype digital phylogenies from model output (algorithm detailed in Text S1, section S3.3, Figures S7 and S8), as summarized in Figure 2. Virus diversity is quantified in our model by the number of different antigenic strains at a given time and provides an analog to the diversity inferred by the scaled effective population size on genetic data. This model also enables assessing the role played by host demography (Hypothesis 5) on the maintenance of virus diversity. If these factors are to explain the observed differences between human and avian strains described above, then we expect to observe rapid population turnover and an absence of genetic diversity in a host population parameterized for humans [4],[5],[24], whereas a model parameterized for birds should show broad coexistence of viral strains (parameters are detailed in Table S3). Because the modeling framework we adopt may give rise to either restricted or expansive antigenic diversity depending on epidemiology [23], the inferences one draws are not a result of prejudicial selection of model parameters or functional forms.
An innovative aspect to this model is our formulation of transmission. AIV transmission has been thought to be predominantly fecal-oral, which has been considered as essentially direct because of (i) the proximity between susceptible and infected birds needed for infection and (ii) the scaling of transmission with the duration of infectivity. Furthermore, recent research points to direct bird-to-bird transmission via the respiratory route [25]. Evidence is accumulating, however, to suggest that an additional transmission route is possible via long-lived viruses in environmental reservoirs [26]–[30], effectively giving rise to a second (longer) time scale over which transmission can occur. This hypothesis is based in part on the routine isolation of AIVs from mud samples, soil swabs [26], unconcentrated lake water [31], feathers [32], and the observation of prolonged virus durability in water [9],[33]–[36] and other media [37]. Virus durability is commonly quantified by Rt, which is the time required to reduce infectivity by 90%, and may vary from a couple of days to several months [33]. Rt is determined both by physical environmental conditions, notably temperature, pH, and salinity [33],[34], as well as by subtype identity [38]. Consequently, Hypothesis 6 suggests that environmental transmission could act on a distinctly longer time scale than direct fecal-oral transmission, thereby significantly impacting virus diversity and phylogenetic structure through frequent re-seeding of the avian virus gene pool, as illustrated in Figure S1.
To quantify the influence of Hypothesis 5 (host demography) on influenza virus diversity, we first parameterized our model to mimic the within-subtype dynamics of human influenza, assuming only direct transmission. Seeding simulations with only a few antigenic variants, we observed the continual replacement of a dominant strain by new antigenic variants (Figure 3A), driven by selective pressure to escape herd immunity in the host population, as empirically observed [4],[5],[24]. The direct measure of antigenic diversity generated by our model (Figure 3D; six antigenic strains coexist on average) is consistent with our estimates of scaled effective population size of human influenza (Figure 1E or 1F). The resulting inferred phylogenetic tree from our model output (Figure 3G) is also “ladder-like,” characteristic of the strong immune escape signature observed in data (Figure 1I or 1J).
We then addressed demographic explanations by exploring the impact of host biology alone, reparameterizing the model to take into account the reduced lifespan, increased fecundity, and seasonal breeding of birds compared with humans. Model output remained qualitatively unaffected, demonstrating continuous antigenic evolution (Figure 3B), with low-standing antigenic diversity (Figure 3E; five strains coexist on average) and a slender trunk in the phylogenetic tree (Figure 3H). Thus, in this model, host demographic properties alone do not strongly influence levels of genetic variation. In contrast, we found that the inclusion of environmental transmission dramatically increased standing antigenic diversity of AIVs (Figure 3C and 3F; 60 strains coexist on average), resulting in both immune selection and virus diversification (Figure 3I). The phylogenetic tree contains lineages that would have gone extinct in the absence of environmental transmission, demonstrating the punctuation of antigenic evolution with reintroduction of past dominant variants, to which there is little immunity in the population (Figure S1).
Next, we carried out sensitivity analyses spanning the parameter space of the most common influenza systems, including swine and equine influenza (Figure 4). We found that the long natural lifespan (and low fecundity) of free living mammalian hosts sustain the selective pressure exerted by herd immunity in the population on dominant strains. Indeed, in the presence of long lifespan and associated long-lived immunity, even substantial levels of environmental transmission do not dramatically increase antigenic diversity. Reduced host lifespan, however, leads to a faster turnover of the population, reducing the selective impacts of herd immunity. As shown in Figure 4, it is the combination of high lifetime fecundity and environmental transmission that produces dramatic increases in genetic diversity and the coexistence of distantly related viral lineages. In Text S1 (section S5, Figures S11, S12, S13, S14, S15, Table S4), we present results of sensitivity analyses to demonstrate the robustness of this broad conclusion to changes in assumed duration of immunity, the strength of cross-protection, the infectious period, direct transmission rate, and the mutation rate. This result also shows very little variation in stochastic realizations of the model.
Four testable predictions arise from our model (Figure 4), three of which can be explored using our existing data set. First, viruses with greater environmental durability are predicted to exhibit greater genetic diversity when host lifespan is short (Figure 5A). Second, increasing viral durability in the environment is predicted to facilitate the reintroduction of past virus variants and hence to correlate with the estimated time to the most recent common ancestor (TMRCA). Third, the variable nature of indirect transmission chains via the environmental reservoir [39],[40] is predicted to generate greater variability in branch lengths. Fourth, and finally, environmental transmission should increase the frequency of co-infection events between antigenically distant viruses (Figure 3J–L) and therefore the number of co-circulating subtype combinations.
We tested for the presence of the first association using phylogenies inferred for 10 different avian influenza HA subtypes isolated from North American wild birds (Text S1, section S2). For comparison, we performed a parallel analysis of the most prevalent subtype observed in human (H3), equine (H3), and swine (H1) influenza viruses. Recalling that HA durability correlates with HA genetic diversity (Table 1 and Figure S10), we further observed that scaled effective population size () increases with HA durability either estimated at a fixed temperature (Figure 5A) or averaged over a season (Figure S16 and Table S15). Here, Ne represents the size of an idealized population corresponding to observed levels of genetic diversity, and τ represents the generation time of the virus. Because long-term transmission chains resulting from environmental durability should directly increase AIV generation time (the average time between infections), a correlation between and experimentally measured environmental durability is evidence for the environmental transmission hypothesis. An increase in is expected to impact the time it takes for the sampled viruses to coalesce to a common ancestor. Consistent with the second model prediction, we observed that TMRCA, quantified by tree height here, increased with HA durability (Figure 5B). Third, we found that variance in branch lengths correlated with HA durability across subtypes (Figure 5C), as predicted.
One final model prediction, that co-infection frequency should increase with environmental durability, is of great evolutionary relevance, as co-infection is necessary for reassortment, which may be a prerequisite to the evolution of pandemic strains [17],[41]. Although co-infections are expected to occur infrequently and mainly between related strains in humans (Figure 3J), environmental transmission in avian communities allows for substantially more frequent co-infections, especially between antigenically distant variants (Figure 3L). This is due to effects of environmental transmission increasing the propensity for co-infection (an empirically supported phenomenon [42]), together with the high antigenic diversity generated by environmental transmission (Figure 4). To test this prediction, a regression of subtype-specific HA diversity against the entropy of associated NA subtypes failed to detect any significant relationship (r = 0.15, p = 0.65).
Our study identified the mechanisms that act to determine hemagglutinin genetic diversity in avian influenza viruses. In particular, the analyses reveal that our hypotheses act in concert to shape the phylodynamics of AIVs. These results are consistent with prior studies that have examined each mechanism in isolation. For instance, it has been shown that strong spatial structuring is an important factor in the phylogeography of these viruses [20],[21]. Similarly, the modest association between genetic diversity and host species number is known [20]. Although it is widely hypothesized that increasing antigenic evolution decreases genetic diversity across human subtypes [43] and between human and swine H3 [44], our research has also demonstrated that increasing avian immune selective pressure acts to reduce influenza virus diversity.
Our most surprising empirical finding is that HA genetic diversity increases with virus durability, as measured in experimental assays, across AIV subtypes (Table 1 and Figure 5). The corresponding theoretical result is that, in short-lived hosts, increasing the frequency of environmental transmission results in greater equilibrium levels of viral genetic diversity (Figures 3 and 4). Thus, both empirical and theoretical results suggest that environmental transmission acts in wild bird populations to increase avian influenza genetic diversity. As emphasized throughout, our results do not exclude a role for additional mechanisms (e.g., Hypotheses 1–3), but establish statistically that the size of the effects of subtype-specific amino acid substitution rates and environmental durability are largest.
Two limitations of this study warrant further investigation. First, a detailed understanding of cross-immunity remains an important empirical limitation to any study of avian influenza evolution. Establishing the duration and extent of protective immunity against heterotypic viruses, particularly, remains a priority. At present, comparative data to test this idea directly are lacking, although work showing the absence of immunity conferred by DNA subtype-specific vaccines to challenge strains from other subtypes [45] suggests that the effect of cross-immunity will be limited. Second, our model assumed a one-dimensional strain space for the practical purpose of numerical tractability [23]. Future work should investigate the effects of incorporating other structures of cross-immunity [4],[5] into epidemiological models of across-species influenza. Similarly, our findings show the relative importance of environmental persistence on phylodynamics of AIVs. Crucially, viral persistence may also occur in nonaquatic environments, including lake sediment, feathers, and feces [26],[32],[37].
Elsewhere, it has been shown that the environmental reservoir can be a crucial source for sparking off annual outbreaks [39] and may have an impact on interannual AIV durability [46]. As we have demonstrated here, another consequence of this feature of AIV transmission is broad strain coexistence that is similar to an ecological phenomenon called the “storage effect” [47],[48]. This effect has been identified for soil bacteria, where dormancy is thought to generate a high level of microbial diversity [49]. Dramatically, one of the main predictions of the storage effect theory is that large fluctuations in recruitment rates are expected of low-density species [48]. Indeed, as shown in Figure 3C, at any given time, the dominant strain may have first appeared far in the past, and as shown in Figure S9, this pattern of dominance is not predictable. Within an epidemiological context, this suggests that unpredictable outbreaks of rare subtypes may occur due to the absence of herd immunity.
Finally, our findings have practical implications for the management of influenza in wild birds. Particularly, our results indicate that—in addition to movement restrictions [50] and measures aimed at population size [51]—considering an environmental dimension to AIV control may be advisable [31]. Particularly, contaminated environments may remain infectious for an extended period following the cessation of transmission among hosts. If complete elimination of the virus is desired, then environmental decontamination may be required. Because of the outer lipid envelope associated with influenza viruses, chlorination has been proposed as a potentially effective method for decontamination [52],[53]. Given the impossibility of large-scale field trials, simulation exercises using models such as we report here may be crucial for determining whether such methods are indeed practically feasible.
The human epidemiological time series presented in Figure 1 is the death rates attributed to pneumonia and influenza reported in the United States [54]. This measure is known to correlate with human influenza activity [54], enabling qualitative description of population-wide influenza transmission. The subtype dominance patterns focus also on the United States and have been estimated through an annual sampling performed by the Center for Disease Control and Prevention [4]. Avian epidemiological time series and subtype dominance has focused on a duck population sampled in Alberta, Canada [55].
For the phylogenetic analyses presented in Figure 1, we focused on the HA gene for 10 subtypes sampled from wild bird species in North America between 1976 and 2001. We restricted our attention to this period because of the availability of parallel subtype-specific AIV prevalence data. For the remaining phylogenetic analyses (such as those presented in Table 1 and Figure 5), we examined sequence data from 1976–2013 from North America (United States and Canada) for avian, swine, and equine subtypes (Figures S2, S3, S4, S5, S6 and Table S1). For human influenza viruses, we considered only sequences from Memphis, Tennessee, in order to use a comparable number of sequences. Data for avian subtypes are available from the Dryad Digital Repository (http://dx.doi.org/10.5061/dryad.dryad.8ct18 [56]). Phylogenetic trees have been computed using the software BEAST [57] assuming a strict molecular clock [58], a site heterogeneity that is gamma distributed, a HKY substitution model, and a Bayesian Skyline Plot (BSP) with 10 groups [59]. The number of replicates was adjusted to maximize effective sample size; 5 millions replicates were used for burn-in.
We measured nucleotide substitution rate at third position sites as a proxy for nucleotide mutation, as evolution at third position sites should primarily represent synonymous change [60]. We also measured amino acid substitution rate, which is affected by both nucleotide mutation rate and selective effect of mutations. In the regression analysis, we find that amino acid substitution rate has a much stronger correlation with than does third position nucleotide substitution rate (coefficients −0.99 versus 0.27; Table 1). This suggests that it is the differing levels of selection on different HAs that determines viral diversity, rather than differing intrinsic mutation rates. The finding of a strong negative correlation between amino acid substitution rate and is consistent with the action of position selection driving amino acid replacements and purging diversity from viral populations [16].
Data on environmental durability for each subtype come from experimental data [38]. For each virus subtype, infective virions were diluted 1∶100 in water samples. The inoculated water samples were then divided into 3.0 ml aliquots in 5.0 ml polystyrene tubes and placed in incubators set to the appropriate treatment temperature. For each virus temperature trial, the viral inoculated water was sampled at the time of viral inoculation and at a second time point postinoculation. Titrations at all time points were performed in duplicate. The second postinoculation time point varied with each trial and was determined based on prior estimates of the time required for the titer of the virus in the water sample to be reduced by at least 1 log10 TCID50/ml [33],[35]. Duplicate 0.5 ml samples of AIV-inoculated water were diluted 1∶1 by addition of 0.5 ml of 2× serum-free MEM. Ten-fold dilutions (10−1 to 10−8) were then made in 1× MEM supplemented with antibiotics. These titers were used to estimate Rt as the time required to reduce infectivity by 90%, assuming a linear association.
The goal of regression analysis was to estimate the size of statistical effect on AIV diversity of covariates corresponding to alternative causal hypotheses. As described above, a covariate corresponding to each hypothesis was developed and assessed for each antigenic subtype. These covariates were (i) FST (the proportion of genetic variance contained in a subpopulation relative to the total variance) of each subtype (geographic hypothesis), (ii) number of species where strains have been sampled (host diversity hypothesis), (iii) subtype-specific nucleotide mutation rate (genetic hypothesis), (iv) amino acid substitution rates characterizing an immune selective pressure (immunological hypothesis), and (v) environmental durability Rt under natural physical conditions (temperature, 20°C; salinity, 0; pH 7.2; epidemiological hypothesis). Due to correlations among these variables, univariate analysis was not considered to provide reliable estimates of covariate effects. But ordinary least squares multiple regression would be equally ill-advised, resulting in weakly identifiable parameters and variance inflation. A generic solution to this problem is provided by penalized least squares models, such as ridge regression and elastic-net regression. These methods introduce a new estimator, which differs from the maximum likelihood estimator by an additional penalty. In effect, the penalized estimator trades a small amount of bias for a large reduction in the variance of the estimated coefficients. We chose to use elastic-net regression, which takes the maximum likelihood and ridge regression estimators as limit cases and therefore can be fine-tuned to balance the bias-variance tradeoff. Fitting of an elastic-net regression model requires the estimation of an additional tuning parameter (the penalty coefficient), which was numerically selected using cross-validation following [19]. The outcome of this procedure is a statistical model with coefficients shrunk to minimize generalization error. Covariates for which shrunk coefficients are zero can be inferred to have no effect.
The individual-based model developed here has been shown to generate evolutionary dynamics that are not statistically distinguishable from the classic SIR model in the limiting cases where analogous mathematical models can be still formulated [22]. Its main algorithm is detailed in Text S1 (section S3.1). Only nonneutral antigenic mutations have been explicitly considered within the model. The reconstruction of neutral mutations to infer the digital phylogenies has been implemented in a second step, detailed in Text S1, section S3.3. To avoid definitive extinctions, immigration of infectious individuals was included, with immigrant strains randomly selected according to the proportion of each variant present during the previous epidemic in order to avoid a strong influence of infectious immigration. Simulations start with four different strains far enough to avoid cross-immunity between them (see Text S1, section S3.2).
For human settings, we have assumed a constant population size of 106 individuals, with a mean lifespan of 80 y and a transmission rate of β(t) = 7.8.105(1+0.035cos(t)) [5]. The avian community is assumed to contain 104 individuals (host lifespan is 4 years) [39] with a seasonal demography integrated through a fluctuating birth rate b(t) = b(1+0.8·sin(t)) [30]. In both cases, β(t) has been chosen to ensure on average, as reported from previous studies [4],[30],[39]. Environmental transmission is characterized by an uptake rate of /L = 6.73 [30] and an environmental durability ξ(t) = 20×(1+0.9·sin(t)) (20 d on average [38]). To infer digital phylogenies, 100 strains have been sampled over the last 25 y from simulation runs.
|
10.1371/journal.pgen.1000017 | SIRT1 Inhibition Alleviates Gene Silencing in Fragile X Mental Retardation Syndrome | Expansion of the CGG•CCG-repeat tract in the 5′ UTR of the FMR1 gene to >200 repeats leads to heterochromatinization of the promoter and gene silencing. This results in Fragile X syndrome (FXS), the most common heritable form of mental retardation. The mechanism of gene silencing is unknown. We report here that a Class III histone deacetylase, SIRT1, plays an important role in this silencing process and show that the inhibition of this enzyme produces significant gene reactivation. This contrasts with the much smaller effect of inhibitors like trichostatin A (TSA) that inhibit Class I, II and IV histone deacetylases. Reactivation of silenced FMR1 alleles was accompanied by an increase in histone H3 lysine 9 acetylation as well as an increase in the amount of histone H4 that is acetylated at lysine 16 (H4K16) by the histone acetyltransferase, hMOF. DNA methylation, on the other hand, is unaffected. We also demonstrate that deacetylation of H4K16 is a key downstream consequence of DNA methylation. However, since DNA methylation inhibitors require DNA replication in order to be effective, SIRT1 inhibitors may be more useful for FMR1 gene reactivation in post-mitotic cells like neurons where the effect of the gene silencing is most obvious.
| Fragile X syndrome is the leading cause of heritable intellectual disability. The affected gene, FMR1, encodes FMRP, a protein that regulates the synthesis of a number of important neuronal proteins. The causative mutation is an increase in the number of CGG•CCG-repeats found at the beginning of the FMR1 gene. Alleles with >200 repeats are silenced. The silencing process involves DNA methylation as well as modifications to the histone proteins around which the DNA is wrapped in vivo. Treatment with 5-azadeoxycytidine, a DNA methyltransferase inhibitor, reactivates the gene. However, this reagent is toxic and since no DNA demethylase has been found in humans, methylation inhibitors are not useful in cells like neurons that no longer divide. We show here that splitomicin is also able to reactivate the Fragile X allele. It does so by inhibiting a protein deacetylase, SIRT1, thus favoring the action of another enzyme, hMOF that reverses the SIRT1 modification. We also found that 5-azadeoxycytidine acts, at least in part, by reversing the effect of SIRT1. However, since splitomicin reactivation occurred without DNA demethylation, DNA replication is not necessary for its efficacy. Thus, unlike DNA methylation inhibitors, SIRT1 inhibitors may be able to reactivate Fragile X alleles in neurons.
| The most common cause of Fragile X mental retardation syndrome (FXS) is the silencing of the FMR1 gene that occurs when the number of CGG•CCG-repeats in its 5′ untranslated region (5′ UTR) exceeds 200 [1],[2]. The net result is a deficiency in the FMR1 gene product, FMRP, a protein that regulates the translation of mRNAs important for learning and memory in neurons. How repeats of this length cause silencing is unknown. However, since the sequence of the promoter and open reading frame of these alleles is unchanged, the potential exists to ameliorate the symptoms of FXS by reversing the gene silencing.
The extent of silencing is related to the extent of methylation of the 5′ end of the gene [3],[4],[5]. Treatment of patient cells with 5-aza-dC, a DNA methyltransferase inhibitor, decreases DNA methylation and this is accompanied by partial gene reactivation [4],[5]. However, this compound has 2 major drawbacks: it is extremely toxic and it requires DNA replication to be effective. This would clearly limit its usefulness in vivo, particularly in post-mitotic neurons where the FMRP deficiency is most apparent. It also leaves open the question of whether DNA demethylation is necessary for gene reactivation to occur, a situation that for the reasons just mentioned, would severely limit the likelihood that gene reactivation would ever be a viable approach to treating FXS.
While the silenced gene is associated with overall H3 and H4 hypoacetylation, lysine 4 and 9 of histone H3 are the only 2 specific modifiable sites that have been examined thus far. In individuals with FXS, the levels of histone H3 acetylated at K9 (H3K9Ac) and H3 dimethylated at K4 (H3K4Me2) are decreased relative to the normal gene while the level of H3K9 dimethylation (H3K9Me2) is increased [5],[6],[7]. By analogy with other genes that have been studied more extensively, we would expect that there are a number of other histone residues that are differentially methylated or acetylated, when the FMR1 gene is aberrantly silenced.
The acetylation state of the histones associated with a particular genomic region is thought to play a critical role in regulating gene expression. The level of acetylation is dependent on the dynamic interplay of histone acetyltransferases (HATs) and histone deacetylases (HDACs). HDACs are sometimes divided into 4 functional classes based on sequence similarity. Class I (HDAC1, 2, 3, and 8) and class II (HDAC4, 5, 6, 7, 9, and 10) HDACs remove acetyl groups through zinc-mediated hydrolysis. Class III HDACs, which includes SIRT1, catalyze the deacetylation of acetyl-lysine residues by a mechanism in which NAD+ is cleaved and nicotinamide, which acts as an end product inhibitor, is released. Class IV HDACs are HDAC11-related enzymes that are thought to be mechanistically related to the Class I and II HDACs. To date, only inhibitors of Class I, II and IV HDACs have been tested for their ability to reactivate the FMR1 gene in FXS cells [4],[6],[8]. These HDAC inhibitors (HDIs), which include TSA and short-chain fatty acids like phenylbutyrate, have a much smaller effect on FMR1 gene reactivation than 5-aza-dC when used alone, although some synergistic effect was noted when these compounds were used in conjunction with 5-aza-dC [5],[6],[7],[9].
Recently, it has become apparent that not only do some HDACs act preferentially on specific lysines on different histones, but they also target certain genes for deacetylation [10]. Thus the available data did not rule out a role for HDACs, specifically Class III HDACs, in gene silencing in FXS. We show here that SIRT1, a member of the Class III HDAC family, plays an important role in silencing of FMR1 in the cells of Fragile X patients acting downstream of DNA methylation. Furthermore we show that SIRT1 inhibitors result in increased FMR1 transcription. This increase is associated with an increase in H4K16Ac and H3K9Ac but does not involve DNA demethylation or an increase in H3K4 dimethylation.
Nicotinamide (Vitamin B3), an end product inhibitor of NAD+-dependent enzymes like the Class III HDACs [11], increased FMR1 expression of a lymphoblastoid cell line from a Fragile X patient with a partially methylated FMR1 gene (GM06897) [12],[13]. Fifteen millimolar nicotinamide increased FMR1 mRNA levels by ∼3-fold while having little or no effect on the amount of FMR1 mRNA produced in normal cells (Figure 1A). A much smaller effect was seen in GM03200B cells in which the FMR1 gene is more heavily methylated [12],[13] and makes much less FMR1 mRNA (too small to see on the scale of the graphs shown in Figure 1A).
Splitomicin, a compound with a saturated six-membered lactone ring, is a more specific inhibitor of Class III HDACs and is thought to have a mechanism distinct from that of nicotinamide, inhibiting these enzymes by competing for binding of the acetylated substrate [14]. Splitomicin not only increased FMR1 mRNA levels in GM06897, but it produced a 200–600-fold increase in the amount of FMR1 mRNA in cell lines like GM03200B that were only minimally responsive to 15 mM nicotinamide (Figure 1B). This corresponded to a final FMR1 expression level that was ∼15–25% of normal, depending on which normal cell line was used for comparison. This level of activation was comparable to that achieved with 10 µM 5-aza-dC, an inhibitor of DNA methylation and much higher than the level of activation seen with TSA (Figure 2). The extent of activation was impressive given the low potency of splitomicin (in the micromolar range) and its relative instability (it has a half-life of 30 minutes at neutral pH [14]). A much smaller level of reactivation was seen with GM09145 and GM04025, lymphoblastoid cell lines that are more heavily methylated [12],[13] and that make less FMR1 than GM03200B (Figure 1C). A similar low level of reactivation was seen for 2 fibroblast cell lines that make very little FMR1 mRNA in the absence of splitomicin (Figure 1D). The simplest interpretation of these data is that a class III HDAC is involved in downregulating FMR1 expression from full mutation alleles. As has been reported for 5-aza-dC, the extent of reactivation is inversely related to the extent of silencing [6]. Whether the failure to completely reactivate the FMR1 gene with either drug reflects a suboptimal dosing strategy or the limits of what these classes of compounds can accomplish remains to be seen.
The ∼2-fold increase in FMR1 mRNA seen in GM06897 treated with 300 µM splitomicin is accompanied by a ∼2-fold increase in FMRP (Figure 2B and 2C). However, for cell lines where the FMR1 gene is more heavily methylated and that make no detectable FMRP, splitomicin did not result in the production of detectable levels of the FMR1 gene product (Figure 2B). The cell lines GM03200B, GM09145 and GM04025 are not only more heavily silenced than GM06897 but they also have more repeats (GM06897 has 477 repeats compared to 530 and 645 for GM03200B and GM04025 respectively). The failure to detect FMRP in these cells may reflect some combination of the low level of gene reactivation with the difficulty translating long CGG-repeat tracts previously reported for lymphocytes and lymphoblastoid cells [15],[16],[17],[18],[19].
Of the known class III HDACs, only SIRT1 is predominantly nuclear [20]. In order to assess whether SIRT1 was involved in FMR1 gene silencing, we transfected plasmids encoding a human SIRT1 protein and a dominant negative version of this construct (dnSIRT1) [21] into fibroblast cells from 3 different males, 1 who was unaffected and 2 with FXS. Fibroblasts were chosen because of the relative efficiency of transfection compared to lymphoblastoid cells. Transfection of the FXS fibroblasts (GM05131 and GM05848) with the normal SIRT1 construct led to a decrease in FMR1 expression from the low level seen in untransfected cells. In contrast a large increase in FMR1 expression was seen when the dnSIRT1 construct was used (Figure 3). This is consistent with a negative effect of SIRT1 on FMR1 transcription. Overexpression of these constructs only had a small effect on the level of FMR1 expression in unaffected individuals analogous to what was seen with nicotinamide and splitomicin.
To examine whether the effect of SIRT1 was direct or indirect, we carried out ChIP assays using an anti-HA antibody on a FXS cell line transfected with a construct encoding the HA-tagged SIRT1 [21]. The HA-tagged SIRT1 was enriched on the FMR1 allele in FXS cells compared to normal alleles (Figure 4).
SIRT1 binding to the promoter would be consistent with a role of this deacetylase in modification of the chromatin associated with the FMR1 gene in FXS cells. We therefore investigated the chromatin changes caused by splitomicin treatment using ChIP with antibodies to H3K9Ac and H4K16Ac since these are the major residues deacetylated by SIRT1 in vitro [22]. We also examined the levels of H3K4Me2, which is a mark of active chromatin that has been shown to increase when FXS alleles are reactivated with 5-aza-dC [7]. We examined the region upstream of the start of transcription and a region of exon 1 downstream of the repeat, with and without, splitomicin treatment. To better understand the differences between gene reactivation mediated by splitomicin and that mediated by 5-aza-dC we also examined the same histone modifications in these cells after 5-aza-dC treatment.
Both the promoter and exon 1 from a normal allele had higher levels of H3K9Ac and H3K4Me2 than the heavily silenced FMR1 full mutation allele, consistent with previous reports (Figure 5A and 5B, left and center panels). In unaffected cells splitomicin had little, if any, effect on the level of H3K9Ac on either the promoter or exon 1 (Figure 5A and 5B, left panel). However, splitomicin treatment of FXS cells increased H3K9Ac on ∼2-fold on the promoter and on ∼15-fold on exon 1. The net result of this increase is that H3K9Ac levels in FXS cells treated with splitomicin are very similar to that seen in normal cells. This suggests that SIRT1 is responsible for the hypoacetylation of H3K9 seen on FXS alleles, consistent with the observed in vitro properties of SIRT1 [22]. In contrast, 5-aza-dC had no effect on H3K9Ac in this region. The opposite situation was seen with H3K4Me2, in that splitomicin had no effect while 5-aza-dC caused a large increase in H3K4Me2 levels on exon 1 of the FXS allele (Figure 5B, center panel). However, both splitomicin and 5-aza-dC increased the levels of H4K16Ac on both the promoter and exon 1 of the FXS allele (Figure 5A and 5B, right panel). This suggests that DNA methylation and SIRT1 may act in the same or overlapping pathways and that this modification may play a key role in FMR1 gene silencing.
To assess the contribution of H4K16 acetylation to splitomicin-mediated FMR1 gene reactivation, we examined the effect of hMOF, a histone acetyltransferase that specifically targets H4K16 [23], on splitomicin-treated patient cells. As can be seen in Figure 6, transfection of patient fibroblasts with a dominant negative version of hMOF completely blocked the splitomicin-mediated increase in FMR1 mRNA, confirming the importance of H4K16 acetylation in FMR1 gene reactivation.
To examine the contribution of DNA demethylation to splitomicin-mediated gene reactivation we used an assay that monitors a region containing 8 CpG residues that is located just upstream of the CGG•CCG-repeat in the FMR1 gene [24]. Demethylation of a single cytosine produces a 0.5°C drop in the Tm of the PCR product obtained after bisulfite treatment. Reactivation with splitomicin did not change the Tm of the PCR product (Figure 7), suggesting that little, if any, demethylation occurred in this region. DNA demethylation-independent gene reactivation by splitomicin has also been seen in certain tumor suppressor genes aberrantly silenced in cancer cells [25].
In contrast, when these cells are treated with 5-aza-dC the Tm of the PCR product was indistinguishable from the results obtained from unaffected individuals (Figure 7). This is consistent with previous reports of the almost complete demethylation of the promoter by this treatment [4],[6],[9],[26].
We have shown that SIRT1, a class III HDAC, is involved in repeat-mediated FMR1 gene silencing via the deacetylation of H3K9 and H4K16. Our data suggests that deacetylation of H4K16 is also one of the major downstream consequences of DNA methylation. Since SIRT1 inhibition is able to reactivate the gene without affecting DNA demethylation, DNA methylation is not dominant over chromatin modifications like H4K16Ac with regard to gene expression. Furthermore, it demonstrates that DNA demethylation is not necessary for relieving gene silencing. This resembles the situation in Friedreich ataxia, another Repeat Expansion Disease, in which expanded alleles that are also aberrantly methylated at the DNA level [27], can be reactivated using an HDI alone [28].
The increased acetylation of H4K16 seen after treatment with both 5-aza-dC and splitomicin is important since the H4K16 acetylation status is thought to be a key determinant of chromatin accessibility [29]. However, the outcomes of the 2 treatments are not completely equivalent. DNA demethylation by 5-aza-dC is accompanied by an increase in H3K4Me2 that is not seen with splitomicin treatment. In contrast, splitomicin, but not 5-aza-dC, causes acetylation of H3K9. One interpretation of our data is that silenced alleles are associated with a methyl-binding protein or protein complex (MeBP) that binds to the methylated promoter and recruits SIRT1 (Figure 8). SIRT1 in turn deacetylates H3K9, H4K16 and potentially other residues as well. DNA demethylation causes the dissociation of the MeBP-SIRT1 complex from the promoter and creates conditions that favor the recruitment of H3K4 methylases and hMOF which specifically acetylates H4K16, but does not facilitate recruitment of a HAT that uses H3K9 as a substrate (Figure 8A). Splitomicin treatment, on the other hand, inhibits SIRT1 while leaving the promoter methylated. This helps generate a chromatin context conducive to recruiting both hMOF and an H3K9 HAT, but not an H3K4 methyltransferase (Figure 8B). Despite the differences in the final histone modification profile, the extent of gene reactivation resulting from the use of these compounds is similar and they show little additive effect when used in combination (data not shown). This raises the possibility that the most significant action of both compounds is exerted via the acetylation of H4K16 with both H3K4Me2 and H3K9Ac having little direct effect on gene expression.
Since the effect of splitomicin is not dependent on DNA replication, SIRT1 inhibitors may be more useful than 5-aza-dC for reversing FMR1 gene silencing in neurons which no longer divide and where the absence of FMRP is most debilitating. However, there are significant barriers to using SIRT1 inhibitors to treat FXS. Firstly, Sir2p, the yeast homolog of SIRT1, plays a role in the extension of lifespan in yeast [30] raising the possibility that SIRT1 inhibition may reduce lifespan in humans. However, there is some evidence that SIRT1 actually limits lifespan in mammals, at least in response to chronic genotoxic stress [31]. Furthermore, SIRT1 inhibition sensitizes cancer cells to apoptosis while sparing normal cells, making HDAC III inhibitors promising anti-cancer drugs [32]. It could also be argued that inhibition of HDACs could lead to inappropriate expression of other genes, which could be deleterious. However several HDIs are already approved for use in humans including dihydrocoumarin, an FDA approved food additive and valproate, a broad spectrum HDI, that has been used for decades in the treatment of epilepsy and is also an effective mood stabilizer. Today Valproate is one of the most highly prescribed antiepileptic drugs [33] and is already used in Fragile X patients to treat seizures, aggression and depression [34].
The fact that RNA with long CGG-repeat tracts is thought to be responsible for the Fragile X associated tremor and ataxia syndrome, a late onset neurodegenerative disorder seen in carriers of FMR1 premutation alleles [35], is a more general problem applicable to any gene reactivation approach for treating FXS. However, some HDIs have actually been shown to be neuroprotective [36],[37] and to expedite the recovery of learning and memory lost as a result of induced neurodegeneration [38]. Thus the beneficial effects of HDIs may help offset the negative effect of the expression of long CGG-repeat tracts.
The final impediment to gene reactivation approaches is the difficulty translating FMR1 transcripts with long CGG-tracts that has been seen in cells like lymphocytes and lymphoblasts [15],[16],[17],[18],[19]. However, there is reason to think that the translation difficulties do not affect all cells equally. For example, in Fragile X embryonic stem cells where the repeat is still unmethylated, both FMR1 mRNA and FMRP are made [39]. Furthermore we have shown that the negative effect of the repeats on translation is more severe in some parts of the mouse brain than others [16]. This is consistent with the fact that individuals with unmethylated full mutations show only mild symptoms of FXS [40],[41],[42]. It could thus be argued that when the FMR1 gene is not silenced, translation occurs at adequate levels in those parts of the brain critical for learning and memory. Even in lymphocytes and lymphoblastoid cells with ∼400 repeats some FMRP is made without treatment ([43] and this manuscript). The fact that even the GM06897 lymphoblastoid cell line, which has 477 repeats, makes some residual FMRP and that FMRP levels increase when the cells are treated with splitomicin, raises the possibility that increased RNA production may lead to increased FMRP production in the ∼40% of individuals with FXS who have repeats of <500 (Sally Nolin, personal communication). Even in lymphoblastoid cells there have been reports of FMRP production in cell lines with >800 repeats after reactivation with 5-aza-dC [4]. New SIRT1 inhibitors with higher stability, selectivity or potency [44] may allow the level of FMR1 transcription from previously silenced alleles to approach that seen in carriers of unsilenced full mutations. Since HDIs do not require DNA replication to be effective, this class of compounds may thus have therapeutic potential at least in that subset of individuals with repeat numbers that do not preclude translation.
Lymphoblastoid cells (GM02168, GM06895) and fibroblasts (GM00357) from unaffected males and lymphoblastoid cells (GM03200B, GM04025, GM09145) and fibroblasts (GM05131 and GM05848) from males with FXS were obtained from the Coriell Cell Repository (Camden, NJ). The antibodies used in this study were obtained from the following sources: anti-acetyl-Histone H4 (Lys 16) (Cat. #: ab1762) and anti-HA-tag (Cat. #: ab9110) were purchased from Abcam (Cambridge, MA); anti-acetyl-Histone H3 (Lys9) (Cat. #: 07-352), anti-dimethyl-Histone H3 (Lys 4) (Cat. #: 07-030) and anti-rabbit Ig were purchased from Millipore (Temecula, CA). Splitomicin and TSA were obtained from Tocris (Ellisville, MO). Nicotinamide and 5-aza-dC were obtained from Sigma (St. Louis, MO). The mutant human MOF (hMOF) construct in pcDNA3 was a kind gift of Arun Gupta (Washington University School of Medicine, St. Louis, MO). The pCRUZ-HA vector, pCRUZ-HA-SIRT1 and a dominant negative version of this construct were kindly provided by Toren Finkel (NHLBI, NIH, Bethesda, MD).
Lymphoblastoid cells were cultured in RPMI medium supplemented with 10% fetal bovine serum and 100 units each of penicillin and streptomycin (Invitrogen, Gaithersburg, MD). Fibroblasts were cultured in Minimum Essential Medium supplemented with 1% Glutamax, 10% fetal bovine serum and 100 units of penicillin and streptomycin (Invitrogen). All cells were grown at 37°C in 5% CO2. Cells were treated where indicated with either 300 µM or 700 µM splitomicin, 15 mM nicotinamide, or 3 µM TSA for 24 hours or 10 µM 5-aza-dC for 72 hours. Transfection of fibroblasts was carried out using Fugene 6 (Roche USA, Nutley, NJ) according to the supplier's instructions.
Total RNA was isolated from the cell lines using Trizol (Invitrogen) and reverse transcribed using SuperScript™ III RT First Strand Synthesis system for RT-PCR (Invitrogen), as per the manufacturer's instructions. Real time PCR was carried out using an ABI 7500 FAST PCR system (Applied Biosystems, Foster City, CA) using TaqMan™ Universal PCR master mix and FMR1 and GUS Taqman probe primer mixes (Applied Biosystems). For quantitation the comparative threshold (Ct) method was used with normalizing to GUS. The fold change was calculated by comparing the normalized treated versus untreated Ct values.
The ChIP assay kit from Upstate was used according to the manufacturer's instructions with slight modifications as previously described [27]. The amount of FMR1 promoter and exon 1 DNA immunoprecipitated with each antibody was determined using quantitative real time PCR as described below. Real time PCR was carried out using an ABI 7500 FAST PCR system and the Power SYBR™ Green PCR kit (Applied Biosystems). For amplification of the promoter region Promoter-F (5′-ACAGTGGAATGTAAAGGGTTG-3′) and Promoter-R (5′-GTGTTAAGCACTTGAGGTTCAT-3′) were used. This primer pair amplifies the 140 bp region from 146800256–146800396 of the human genome sequence (March, 2006 assembly, http://genome.ucsc.edu/cgi-bin/hgBlat) which terminates 736 bp upstream of the 3′ most transcription start site. For amplification of exon 1, the primer pair Exon1-F (5′-CGCTAGCAGGGCTGAAGAGAA-3′) and Exon1-R (5′-GTACCTTGTAGAAAGCGCCATTGGAG-3′) was used. This primer pair amplifies the region 146801368–146801444 of the human genome sequence that corresponds to the region in exon 1 236–311 bp downstream of the transcription start site. All experiments were done in triplicate. The ChIP experiments were performed in triplicate and each PCR reaction was done in duplicate. The immunoprecipitated DNA was expressed relative to the amount of input DNA that constituted 10% of the original material. GAPDH was used for normalization using hs_GAPDH exon1F1 primer (5′-TCGACAGTCAGCCGCATCT-3′) and hs_GAPDH intron1R1 (5′-CTAGCCTCCCGGGTTTCTCT-3′).
Genomic DNA from cell lines was bisulphite modified according to standard procedures except that the bisulphite treatment was carried out overnight at 55°C. The methylation status of the promoter was determined as previously described [24].
SDS protein gel electrophoresis and Western blotting of protein extracts was carried out using standard procedures. Anti-FMRP antibody (MAB2160, Millipore) was used to detect FMRP. Anti-β-actin antibody (Abcam) was used to normalize the FMRP levels for variations in protein loading. Detection of antibody binding was carried using an ECL™ kit (Amersham, Buckinghamshire, UK) according to the manufacturer's instructions. The amount of FMRP and β-actin were determined by standard densitometry. The increase in FMRP was calculated based on the average of 3 independent experiments. |
10.1371/journal.pgen.1006968 | Kek-6: A truncated-Trk-like receptor for Drosophila neurotrophin 2 regulates structural synaptic plasticity | Neurotrophism, structural plasticity, learning and long-term memory in mammals critically depend on neurotrophins binding Trk receptors to activate tyrosine kinase (TyrK) signaling, but Drosophila lacks full-length Trks, raising the question of how these processes occur in the fly. Paradoxically, truncated Trk isoforms lacking the TyrK predominate in the adult human brain, but whether they have neuronal functions independently of full-length Trks is unknown. Drosophila has TyrK-less Trk-family receptors, encoded by the kekkon (kek) genes, suggesting that evolutionarily conserved functions for this receptor class may exist. Here, we asked whether Keks function together with Drosophila neurotrophins (DNTs) at the larval glutamatergic neuromuscular junction (NMJ). We tested the eleven LRR and Ig-containing (LIG) proteins encoded in the Drosophila genome for expression in the central nervous system (CNS) and potential interaction with DNTs. Kek-6 is expressed in the CNS, interacts genetically with DNTs and can bind DNT2 in signaling assays and co-immunoprecipitations. Ligand binding is promiscuous, as Kek-6 can also bind DNT1, and Kek-2 and Kek-5 can also bind DNT2. In vivo, Kek-6 is found presynaptically in motoneurons, and DNT2 is produced by the muscle to function as a retrograde factor at the NMJ. Kek-6 and DNT2 regulate NMJ growth and synaptic structure. Evidence indicates that Kek-6 does not antagonise the alternative DNT2 receptor Toll-6. Instead, Kek-6 and Toll-6 interact physically, and together regulate structural synaptic plasticity and homeostasis. Using pull-down assays, we identified and validated CaMKII and VAP33A as intracellular partners of Kek-6, and show that they regulate NMJ growth and active zone formation downstream of DNT2 and Kek-6. The synaptic functions of Kek-6 could be evolutionarily conserved. This raises the intriguing possibility that a novel mechanism of structural synaptic plasticity involving truncated Trk-family receptors independently of TyrK signaling may also operate in the human brain.
| A long-standing paradox had been to explain how brain structural plasticity, learning and long-term memory might occur in Drosophila in the absence of canonical Trk receptors for neurotrophin (NT) ligands. NTs link structure and function in the brain enabling adjustments in cell number, dendritic, axonal and synaptic patterns, in response to neuronal activity. These events are essential for brain development, learning and long-term memory, and are thought to depend on the tyrosine-kinase function of the NT Trk receptors. However, paradoxically, the most abundant Trk isoforms in the adult human brain lack the tyrosine kinase, and their neuronal function is unknown. Remarkably, Drosophila has kinase-less receptors of the Trk family encoded by the kekkon (kek) genes, suggesting that deep evolutionary functional conservation for this receptor class could be unveiled. Here, we show that Kek-6 is a receptor for Drosophila neurotrophin 2 (DNT2) that regulates structural synaptic plasticity via CaMKII and VAP33A. The latter are well-known factors regulating synaptic structure and plasticity and vesicle release. Furthemore, Kek-6 cooperates with the alternative DNT2 receptor Toll-6, and their concerted functions are required to regulate structural homeostasis at the NMJ. Our findings suggest that in mammals truncated Trk-family receptors could also have synaptic functions in neurons independently of Tyrosine kinase signaling. This might reveal a novel mechanism of brain plasticity, with important implications for understanding also the human brain, in health and disease.
| Brain plasticity, neurotrophism in development, structural and synaptic plasticity during learning and long-term memory in humans critically depend on the receptor TrkB binding its neurotrophin (NT) ligand BDNF [1,2]. During development, NTs and Trks regulate neuronal number and connectivity; subsequently BDNF and TrkB establish a reinforcing positive feedback loop promoting synaptic potentiation, and regulate the dynamic generation and elimination of synaptic boutons and dendritic spines in response to activity [1,3–5]. Thus, NTs and Trks are fundamental to linking structure and function in the brain, and this is thought to depend mostly on the tyrosine kinase (TyrK) function of Trks. Through its intracellular TyrK domain, TrkB activates the Ras/MAPKinase, PI3Kinase/ AKT and PLCγ signaling pathways downstream [1,2]. Pre-synaptic targets include Synapsin, to trigger vesicle release [5]. Post-synaptic targets include NMDAR and CREB, essential for long-term potentiation, learning and long-term memory [1,5]. Paradoxically, full-length TrkB decays postnatally and TrkB homodimers are not formed in the adult mammalian brain [6–10]. Instead, the most abundant adult isoform is truncated TrkB-T1 lacking the TyrK [8–10]. Mutant mice lacking TrkB-T1 have anxiety, and in humans alterations in TrkB-T1 are linked to severe mental health disorders [11–13]. However, the neuronal functions of the truncated Trk isoforms are poorly understood.
No canonical, bona fide, full-length, TyrK Trk receptors have been found in Drosophila. However, neurotrophism, structural and synaptic plasticity, learning and long-term memory all occur in fruit-flies, implying that either in the course of evolution insects and humans found different molecular solutions to elicit equivalent functions, or that undiscovered mechanisms contribute to brain plasticity in both humans and fruit-flies. Finding out what happened to the Trks in Drosophila is important, as it could uncover novel fundamental mechanisms of structure-function relationships in any brain.
Trk receptors have long been searched for in Drosophila. Trks (TrkA,B,C) bear the unique combination of Cysteine Rich Repeats (CRR), Leucine Rich Repeats (LRR) and Immunoglobulin (Ig) domains extracellularly, and an intracellular TyrK domain (Fig 1A). Original searches focused on the TyrK domain, and identified DTrk and Dror as candidate Drosophila Trk homologues, but these are unlikely to bind neurotrophins. DTrk, also known as Off-Track (Otk)[14,15] lacks the LRR and CRR modules, it is kinase-dead and binds Semaphorins (Fig 1A). Dror/Dnrk, like all Ror-family receptors, has an extracellular Frizzled/Kringle module instead [16–18](Fig 1A). Subsequent proteomic analyses found no full-length, canonical Trk orthologues with the combination of LRR, CRR and Ig modules extracellularly and a Trk-family TyrK intracellularly in Drosophila [19–21]. However, phylogenetic analysis of the Trk-receptor superfamily identified the Drosophila Kekkons (Keks), lacking an intracellular TyrK domain, as closely related to the Trks [22](Fig 1A and 1B). Trks and Keks both belong to the LIG family of proteins that contain extracellular ligand-binding LRR and Ig motifs [22,23]. There are 38 LIGs in humans, and amongst these are transmembrane proteins with a divergent intracellular domain lacking a TyrK or any conserved motifs [22]. There are 9 LIGs in Drosophila. Phylogenetic analysis clusters mammalian AMIGO, LINGO, LRIG and LRRC4 in one clade with Drosophila Lambik, and mammalian Trks in a separate clade together with Drosophila Keks (Kek1-6) [22](Fig 1B). Keks are more similar to the Trks than all other vertebrate LIGs are to each other [22]. Keks have only been found in insects, and thus are remnant, conserved Trk-like receptors in fruit-flies.
There are 6 Keks in Drosophila, with a cytoplasmic tail lacking any remarkable conservation, except that Kek-1, -2, -5 and -6 have a PDZ domain that could bind cytoplasmic effectors [24]. Kek-1, -2, -5 and -6 are highly conserved in insects [24]. At least kek-1, kek-2 and kek-5 are expressed in the CNS [25–27]. Only Kek2 has been investigated in the CNS, where it functions as a neuronal activity dependent modulator of synaptic growth and activity [27]. The ligands for the Keks have not been identified.
The prime candidates for Kek ligands are the Drosophila neurotrophins (DNTs). DNT1 (Drosophila neurotrophin 1 also known as Spz2), DNT2 (also known as Spz5) and Spz bear the distinctive and evolutionarily conserved neurotrophin cystine-knot domain of the mammalian neurotrophins [28–36], and they have conserved CNS functions regulating neuronal survival, connectivity and synaptogenesis [35–39]. Spz, DNT2 and DNT1 are known ligands for Toll-1, Toll-6 and Toll-7 receptors of the Toll and Toll-Like Receptor (TLR) superfamily [38,40]. In Drosophila Toll-1, Toll-6 and Toll-7 are required for targeting at the embryonic neuromuscular junction (NMJ), Toll-6 and Toll-8 for larval NMJ growth, and Toll-1, Toll-6 and Toll-7 function as neurotrophin receptors regulating neuronal survival and death, connectivity and behaviour [35–38,41–44]. In Drosophila, the NMJ is glutamatergic, and undergoes plasticity and potentiation, thus resembling mammalian central synapses. The NMJ is the standard context in which to investigate synaptic structural and functional plasticity in Drosophila [45]. Given that NT family ligands can bind multiple receptor types, and that receptors and ligands tend to co-evolve [17], conservation of the extracellular ligand-binding domain of Trks and Keks suggested Keks could potentially function as DNT receptors in flies.
Here, we investigated whether Kek-6 might function as a DNT receptor in Drosophila, at the glutamatergic NMJ synapse.
To investigate if Drosophila LIGs might function as DNT receptors in the CNS, we first looked at their expression in embryos. CG15744 did not reveal expression above background in the ventral nerve cord (VNC) (Fig 1C); lambik and CG16974 mRNAs were absent from the VNC, but might be expressed in the Peripheral Nervous System (PNS) or muscle, respectively (Fig 1C and 1D). kek1 and kek2 were expressed in VNC cells, and possibly in PNS cells too (Fig 1D), as previously shown [25,27]. kek-3 and kek-4 transcripts were not detected in embryonic CNS. kek-5 expressed in the VNC [26], and we confirmed this with a kek5-GAL4 reporter, which also revealed PNS expression (Fig 1D). kek-6 transcripts were abundant in the VNC (Fig 1D). Thus, amongst the 9 LIGs, Kek-1, -2, -5, and -6 could function in the CNS.
To test whether Keks could function downstream of DNTs in vivo, we took advantage of the cold semi-lethality of DNT141 DNT2e03444 double mutants [38], and asked whether it could be rescued with the over-expression of keks in neurons. Over-expression of kek-1 and kek-4 in neurons (with elavGAL4) did not rescue, and kek-2 and kek-6 did most prominently (Fig 1E, S1 Table). Thus, Kek-2 and Kek-6 could function downstream of DNTs.
To ask whether DNT ligands could bind Keks and induce signaling, as Keks lack the TyrK, to enable a signaling readout, we generated chimaeric receptors formed of the extracellular and transmembrane domains of the Keks fused to the intracellular domain of Toll-6 (Fig 2A). Toll-6 uses a conserved TIR domain to activate Dif/NFκB signaling downstream, which can be measured through the activation of drosomicin-luciferase (dros-luc) [38]. We used S2 cells stably transfected with dros-luc, transfected them with kek-Toll-6 chimaeric receptors, and tested whether stimulation with purified cleaved DNT2 (DNT2-CK) could induce Dif/NFκB signaling. The chimaeric receptors targeted correctly to the S2 cell membrane (Fig 2A). Stimulation with DNT2-CK in pDONR transfected controls induced dros-luc, as S2 cells express multiple Toll receptors [36,38] (Fig 2A). Stimulation with DNT2-CK of cells transfected with kek3,4,5,6-Toll-6 chimaeras had an effect comparable to the induction by stimulated full-length Toll-6, and Kek3-Toll-6 and Kek6-Toll-6 chimaeras responded more robustly (Fig 2A, S1 Table). We could not generate Kek-1 and Kek-2 chimeric receptors, thus a potential interaction with them cannot be ruled out. Thus, DNT2 can interact physically with at least Kek-3 and Kek-6.
To verify whether Kek-6 could bind DNT2, we carried out co-immunoprecipitations. S2 cells were co-transfected with HA-tagged kek-6 and V5-tagged full-length DNT2. Precipitating Kek6-HA with anti-HA, brought down DNT2-V5 detected with anti-V5 (Fig 2B). Conversely, precipitating DNT2 with anti-V5, also brought down Kek6 detected with anti-HA (Fig 2C). Thus, Kek6 can bind DNT2. To test whether Kek-6 might also bind DNT1, we co-transfected S2 cells with kek6-HA and full-length DNT1-V5. Precipitating DNT1 brought down Kek-6 (Fig 2C). Thus, Kek-6 can bind both DNT2 and DNT1. To test if DNT2 could bind other Keks, S2 cells were co-transfected with kek5-Flag and DNT2-V5, and precipitating DNT2 also brought down Kek-5 (Fig 2D). Similarly, S2 cells were co-transfected with kek2-Flag and DNT2-HA, and precipitating Kek-2 also brought down DNT2 (Fig 2E). However, upon co-transfection, precipitating Kek3-Flag failed to robustly bring down DNT2-HA (Fig 2F). Thus, DNT2 can equally bind Kek-5 and Kek-2, but cannot bind Kek-3 as well. These data demonstrate that DNT ligands bind Keks, that binding is somewhat promiscuous, and that DNTs might preferentially bind the CNS-specific Kek-2, -5 and -6.
As Kek-6 was widely expressed in the CNS, rescued the semi-lethality of DNT1 DNT2 double mutants, and bound DNT ligands in both a signaling assay and in co-immunoprecipitations, from this point we focused on Kek-6. We asked whether Kek-6 could function as a DNT receptor.
For a functional in vivo analysis we focused on Kek-6. kek-6 CNS expression was examined in larvae using kek6MIMIC13953 flies bearing a GFP insertion into the kek-6 coding region (hereafter named Kek6GFP). Kek6GFP was found in HB9+(Fig 3A) and Eve+(S1A Fig) inter-neurons and motoneurons, and excluded from Repo+ glia (S1B Fig) in third instar larval VNCs. Kek6GFP was present in motoneuron terminals at the neuro muscular junction (NMJ) of third instar larvae and not in the muscle (Fig 3B and 3C), revealing NMJ6/7 synaptic boutons, surrounded by post-synaptic Dlg (i.e. Drosophila PSD95) (Fig 3C). To verify this, we swapped MIMIC for GAL4 using RMCE, and drove expression of the membrane tethered FlyBow reporter. kek6>FlyBow revealed abundant signal in CNS neurons, axons and dendrites, and in the cell body clusters of RP3,4,5 motoneurons (Fig 3D). kek6>FlyBow was also present in motoraxons reaching the NMJ, and in synaptic boutons (Fig 3F). No prominent signal was detected in the muscle above background levels. However, data do not exclude potential post-synaptic distribution in the CNS. To conclude, kek-6 is expressed pre-synaptically in motoneurons (Fig 3E).
DNT2 transcripts are expressed in the larval body wall muscles, and localised to post-synaptic boutons [39]. To test if DNT2 could function retrogradely, we generated a tagged form of full-length DNT2 with GFP at the C-terminus, and over-expressed DNT2-FL-GFP in the muscle with MhcGAL4. DNT2-FL-GFP would result in secretion of mature DNT2-CK-GFP [36], although not all protein might get cleaved and secreted. Over-expression of DNT2-FL-GFP in muscle resulted in the localization of GFP pre-synaptically in boutons, surrounded by the post-synaptic marker Dlg (Fig 3G). DNT2-FL-GFP also colocalised with the motoneuron marker FasII in boutons (Fig 3H, 3I and 3K) and motoraxons (Fig 3I, 3J and 3K). Thus, DNT2 produced in muscle could get distributed to the motoneuron, consistent with a retrograde function.
To investigate the in vivo functions of Kek-6 and DNT2, we generated kek-6 and DNT2 null mutant alleles by FRT-mediated recombination between PiggyBac insertions [46] (S2 Fig). Neither kek634/Df(3R)ED6361 or DNT237/Df(3L)6092 loss of function mutants, nor kek6–/–DNT2–/–double mutants, affected viability. We analysed larval locomotion using FlyTracker software to trace trajectories and measure crawling speed [38], and found that kek634/Df(3R)ED6361 mutant larvae crawled more slowly than controls (Fig 4A and 4B, S1 Table), and DNT237/Df(3L)6092 larvae crawled even slower (Fig 4A and 4B, S1 Table). kek6–/–DNT2–/–double null larvae crawled at similar speeds as DNT2–/–mutants, but travelled shorter distances, moving around the starting spot (Fig 4A and 4B, S1 Table). Overall, DNT2–/–mutants and the double mutants crawled slowest (Fig 4C). Furthermore, kek6–/–and DNT2–/–single mutants and kek6–/–DNT2–/–double mutants spent considerably longer times than controls not moving (Fig 4D). However, when crawling, both single mutants, and kek6–/–DNT2–/–double mutants could attain the fastest speeds achieved by controls (Fig 4C and 4E). All genotypes only spent very brief times at these high speeds (Fig 4C and 4E). The shared phenotypes of single and the double mutants, and synergistic effect in the doubles (Fig 4A), suggested that Kek-6 and DNT2 might be functionally linked.
Locomotion phenotypes suggested the NMJ might be affected, so we looked at the muscle 6/7 NMJs, which require DNT1 and 2 [39]. Targeting at the embryonic muscle 6/7 NMJ was affected in kek6–/–mutants and upon kek-6 over-expression in neurons (S3 Fig). In wandering third instar larvae, kek6–/–and DNT2–/–single mutant larvae, and kek6–/–DNT2–/–double mutants, had smaller NMJs than controls, with fewer Ib boutons and shorter axonal terminal length (Fig 4F, S1 Table). All mutant genotypes also had reduced NMJ branching (Fig 4F). Thus, Kek-6 and DNT2 are required for normal NMJ growth. kek6–/–and DNT2–/–single mutant NMJs had higher active zone density, visualized with anti-Brp (Drosophila ELKs) and quantified automatically throughout the NMJ stack of images using DeadEasy Synapse software [39](Fig 4F, S1 Table). Since the NMJs were smaller, this suggested that the increase in active zones was a homeostatic compensation of defective synaptic function, to enable adequate behaviour. Homeostatic adjustments in active zones are a common manifestation of structural plasticity at the NMJ [47]. Remarkably, the increase in active zone density did not occur in kek-6–/–DNT2–/–double mutants (Fig 4F, S1 Table), meaning that compensation fails in the double mutants. To further test how Kek-6 affects the synapse, we also visualised Synapsin, which phosphorylates components of the SNARE complex to promote neurotransmitter vesicle release [48], and quantified it automatically using DeadEasy Synapse software [39]. kek6–/–mutants had reduced Synapsin production at synaptic locations (Fig 4F, S1 Table), revealing defective synapse composition. These data show that: (1) Kek-6 is required for appropriate synaptic structure; (2) that in kek-6–/–mutants homeostatic compensation regulates Brp but not Synapsin; (3) and that DNT2 is involved in the compensation mechanism, as it fails in its absence.
Over-expression of kek-6 in motoneurons alone did not alter NMJ size, but increased branching (Fig 5A and S4A and S4B Fig) and active zone density (Fig 5A, S1 Table). It also induced ghost boutons, albeit not significantly (similarly in kek-6–/–mutants, S4C–S4E Fig, S1 Table). Ghost boutons are presynaptic, HRP+ Dlg-negative, immature boutons that fail to get stabilized, and are correlates of increased neuronal activity. Together, these data are consistent with Kek-6 influencing synaptic structure and/or function. Over-expression of DNT2-FL in muscle increased axonal terminal size and branching (Figs 5B and 6B). Furthermore, over-expression of full-length DNT2-FL in muscle, and either full-length DNT2-FL or mature DNT2-CK in motoneurons (with D42GAL4), also increased active zones (Fig 5B and 5C, S1 Table). Thus, DNT2 can affect NMJ size and synaptic structure.
Altogether, these data showed that both Kek-6 and DNT2 are required: (1) for NMJ growth, (2) for appropriate synaptic structure. Furthermore, whereas kek-6 cannot promote NMJ growth, DNT2 can.
Data suggested that DNT2 may function as a retrograde ligand for Kek-6. To further test this, we used epistasis analysis. Over-expression of kek6 in motoneurons rescued the NMJ mutant phenotypes of bouton number and axonal length of kek634/Df(3R)6361 mutants (Fig 6A, S1 Table), demonstrating that the kek6–/–mutant phenotypes were specific. Over-expression of kek6 in motoneurons rescued the NMJ phenotypes of DNT237/Df(3L)6092 mutants and kek6–/–DNT2–/–double mutants (Fig 6B and 6C, S1 Table), demonstrating that Kek-6 functions downstream of DNT2. Over-expression of untagged DNT2-FL in muscle (with MhcGAL4) increased Dlg+ bouton number (Fig 6B), and importantly, this was rescued by kek-6 loss of function (Fig 6B, S1 Table). Since Kek-6 is expressed in motoneurons and functions pre-synaptically, and DNT2 was over-expressed in muscle, this demonstrates that DNT2 is a retrograde ligand for Kek-6. Altogether, our data demonstrate that DNT2 is a retrograde ligand for Kek-6 at the NMJ and that DNT2 and Kek-6 are required together for synaptic structure and NMJ growth.
Intriguingly, our data also suggested that DNT2 had additional functions compared to kek-6. Firstly, the homeostatic compensation of active zones seen in kek6–/–mutants did not occur in kek-6–/–DNT2–/–double mutants (Fig 4F). This suggests that homeostatic compensation depends on an alternative mechanism downstream of DNT2. Secondly, over-expression of DNT2 but not kek-6 increased NMJ size, as both HRP+ axonal terminal length (Fig 5B) and bouton number (Fig 6B) increased when DNT2-FL was over-expressed from muscle. This suggests that a second mechanism downstream of DNT2 can influence NMJ growth. DNT2 is a known ligand of Toll-6 [38], and Toll-6 and -8 are required in motoneurons for NMJ growth and active zone formation[37,43]. So this raised two questions: do Toll-6 and Kek-6 interact functionally as DNT2 receptors at the NMJ, and why?
LIG proteins and truncated Trk isoforms can function as ligand sinks or dominant negative co-receptors that abrogate signaling, e.g. by full-length Trks [6,7]. Thus, Kek-6 might inhibit Toll-6 function.
Toll-6 was shown to function at the larval NMJ [43], but its expression here had not been reported. Using a MIMIC insertion into the intronless coding region of Toll-6, we generated Toll-6GAL4 flies by RMCE. Toll-6>mCD8-GFP was distributed along the motoraxon of the muscle 6/7 NMJ (Fig 7A). Like kek-6–/–and DNT2–/–mutants, Toll-6MIO2127/ Df(3L)BSC578 mutants had smaller NMJs, with fewer 1b boutons (Fig 7B), shorter HRP+ axonal length (Fig 7B, S1 Table) and reduced branching (Fig 7B) than controls. This confirms that Toll-6 is required for NMJ growth. Contrary to kek-6–/–mutants, Toll-6–/–mutants had reduced Brp+ active zones compared to controls (Fig 7B), meaning that Toll-6 is required for active zone formation. kek-6–/–Toll-6–/–double mutant larvae also had small NMJs (Fig 7B, S1 Table), with reduced branching complexity (Fig 7B), both consistently more pronounced than the single Toll-6 mutants when compared to controls. Importantly, Brp+ active zones remained lower than in controls and comparable to the levels of Toll-6–/–single mutants (Fig 7B). Thus, the compensatory increase in active zones observed in kek-6–/–single mutants did not occur in the absence of Toll-6. Conversely, over-expression of Toll-6 did not alter NMJ size, but increased active zones (Fig 7B). Thus, Toll-6 can induce Brp+ active zone formation. Together, these data show that active zone formation and homeostasis depend on Toll-6.
Evidence indicated that Brp depends on Toll-6 and not Kek-6, but both Kek-6 and Toll-6 could induce Brp+ active zones. This could mean that Kek-6 might induce Brp+ indirectly via Toll-6. To test this, we asked whether Kek-6 could induce active zone formation in the absence of Toll-6. Over-expression of kek-6 presynaptically in motoneurons in Toll-6–/–mutants increased Brp+ (Fig 7B, yellow). This demonstrates that Kek-6 can induce active zone formation independently of Toll-6.
These results revealed that: 1) Loss of both kek-6–/–and Toll-6–/–more robustly reduced NMJ size and complexity than loss of Toll-6–/–alone, suggesting that Kek-6 contributes to NMJ growth via an alternative mechanism independently of Toll-6. 2) NMJ growth requires DNT2, Kek-6 and Toll-6. However, over-expression of DNT2 had a much stronger effect than over-expressing either kek-6 or Toll-6 alone. This means that NMJ growth is regulated by the concerted functions of both DNT2 receptors, Toll-6 and Kek-6. 3) Synaptic structure requires Toll-6 and Kek-6, but via distinct parallel mechanisms. 4) The compensatory increase in active zones observed in kek-6–/–and DNT2–/–single mutants did not occur in kek-6–/–DNT2–/–or kek-6–/–Toll-6–/–double mutants, meaning that the concerted functions of Toll-6 and Kek-6 are required for NMJ structural synaptic homeostasis. 5) Since Toll-6 is required for active zone formation and NMJ growth, and over-expression of kek-6 increased active zones and did not compromise NMJ growth, Kek-6 does not function as a ligand sink, an inhibitor or dominant negative co-receptor to abrogate Toll-6 function.
The data suggested that Kek-6 and Toll-6 can function together forming a receptor complex for DNT2. To test whether Kek-6 and Toll-6 might physically interact, we carried out co-immunoprecipitations. S2 cells were co-transfected with the active forms Toll-6CY-Flag and Kek6-HA, and we found that precipitating Toll-6CY with anti-Flag brought down Kek6 detected with anti-HA (Fig 7C). Similarly, precipitating Toll-7CY also brought down Kek6 (Fig 7C). Thus, Kek6 can bind Toll-6, and also Toll-7.
Altogether, these data showed that retrograde DNT2 can bind Toll-6 and Kek-6, which can interact pre-synaptically to form a receptor complex. They cooperatively promote NMJ growth and regulate synaptic structure. This raised a new question: if Kek-6 could influence NMJ growth and synaptic structure independently of Toll-6, and without a TyrK, how might it function?
To find out how Kek-6 might function, we carried out pull-down assays to isolate candidate factors binding its intracellular domain. S2 cells were transfected with kek6-Flag, and anti-Flag coated beads were used to expose Kek-6 to cell lysates from either S2 cells or wild-type adult fly heads, and bound proteins were isolated by SDS-PAGE followed by mass spectrometry (Fig 8A and 8B). Candidates were identified as proteins present in Kek6-Flag samples and absent from non-transfected mock controls, and if identified from multiple peptides (S2 and S3 Tables). Prevalent amongst these were proteins involved in vesicle trafficking, axonogenesis, dendrite morphogenesis and synaptic function (Fig 8C). Amongst the top hits were CaMKinase II identified from fly heads, and VAP33A from both S2 cells and heads (Fig 8C). CaMKII functions both as a kinase and a scaffolding protein, to promote structural synaptic plasticity. Post-synaptically, it phosphorylates and recruits AMPAR and NMDAR to the post-synaptic density, leading to post-synaptic potentiation, and pre-synaptically it localizes to active zones and phosphorylates Synapsin and other SNARE complex proteins, triggering neurotransmitter release [48–51]. VAP33A is a Vamp Associated Protein, with evolutionarily conserved functions in exocytosis and vesicle trafficking, at synapses and in wider contexts [52]. To validate these two candidates as downstream effectors of Kek-6, we carried out co-immunoprecipitations. We co-transfected S2 cells with Kek6-Flag and either CaMKII or VAP33A tagged with HA. Precipitating Kek-6-Flag, brought down CaMKII-HA (Fig 8D). Similarly, precipitating Kek-6-Flag also brought down VAP33A-HA (Fig 8D). Thus, Kek-6 can bind CaMKII and VAP33A.
To validate the functional relationship between Kek-6 and CaMKII in vivo, we asked whether altering Kek-6 function would affect CaMKII activation in brain. We tested for the constitutively active state of CaMKII, corresponding to phosphorylation at Thr286—T287 in Drosophila—using antibodies that detect pCaMKIIT287. In the heads of kek634/Df(3R)ED6361 mutant adult flies the relative levels of pCaMKIIT287, normalised over non-phosphorylated CaMKII, decreased (Fig 8E, S1 Table). Conversely, over-expression of kek-6 in retina (with GMRGAL4) increased CaMKII phosphorylation (Fig 8E, S1 Table). Over-expressing DNT2-FL had the same effect (Fig 8E). Thus, Kek-6 is required for CaMKII activation, and both Kek-6 and DNT2 can activate CaMKII downstream.
Next we asked whether CaMKII and VAP33A might function downstream of Kek6 at the NMJ. Unfortunately, we were not able to get antibodies to inactive CaMKII to work reliably at the NMJ for normalisation, so we visualised constitutively activated CaMKII with anti-pCaMKIIT287 and quantified it automatically, using DeadEasy Synapse [39]. DeadEasy Synapse detected the increase in pCaMKII caused by the over-expression of activated CaMKIIT287 in motoneurons (D42>CaMKIIT287D, S5A and S5C Fig). Over-expression of CaMKIIT287 pre-synaptically in motoneurons increased axonal length, Ib bouton number and active zone density (S5A and S5C Fig, S1 Table). Conversely, inhibiting CaMKII only pre-synaptically by over-expressing the CaMKII phosphorylation inhibitor Ala [53] in motoneurons resulted in smaller NMJs and reduced active zones (S5A and S5B Fig, S1 Table), as previously reported [54]. Thus we asked whether Kek-6 influenced CaMKII at the NMJ. Pre-synaptic over-expression of kek-6 in motoneurons increased pCaMKIIT287 levels (D42>kek6 Fig 9A, S1 Table). This increase was rescued by the pre-synaptic over-expression of Ala together with kek-6 (D42>kek6, Ala) showing that this phenotype was specific (Fig 9A, S1 Table). However, pre-synaptic over-expression in motoneurons of Ala alone (D42>Ala) did not result in a detectable reduction in pCaMKII in this case (Fig 9A, S1 Table). To try an alternative approach, we knocked-down pre-synaptic CaMKII expression with RNAi in motoneurons, and this decreased overall pCaMKIIT287 levels, causing a stronger effect than Ala (D42>CaMKII-RNAi, Fig 9B S1 Table). Like Ala, pre-synaptic CaMKII-RNAi knock-down together with kek-6 over-expression in motoneurons also rescued pCaMKIIT287 levels, and reduced them further than controls (D42>kek6, CaMKII-RNAi, Fig 9B, S1 Table). This demonstrates that CaMKII functions downstream of Kek-6 in motoneurons. Together, these data show that Kek-6 function results in the activation of CaMKII pre-synaptically.
Pre-synaptic activated CaMKII localizes active zones [49,55], so next we asked whether CaMKII was required for the increased active zones caused by kek-6 gain of function. Pre-synaptic CaMKII inhibition with Ala or knock-down with RNAi in motoneurons decreased active zones (S5A and S5B Fig and Fig 9D, S1 Table) and over-expression of activated CaMKII in motoneurons increased active zones (S5A and S5C Fig, S1 Table). Remarkably, pre-synaptic over-expression in motoneurons of Ala or CaMKII-RNAi together with kek-6, rescued the increase in active zones caused by kek-6 over-expression (Fig 9C and 9D, S1 Table). These showed that CaMKII is required downstream of Kek-6 for active zone formation. Furthermore, in kek6–/–mutants and kek6–/–DNT2–/–double mutants the levels of pCaMKIIT287 decreased (Fig 10A, S1 Table), showing that Kek-6 is required for CaMKII activation at the NMJ. We found no significant effect of DNT2 loss or gain of function on pCaMKIIT287 levels at the NMJ. Importantly, over-expressing activated CaMKIIT287D pre-synaptically rescued the phenotypes of decreased axonal length and reduced Ib boutons of kek6–/–(Fig 10B) and DNT2–/–(Fig 10C) single mutants, and kek6–/–DNT2–/–double mutants (Fig 10D, S1 Table). This means that the mutant phenotypes were caused, at least partly, by decreased CaMKII activation. Together, these data demonstrate that Kek-6 and DNT2 function in concert upstream of CaMKII to regulate NMJ size and active zones.
To test the functional link between Kek-6, DNT2 and VAP33A, we used genetic epistasis. Loss of function VAP33AG0231 mutants had fewer Ib boutons (Fig 11A and 11B), and pre-synaptic over-expression of VAP33A increased Ib bouton number (Fig 11A, 11D and 11E, S1 Table), as previously reported[52]. These phenotypes were also shared with alterations in kek-6 and DNT2 levels, consistent with common functions. Importantly, neuronal over-expression of VAP33A rescued bouton number in kek-6–/–(Fig 11A and 11C) and DNT2–/–(Fig 11A and 11D) single mutants (S1 Table). Thus, VAP33A functions downstream of Kek-6 and DNT2. Interestingly, kek-6–/–DNT2–/–double mutants rescued the increase in Ib boutons caused by VAP33A over-expression, restoring bouton number down to control levels (Fig 11A and 11E, S1 Table). This suggests that VAP33A may be required both for pre-synaptic vesicle release downstream of Kek-6, and post-synaptic secretion of DNT2.
To conclude, altogether these data show that CaMKII and VAP33A function downstream of DNT2 and Kek-6 in motoneurons.
Drosophila homologues of the Trk receptor family had long been sought. This was important to find fundamental principles linking structure and function in any brain. Here, we show that Keks are Trk-family receptors lacking a TyrK, for Drosophila neurotrophin ligands (DNTs) (Fig 12A). Kek-2, -5 and -6 are expressed in the CNS, and genetic rescues, a signaling assay and co-immunoprecipitations demonstrate that they bind DNTs promiscuously. For in vivo analyses we focused on Kek-6. We demonstrate that motoneurons express kek-6 pre-synaptically, and Kek-6 binds DNT2, which is produced by the muscle from post-synaptic boutons (Fig 12B). kek-6–/–and DNT2–/–share mutant phenotypes, and using genetics we demonstrate they are functionally linked in vivo. Kek-6 can interact physically with the alternative DNT2 receptor, Toll-6. Most likely Kek-6 and Toll-6 function as a DNT2 receptor complex (Fig 12C). We show that: (1) The concerted functions of Toll-6 and Kek-6 are required to regulate NMJ growth. All single mutants–kek-6–/–, Toll-6–/–and DNT2–/––decreased NMJ size and complexity (Fig 12D and 12E), but only over-expression of DNT2 could induce a dramatic increase in NMJ size (Fig 12F). This shows that the compound activation of both receptors by DNT2 has a stronger effect on NMJ growth that activating each alone (Fig 12F). (2) Kek-6 and Toll-6 can each regulate active zone formation via alternative pathways, as Kek-6 can promote active zone formation independently of Toll-6 (Fig 12C). The ability of both Kek-6 and Toll-6 to regulate active zones enables homeostatic compensation in single mutants (Fig 12D). We show that Kek-6 functions by recruiting CaMKII and VAP33A to a pre-snaptic downstream complex and induces CaMKII activation. Epistasis analysis demonstrated that Kek-6 functions in motoneurons downstream of DNT2 and upstream of CaMKII and VAP33A to promote both NMJ growth, and active zone formation (Fig 12C). We conclude that at the NMJ, Kek-6 is a pre-synaptic receptor for DNT2, regulating structural synaptic plasticity (Fig 12A–12C).
Despite abundant evidence that retrograde signals and positive feedback loops regulate structural synaptic plasticity at the Drosophila glutamatergic NMJ, the identification of retrograde factors had been scarce [45,56]. This created a void for the general understanding of structural synaptic plasticity across different brain types. BDNF is a key retrograde ligand in mammals, and together with its receptor TrkB they form a positive feedback loop that reinforces synaptic function [1,4]. We have shown that the neurotrophin DNT2 is a retrograde ligand for Kek-6 (Fig 12B). DNT2 is expressed in muscle and both its receptors kek-6 and Toll-6 are expressed in motoneurons. Over-expression of DNT2 in muscle leads to its distribution in pre-synaptic boutons; over-expression of DNT2 in muscle induces NMJ growth and active zone formation pre-synaptically; DNT2 mutants have smaller NMJs; pre-synaptic over-expression of either kek-6, CaMKII or VAP33A can rescue the DNT2 loss of function phenotype; and conversely, kek6 loss of function rescues the phenotype caused by DNT2 over-expression in muscle. Thus, DNT2 is an evolutionarily conserved retrograde factor that regulates structural synaptic plasticity in Drosophila. This finding will have an impact bridging the universal understanding of synaptic plasticity and potentiation.
We demonstrate that keks are the Drosophila Trk-family homologues (Fig 12A). Previous searches for Trk-family receptors in fruitflies had focused on the Tryrosine Kinase domain, and did not identify any bona fide candidates [14–16,18]. Subsequent proteomic analyses confirmed the absence of full-length Trk receptors with a conserved tyrosine kinase in fruitflies [19–21]. Since those earlier searches, Trks in mammals have been found to encode multiple isoforms, most of which lack the tyrosine kinase domain [6,57]. Human TrkB has 100 isoforms, which produce 36 different proteins, of which four are abundant. Importantly, the most abundant isoform in the adult human brain is truncated TrkB-T1, which lacks the tyrosine kinase [6,57]. Paradoxically, the function of TrkB-T1 in neurons in the mammalian brain is unknown. Our findings suggest that truncated Trk family receptors could have an evolutionarily conserved function regulating structural synaptic plasticity.
We demonstrate that Kek-6 can activate CaMKII downstream (Fig 12B and 12C). However, we do not yet know how this may come about. CaMKII activation depends on the intracellular increase in Ca2+ levels, either through Ca2+ channels, or Ca2+ stores through the PLC and IP3 pathways [50,55]. In mammals, TrkB-T1 regulates Ca2+ levels in glia and in the heart, two contexts expressing high levels of TrkB-T1 and no TrkB-FL [58–60]. However, the mechanism by which TrkB-T1 raises Ca2+ is unknown [58,59]. Furthermore, whether this results in the activation of CaMKII is unclear or may be context dependent [58,60]. CaMKII activation in neurons could depend on the increase in Ca2+ with neuronal activity. A potential link of Kek-6 to neuronal activity was revealed by the increase in ghost boutons upon alterations in kek-6 levels (S4 Fig). Furthermore, kek-2 expression is modulated by neuronal activity, appropriate Kek-2 levels at the NMJ are required for normal synaptic structure, and DNT2 can modulate the Na+/K+ ATPase [27,61]. Thus, Kek-6 could function via a PLCβ pathway or by influencing membrane channels, to increase Ca2+ levels and induce CaMKII activation.
The finding that Trk-like receptors can regulate neuronal plasticity independently of kinase signaling is very important. Activation of PLCγ by TrkB TyrK signaling is generally seen as the key mechanism of Trk-depedent plasticity [1,5]. Yet CaMKII is necessary and sufficient for synaptic structural plasticity, LTP and long-term memory, and has wide functions in synaptic organization and homeostasis [49–51]. CaMKII can function as a frequency detector, and cause long lasting changes in synaptic strength, structure and brain plasticity [50]. Thus, the regulation of CaMKII and Ca2+ by truncated Trk-family receptors and Keks means that they could regulate structural synaptic plasticity independently of the canonical TyrK-dependent PLCγ pathway.
Keks are not identical to truncated Trk receptors and may carry out further functions that could be implemented via other routes in mammals. TrkB-T1 has a very short intracellular domain, whereas Kek-2, -5 and -6 have longer intracellular fragments that include a PSD95/Dlg/ZO1 (PDZ) motif [22,24,25]. PDZ domains are involved in scaffolding, and assembly of post-synaptic complexes that regulate the size and strength of synapses [62]. Through its PDZ domain, Kek-6 might recruit synaptic partners and/or Ca2+ channels. Interestingly, truncated TrkC-T1 binds the PDZ-containing protein Tamalin to induce cellular protrusions via Rac[63]. Exploring further signaling mechanisms of Keks vs. truncated Trks could uncover novel mechanisms of synaptic plasticity, perhaps also in the human brain.
LIG proteins and truncated Trk isoforms can function as ‘dominant negative’ co-receptors or ‘ligand sinks’. TrkB-T1 can form dimers with TrkB-FL and abrogate signaling, and bind BDNF rendering it unavailable to TrkB-FL [6,7]. This is thought to modulate kinase-signaling levels by TrkB [64]. Similarly, in Drosophila, Kek-5 is an inhibitor of BMP signaling [26] and Kek-1 an inhibitor of EGFR signaling [65]. Thus, each Kek could function as a specific inhibitor of different signaling pathways. By contrast, our evidence shows that Kek-6 does not function as an inhibitor of Toll-6. Instead, it indicates that Kek-6 and Toll-6 function in concert to regulate NMJ structure and growth (Fig 12C). Firstly, kek-6–/–, Toll-6–/–and DNT2–/–single mutants, and kek-6–/–Toll-6–/–double mutants, all have smaller NMJs, ruling out antagonistic functions between the two receptor types (Fig 12D and 12E). Secondly, whereas over-expression of kek-6 or Toll-6 alone did not increase NMJ size, over-expression of DNT2 did, showing that the activation of both receptor types at once has a stronger effect than each alone (Fig 12F). This also rules out an inhibitory function. Thirdly, Toll-6 is required for active zone formation, and over-expression of either Toll-6 or kek-6 increases active zones (Fig 12C–12F). This implies that Kek-6 does not inhibit Toll-6. Instead, DNT2 can modulate the homeostatic compensation of active zones, via its two receptors (Fig 12D and 12E). Indeed, compensation fails in kek6–/–DNT2–/–and kek-6–/–Toll-6–/–double mutants (Fig 12D and 12E). The data suggest that Kek-6 and Toll-6 influence synaptic structure via alternative, independent pathways. Toll-6 generally functions in neurons via NFκB and JNK [36], and at the NMJ Toll-6 and Toll-8 regulate structural synaptic plasticity through Sarm, JNK and dFOXO [37,43]. Here, we show that instead, Kek-6 functions via CaMKII and VAP33A to regulate active zones and synaptic structure (Fig 12B and 12C). Kek-6 functions independently of Toll-6, as it can induce active zone formation in Toll-6 mutants; the induction of active zones downstream of Kek-6 depends on CaMKII; and there is no evidence that Toll-6 can activate CaMKII on its own. Furthermore, Kek-6 also affects other synaptic factors, like Synapsin. We have shown that Kek-6 and Toll-6 can interact physically, thus could form a receptor complex for DNT2. We conclude that DNT2 regulates NMJ size and active zone formation, by engaging a receptor complex that drives distinct parallel pathways downstream (Fig 12C).
An intriguing observation is that DNTs can bind various Tolls [38] and Keks (this work) promiscuously, and similarly multiple Tolls can bind Kek-6. The biological significance of such promiscuity is not understood. Whether this reflects redundancy will depend on the spatial, cell-type specific and temporal distribution of each ligand and receptor. Redundancy can serve to compensate for deficits, thus conferring robustness to the CNS and resulting behaviour. For instance, homeostatic compensation of active zones vs. NMJ size, which was also observed in Toll-8 mutants[37]. Either way, our findings show that DNTs, Keks and Tolls working together constitute a novel molecular mechanism for synaptic structural plasticity and homeostasis. Whether this could be mirrored by the mammalian NTs, Truncated Trks and TLRs, will be intriguing and important to find out.
To conclude, evolution may have resolved how to implement structural synaptic plasticity through distinct mechanisms in fruit-flies and humans, or perhaps common molecular principles are shared by both, even if not in every detail. We found that in fruitflies, truncated-Trk-like receptors encoded by the Keks bind neurotrophin ligands to regulate structural synaptic plasticity via CaMKII and VAP33A, and the receptor complex also includes Tolls. It is compelling to consider whether such a non-canonical mechanism of neuronal plasticity, downstream of neurotrophins and Trk receptors but independently of kinase signaling, may also operate in humans, as it could uncover novel mechanisms of brain function and brain disease.
Mutants: Df(3R)ED6361 lacks the kek-6 locus (Kyoto Stock Centre), Df(3L)6092, Df(3L)ED4342, DNT141 and DNT2e03444 are described in[35], Df(3L)BSC578 lacks the Toll-6 locus (Bloomington). Mutant null alleles kek634 and kek635, and DNT237, were generated by FRT mediated recombination of the PiggyBac insertion lines [46]: for Kek6: PBac[RB]kek6e000907 and PBac[WH]kek6f05733; for DNT2: PBac[RB]spz5e03444 and PBac[WH]Shabf05893.. Mutants were selected using genetics and PCR, using primers as recommended in [46]. kek634 is a 41.9 kb deletion that just removes the coding region for kek6; DNT237 is a 27 kb deletion that removes the ATG and first exon, most likely resulting in no protein production [36]. VAP33AG0231 is semi-lethal P-lac insertion allele from Bloomington (w P{lacW]VAP33AG0231). Toll-6 mutants used have been described in [38]. GAL4 lines: kek5GAL4 line P[GawB]NP5933 (from BSC); w;;elavGAL4 drives expression in all neurons (insertion on the third chromosome); Toll-6-GAL4 was generated by RMCE from MIMICToll-6MIO2127 and kek-6-GAL4 from MIMICkek-6MI13953; w;;D42GAL4 and w;Toll-7GAL4 drive expression in motoneurons and have been previously described[38]; w; MhcRFP MhcGAL4 drives expression in muscle (Bloomington); w; GMRGAL4 drives expression in retina. UAS lines: to drive expression of each of the keks UAS-lines were made as described below. w;;UAS-DNT2-FL expresses full length untagged DNT2; w;;UASDNT2-CK expresses the mature form of DNT2, i.e. signal peptide+cystine-knot domains; w; UAS-DNT2-FL-GFP expresses full-length DNT2 tagged at the COOH end, and were made as described below. w; 10xUASmyr-Td-Tomato is membrane tethered (gift of B. Pfeiffer); w; UASFlyBow (gift of I.Salecker); w;UASCaMKIIT287D expresses constitutively active CaMKII and w;UASAla expresses the CaMKII inhibitor (gifts of J. Hodge); UASCaMKII-RNAi: y sc v; P{y[+ v+, TRiP.GL00237}attP2/TM3, Sb (BSC); y[1] w[*]; P{w[+mC] = UAS-FLAG.Vap-33-1.HA}2 (Bloomington). Double mutant lines were generated by conventional genetics. MIMIC-GFP lines: y w; MIMICkek6MI13953 and has a GFP insertion in the coding region (BSC). All stocks were balanced over SM6aTM6B or TM6B to identify balancer chromosomes, and all were generated from a yw or w mutant background. In all figures, controls are: (1) F1 from yw x Oregon; (2) y w; GAL4/+
Full length cDNAs for kek1, 2, 3, 4, 5 and 6 were obtained either from cDNA clones (kek1, SD01674; kek2, NB7), by PCR from cDNA libraries (gift of G. Tear; LD: kek5; GH: kek3; kek6), or by reverse transcription PCR from larvae (kek4), using primers designed for Gateway cloning into the pDONR plasmid: kek1 forward: GGGGACAAGTTTGTAC AAA AAA GCA GGC TCA TCC AGG AAA ATG CAT ATC A and reverse: GGGGACCAC TTT GTA CAA GAA AGC TGG GTA GTC AGT TCT TGG TTT GGT TT; kek2: forward: GGGGACAAGTTTGTAC AAA AAA GCA GGC TCA ATG AGT GGT CTG CCA ATC T and reverse: GGGGACCAC TTT GTA CAA GAA AGC TGG GTA AAT GTC GCT GGT TTC CTG GC; kek3: forward: GGGGACAAGTTTGTAC AAA AAA GCA GGC TCA TAT GCG ATG GCA GCG GGA A and reverse: GGGGACCAC TTT GTA CAA GAA AGC TGG GTA GCT CTT GAA AAT ATC CTG TC; kek4: forward: GGGGACAAGTTTGTAC AAA AAA GCA GGC TCA CTA GAC CTT CCG TTC CTT, and reverse: GGGGACCAC TTT GTA CAA GAA AGC TGG GTA TAT TGA GAT ATC AAC ACC AG; kek5: forward: GGGGACAAGTTTGTAC AAA AAA GCA GGC TAG CTA GAC GCA GAC TTA GAG and reverse: GGGGACCAC TTT GTA CAA GAA AGC TGG GTA GAC CTC GGT GCC ATC CTC GC; kek6: forward: GGGGACAAGTTTGTAC AAA AAA GCA GGC TCA ATG CAT CGC AGC ATG GAT C and reverse: GGGGACCAC TTT GTA CAA GAA AGC TGG GTA GAG CGA CAC GAA CTC GCC AG. CaMKII and VAP33A full-length cDNAs were cloned using the Gateway System first into pDONR, and then into pAct5-attR-HA.
For expression in flies under UAS/GAL4 control, cDNAs were subcloned to a Gateway pUASt-attR-mRFP destination vector for conventional transgenesis, injected by BestGene (www.thebestgene.com). For expression in S2 cells, they were subcloned into tagged pAct destination vectors pAct5c-attR-mCFP, pAct5c-attR-FLAG and pAct5c-attR-HA. Constructs used for expression of S2 cells of DNT1 and 2, Toll-6 and 7 were as previously described[38]. Chimaeric Kek-Toll-6 receptors were generated after analyzing the domain composition of the proteins using ProSite (ExPASy), PFAM, SMART, TMHMM and TMPred algorithms and PubMed data. Primers were designed to amplify the sequences that encode the extracellular and transmembrane domains of Kek3–6, including 15 amino acids C-terminal to the Kek transmembrane region, and the intracellular domain of Toll6, at halfway between the transmembrane region and the TIR domain. A unique enzyme site (BamHI or EcoRI) was included in the designed primers to join the kek and Toll-6 sequences, and attB sites were introduced in the primers at the 5′ and 3′ ends of the chimaeric insert for Gateway cloning into pAct5c-attR-3xHA destination vector. Unfortunately, we were not able to generate kek1-Toll-6 and kek2-Toll-6 chimaeric receptor constructs. To clone chimaeric kek3,4,5,6-Toll-6 receptor constructs, reverse primers at the juxtamembrane of keks were as follows: kek3-Toll-6: CGAT-GAATTC-AGGTACAGAGTTCCAGAGAC; kek4-Toll-6: CGCG-GAATTC-TTGCAAATAAGTGTGCTGGC; kek5-Toll-6: CTAT-GGATCC-GCTCATCATGGTGGTGTCCT; kek6-Toll-6: GTAT-GAATTC-ACGCCGGCCTTGTTGGCATG. Toll-6 primers from chimaeras were: Forward primers from juxtamembraneCATG-GAATTC-AACTTCTGCTACAAGTCACC (compatible with kek3, kek4 and kek6 chimaeras) or CATG-GGATCC-AACTTCTGCTACAAGTCACC (compatible with kek5 chimaera) and reverse from C terminus: GGGGACCAC TTT GTA CAA GAA AGC TGG GTC CGC CCA CAG GTT CTT CTG CT.
Un-tagged full-length DNT2 was cloned into the pUASt-attB vector by conventional cloning, DNT2-full length (DNT2-FL) was PCR-amplified from cDNA libraries, and cloned into pUASt-attB for ΦC31 transgenesis [66]. UAS-DNT2-FL-GFP was cloned by Gateway cloning into pUAS-GW-GFP, followed by conventional transgenesis.
Reverse Transcription PCR was carried out following standard procedures after mRNA straction from wandering larvae, using the following primers: (1) Control: GAPDH: GAPDH Fw TCACCACCATTGACAAGGC; GAPDH Rev:; CGGTAAGATCCACAACGGAG; (2) DNT2: DNT2 0.5kDa Rwd TGGAAACCCGCTCTTTGTCAG; DNT2 0.9kDa Rev:.TGATGAACTGCGATGTCGTCT. The genotypes of larvae tested were y w control, and transheterozygous mutants DNT237/Df(3L)6092.
For Dif signaling assays using chimaeric kek-Toll-6 receptors, S2 cells stably transfected with drosomycin-luciferase were maintained at 27°C in air in Insect-Xpress medium (Lonza) supplemented with penicillin/streptomycin/L-glutamine mix (Lonza) and 10% foetal bovine serum (Lonza). 1ml suspended cells were passaged every two days into 4ml fresh medium. 3x106 cells in 2ml media were seeded per well of a 6-well plate 24 hours prior to transfection. Per well of experiment, 250μl serum-free media, 3μl TransIT-2020 (Mirus), 2μg of HA tagged chimaeric receptors kek-toll6-HA plus 1μg of pAct-renilla-luciferase were incubated at room temperature for 30 minutes, supplemented with 350μl serum-free media and added to aspirated cells. After 4 hours, transfection mixture was removed and 2ml supplemented medium added. All experiments were conducted 48 hours after transfection; for imaging membrane targeting of Kek-Toll6-HA protein S2 cells were starved for 6h prior to fixation.
For signaling assays, S2 cells were stimulated with 50nM/well of purified baculovirus DNT2 protein (generated as previously described[38]) and Dif signaling quantified by luminescence. DNT2 was added 48 hours after transfection and luminescence quantified 24 hours after DNT stimulation. Transfected and stimulated cells were pelleted from single wells, resuspended in 400μl media and separated into three 50μl aliquots in an opaque 96-well plate. 40μl of Firefly Luciferase Substrate (Dual-Glo Luciferase Assay System; Promega) was added per 50μl aliquot, incubated for 10 minutes, and luminescence measured using a Mithras LB 940 Multimode Microplate Reader (Berthold). 40μl Stop & Glo substrate (Dual-Glo Luciferase Assay System; Promega) was added to quench Dif signal and activate Renilla Luciferase. Renilla luminescence data was used to normalize Firefly Luciferase data.
Coimmunoprecipitations were carried out as previously described[38], after transfecting standard S2 cells with the following constructs: (1) Co-IP Kek6-DNT2: pAct5C-Pro-TEV6HisV5-DNT2-CK and pAct5C-Kek6-3xHA; (2) Co-IP Kek6-DNT2 or Kek6-DNT1: pAct5C-Kek6-3xHA, pAct5C-Pro-TEV6HisV5-DNT1-CK-CTD, pAct5C-Pro-TEV6HisV5-DNT2-CK; (3) Co-IP Kek5-DNT2: pAct5C-Kek5-3xFlag and pAct5C-Pro-TEV6HisV5-DNT2-CK. (4) Co-IP Kek2-DNT2: pAct5C-Kek2-3xFlag and pAct5C-DNT2-FL-HA. (5) Co-IP Kek3-DNT2: pAct5C-Kek3-3xFlag and pAct5C-DNT2-FL-HA. (5) Co-IP kek6-Toll-6: pAct5C-Kek6-HA and pAct5C-Toll-6CY-3xFlag, pAct5C-Toll-7CY-3xFlag.(6) Co-IP Kek6-CaMKII: pAct5-CaMKII-HA and pAct-kek-6-FLAG. (7) Co-IP kek6-VAP33A: pAct5-Vap33A-HA and pAct-kek-6-FLAG. 48h after transfection cells were harvested and washed in PBS. Final cell pellets were lysed in 600 μL NP-40 buffer (50 mM Tris-HCl pH:8.0, 150 mM NaCl, 1% Igepal-630) supplemented with protease inhibitor cocktail (Pierce). For V5 and HA immuno-precipitations,500 μL of lysates from single or co-transfected cells were incubated with 1 μg of mouse anti-V5 or 1 μg of mouse anti-HA antibodies overnight at 4°C, then the lysate plus antibody mixtures were supplemented with 25 μL of protein-A/G megnetic beads (Pierce) and incubated for 1 h at room temperature. For Flag immunopreciptations, lysates were incubated with anti-Flag antibody conjugated magnetic beads (Sigma-Aldrich) overnight at 4°C or for 2h at room temperature. Beads were washed thoroughly in NP-40 buffer and/or PBS, or in TBST buffer. Proteins were eluted in 40 μL of 2x Laemmli-buffer and analysed by Western blot following standard procedures.
S2 cells were transfected with 4 μg pAct5C-Kek6-3xFlag expression construct. For controls mock-transfected cells (i.e. no construct) were used. After transfection cells were processed as described above for co-immunoprecipitation. 600 μL of cell lysates were incubated with 20 μL of anti-Flag conjugated magnetic beads overnight at 4°C. Beads were washed thoroughly in PBS, then proteins were eluted in 40 μL 2x Laemmli-buffer for 10 min at room temperature. Eluted proteins were loaded onto 10% polyacryalamide gels. Alternatively, proteins were not eluted after overnight incubation with anti-Flag beads, but they were re-incubated with whole fly (OregonR) head lysates. Here, 60 heads were lysed in 600 μL of NP-40 buffer. After overnight incubation at 4°C, proteins were eluted and analysed as for S2 cell lysate proteins. Thus, eluted proteins were loaded onto 10% polyacryalamide gels, and the gels were stained with Coomassie Blue and cut into several pieces. Gel pieces were subjected to in-gel digestion with trypsin using a standard protocol. Tryptic peptides were analysed by LC-MS/MS using Ultimate 3000 HPLC coupled to a LTQ Orbitrap Velos ETD mass-spectrometer. Peptide separation, mass spectrometric analysis and database search were carried out as specified at the University of Birmingham Proteomics Facility. Candidate binding proteins were identified on these criteria: (1) Proteins were accepted only if they were identified with at least two high confident peptides. (2) Mock-transfected controls and Kek6-3xFlag samples were compared, and only proteins identified from Kek6-3xFlag lysates or heads, but absent form controls, were considered as possible interacting partners for Kek6.
Western blot was carried out following standard procedures using antibodies listed here: mouse anti-V5 (1:5000, Invitrogen), mouse anti-HA (1:2000, Roche), chicken anti-HA (1:2000, Aves), rabbit anti-Flag (1:2000, Sigma-Aldrich), HRP-conjugated anti-mouse IgG (1:5000–1:10000, Vector), HRP-conjugated anti-chicken IgG (1:5000, Jackson Immunoresearch), HRP-conjugated anti-rabbit IgG (1:1000–1:5000, Vector), mouse dCaMKII (Cosmo CAC-TNL-001-CAM 1:1000), rabbit α-p-CaMKIIα (Thr 286) (Santa Cruz sc-12886-R 1:1000), mouse Tubulin DM1A (Abcam ab7291 1:10000).
NMJ preparations were carried out according to [67]. For GFP stainings in MhcGAL4>UAS-DNT2-FL-GFP, L3 larvae were placed in agar plates (2%) and left at 29°C for 90 minutes to potentiate the NMJ before dissections [68]. A hundred and twenty female flies and 50 males were placed in a cage with a removable agar and grape juice plate. On the second day, plates were changed for new ones in the morning and evening, and discarded. On the third day, the plates were changed every 1h 30min and discarded. On the fourth day, the plates were changed every 1h 30min and kept in a 25°C incubator for larvae collection. The next day, hatched L1 larvae from each plate were transferred to a vial, and exactly 40 larvae were placed in each vial. When larvae reached L3 stage (5 days after egg laying), they were dissected in low calcium saline as previously described[67], and fixed for 10 minutes in Bouin’s solution (HT10132 SIGMA). Samples were washed 6 times for 10 minutes in PBT (0.1% of Triton in 1M PBS) to remove fixative solution and kept overnight in blocking solution (10% normal goat serum in 0.1% Triton in PBS 1M). Primary antibodies were incubated overnight at 4°C and samples were washed the following day 8 times for 10 minutes in PBT. Secondary antibodies were incubated for 2h at room temperature and samples were washed 8 times for 10 minutes in PBT. Samples were mounted in Vectashield anti-bleacing medium (Vector Labs) in No.1 coverslips. NMJs were analysed for muscle 6/7 only; segments A3 and A4, left and right, were analysed in each larva.
Antibody stainings in embryos and S2 cells were carried out following standard procedures, using the following primary antibodies at the indicated dilutions: mouse anti-FasII at 1:5 (ID4, Developmental Studies Hybridoma Bank, Iowa); rabbit anti-GFP 1:1000 (Molecular Probes); rabbit anti-GFP 1:4,000 (abcam ab290); mouse anti-Dlg 1:20 (4F3, Developmental Studies Hybridoma Bank, Iowa); rabbit anti-HRP 1:250 (Jackson Immunoresearch); mouse anti-Brp 1:100 (nc82, Developmental Studies Hybridoma Bank, Iowa); rabbit anti-p-CaMKIIT286 1:150 (Santa Cruz), raised against mammalian pCaMKIIT286 detects phosphorylation at Thr287 in Drosophila, the constitutively active form; mouse plain anti-Synapsin at 1:25 (DSHB 3C11). Secondary antibodies were: anti-guinea pig-Alexa 488 at 1:250 (Molecular Probes); biotynilated anti-mouse at 1:300 (Jackson Labs) followed by the ABC Elite Kit (Vector Labs); biotynilated anti-guinea pig at 1:300 (Jackson Labs) followed by Streptavidin-Alexa-488 at 1:400 (Molecular Probes); anti-rabbit-Alexa 488 at 1:250 (Molecular Probes); anti-mouse-Alexa 488 1:250 (Molecular Probes); anti-rabbit-Alexa 647 at 1:250 (Molecular Probes); anti-mouse-Alexa 647 1:250 (Molecular Probes).
In situ hybridizations were carried out following standard procedures, using antisense mRNA probes from 5’ linearised plasmids and transcribed as follows: lambik: a 551 nucleotides fragment was cloned into pDONR with primers: forward GGGGACAAGTTTGTAC AAA AAA GCA GGC TAG AAA CTA CGC ATG AGC CTG and reverse GGGGACCAC TTT GTA CAA GAA AGC TGG GTA CCG CTC AAA TGT CCA CTG T; then linearised with HpaI and transcribed with T7 RNA polymerase. CG15744: a 569 nucleotides fragment was cloned into pDONR with primers: forward GGGGACAAGTTTGTAC AAA AAA GCA GGC TGG ATT GGA TAG CCT TGG TGA and reverse GGGGACCAC TTT GTA CAA GAA AGC TGG GTT TCG CTT CCA TCT CCA TCT C (linearised with HpaI, transcribed with T7); CG16974: a 548 nucleotides fragment was cloned into pDONR with primers: forward GGGGACAAGTTTGTAC AAA AAA GCA GGC TTA TAT GAA TCC CGA AGG CGC and reverse GGGGACCAC TTT GTA CAA GAA AGC TGG GTT TGG GGG GAG TAG ATG GTA A (linearised with HPA1, transcribed with T7); kek1 (SD01674+pOT2 cDNA clone; linearised with EcoRI, transcribed with SP6 RNA polymerase); kek2 (NB7+pNB40; HindIII; T7); kek3 (HpaI; T7); kek4 (GH27420+pOT2; EcoRI; SP6); kek6 (in pDONR, HpaI; T7); CG15744 (in pDNOR, HpaI; T7); CG16974 (in pDNOR, HpaI; T7); lambik (in pDONR, HpaI; T7). Colorimetric reaction was using Alkaline Phosphatase conjugated anti-DIG conjugated.
Wide-field Nomarski optics images were taken with a Zeiss Axioplan microscope, 63x lens and JVC 3CCD camera and Image Grabber graffics card (Neotech) and a Zeiss AxioCam HRc camera and Zen software. Laser scanning confocal microscopy was carried out using inverted Leica SP2 AOBS, upright Leica SP8 or inverted Zeiss LSM710 laser scanning confocal microscopes, with 40x lens, 1024 x 1024 resolution, 0.25μm step for anti-Brp, and 0.5 μm step for the rest. Images were compiled using ImageJ, Adobe Photoshop and Illustrator.
For NMJ data, all images provided in the figures are projections from Z stacks; all the quantitative anlyses were carried out in the raw stacks, not in projections. Muscle surface area was measured from bright field images using ImageJ. Anti-HRP was used to measure total terminal axonal length using ImageJ, and branching points. Boutons were visualized with anti-Dlg, and total boutons as well as separately Is and Ib boutons, were counted manually with the aid of the ImageJ Cell Counter plug-in. Automatic quantification of anti-Brp and anti-pCaMKIIT286 were carried out in 3D throughout the stacks of images using the DeadEasy Synapse ImageJ plug-in that we previously validated [39], and anti-Synapsin was analysed with a slightly modified version.
DeadEasy Synapse has already been described and validated [39]. DeadEasy Synapse first reduces the Poisson noise, characteristic of confocal microscopy images, of each slice in the stack using a median filter. Subsequently, in order to separate signal (e.g. Brp+ active zones) from the background, images are segmented. Since the intensity of the staining varied within each image and from image to image, the maximum entropy threshold method [69] was used to find a local optimal threshold value for each pixel. For this, we used a square window of size 15x15, sufficient to find the local optimal threshold around each pixel in an image. In this way, each pixel is considered part of an active zone if the value of the pixel is higher than the local threshold, otherwise assigned to the background. Since this method is computationally expensive and very low intensity pixels correspond to background, it was possible to reduce the computation time by assigning pixels whose intensity was lower or equal than 20 directly as part of the background and only applying the thresholding method to pixels whose intensity was higher than 20. Finally the volume of the active zones is measured. This method worked just as accurately with anti-Brp, anti-pCaMKIIT286 and anti-Synapsin stainings. Data were normalized to muscle surface area or axonal length.
L3 wandering larvae were placed one at a time on an agar plate (2%) and left it crawl for 40 seconds. Larvae were filmed crawling across the agar plate and then discarded. Plates were cleared before placing another larva. At least 50 larvae were filmed per genotype, and the test was always done in the morning. Films of 400 frames per larva were analysed using FlyTracker software developed in our lab to obtain the trajectory and speed, as previously described[38].
Data were analysed in SPSS Statistics 21 (IBM) and GraphPad Prism 6. Confidence interval was 95% (p<0.05). Categorical data were tested using χ2, and a Bonferroni correction was applied for multiple comparisons. Continuous data were first tested for normality by determining the kurtosis and skewness, and a Levene’s Test was applied to test for homogeneity of variance. Data were considered not normally distributed if absolute kurtosis and skewness values for each genotype were greater than 1.96 x standard error of kurtosis/skewness. Variance of the populations of different samples were considered unequal if Levene’s test for homogeneity of variance gave a p value of <0.05. If samples were normally distributed and variances were equal, Student-t tests were applied for 2-sample type graphs, and One-Way ANOVA was used to compare means from >2 samples. When data were normally distributed, but variances were unequal, Welch ANOVA was used instead. Multiple comparison corrections to normal data were applied using a post-hoc Dunnett test of comparisons to a control, Games-Howell or Bonferroni post-hoc comparing all samples. Non-parametric continuous data were compared using a Mann-Whitney U-test when 2 sample types were being analysed, and a Kruskal-Wallis test for >2 samples, and multiple comparison corrections were applied using a post-hoc Dunn test. See S1 Table for all statistical sample sizes, applied tests and p values.
|
10.1371/journal.pntd.0005573 | Associations between selective attention and soil-transmitted helminth infections, socioeconomic status, and physical fitness in disadvantaged children in Port Elizabeth, South Africa: An observational study | Socioeconomically deprived children are at increased risk of ill-health associated with sedentary behavior, malnutrition, and helminth infection. The resulting reduced physical fitness, growth retardation, and impaired cognitive abilities may impede children’s capacity to pay attention. The present study examines how socioeconomic status (SES), parasitic worm infections, stunting, food insecurity, and physical fitness are associated with selective attention and academic achievement in school-aged children.
The study cohort included 835 children, aged 8–12 years, from eight primary schools in socioeconomically disadvantaged neighborhoods of Port Elizabeth, South Africa. The d2-test was utilized to assess selective attention. This is a paper and pencil letter-cancellation test consisting of randomly mixed letters d and p with one to four single and/or double quotation marks either over and/or under each letter. Children were invited to mark only the letters d that have double quotation marks. Cardiorespiratory fitness was assessed via the 20 m shuttle run test and muscle strength using the grip strength test. The Kato-Katz thick smear technique was employed to detect helminth eggs in stool samples. SES and food insecurity were determined with a pre-tested questionnaire, while end of year school results were used as an indicator of academic achievement.
Children infected with soil-transmitted helminths had lower selective attention, lower school grades (academic achievement scores), and lower grip strength (all p<0.05). In a multiple regression model, low selective attention was associated with soil-transmitted helminth infection (p<0.05) and low shuttle run performance (p<0.001), whereas higher academic achievement was observed in children without soil-transmitted helminth infection (p<0.001) and with higher shuttle run performance (p<0.05).
Soil-transmitted helminth infections and low physical fitness appear to hamper children’s capacity to pay attention and thereby impede their academic performance. Poor academic achievement will make it difficult for children to realize their full potential, perpetuating a vicious cycle of poverty and poor health.
ClinicalTrials.gov ISRCTN68411960
| Children growing up in challenging environments, such as townships in South Africa, are at an increased risk of ill-health associated with sedentary behavior, poor nutrition, growth retardation, and infections with parasitic worms. Negative factors such as limited educational resources, insufficient health care and safety are exacerbating the effects of poverty and, taken together, might cause developmental delays and school failure. A total of 835 school children aged 8–12 years were examined for soil-transmitted helminth infection, physical fitness, selective attention, stunting, household socioeconomic conditions, and food security. Furthermore, children’s academic achievement scores were utilized as a proxy for academic achievement. The multivariate analyses showed that low selective attention was associated with soil-transmitted helminth infection and low shuttle run performance, whereas higher academic achievement was observed in children without soil-transmitted helminth infection and with higher shuttle run performance. Our study suggests that soil-transmitted helminths and low physical fitness hinder children from realizing their full potential.
| Attention skills are relevant for academic foundations and are important for learning [1]. Selective attention is the ability to select and focus on a particular task, while simultaneously suppressing irrelevant or distracting information. Competing information can occur both externally and internally due to visual or auditory distractions or distracting thoughts [2]. Selective attention has been associated with important domains in education, such as language processing [3], literacy [4], and numeracy [5], and hence, plays an important role in academic achievement. A growing body of literature documents that children from low-income households exhibit more attention deficits compared to their higher-income peers [6,7]. Of note, academic achievement depends on multiple factors such as educational opportunity, socioeconomic status (SES), health and nutritional status, family environment [8,9], social competence [10], cognitive skills, and the ability to pay attention [2].
Children growing up in socioeconomically deprived environments face multiple challenges. Essential services, such as health care, sanitation, physical security, electricity, and high quality academic and physical education are often lacking, with serious consequences for children’s psychological and physiological development and wellbeing [11]. Poverty also limits the parents’ ability to provide a responsive, supportive, and safe learning environment [12], and lessens the probability that children will have access to cognitively stimulating materials (e.g., books and toys) [13]. Families with low income often invest most of their resources into covering their basic needs, such as food and housing, and have therefore limited means to invest in the future of their children [14]. Poverty also puts children at risk of chronic malnutrition [14]. Chronic malnutrition causes stunting and has been found to be associated with poor cognitive development resulting in low IQ, and problems with motor development [15]. This, in turn, can impede children’s ability to concentrate, process information, and focus on academic work [16].
Poor living conditions with a lack of clean water, inadequate sanitation, and insufficient hygiene also favor parasitic worm and intestinal protozoa infections [17,18], which may lead to symptoms such as abdominal pain, diarrhea, anemia, growth retardation, reduced physical fitness, cognitive impairment, and poor academic achievement [19,20]. Recent systematic reviews suggest associations between parasitic worm infection and children’s cognitive function and academic performance, but positive effects of mass treatment on cognition or school performance remain elusive [21–24]. A study by Ezeamama et al. [25] found that roundworm (Ascaris lumbricoides) infection was associated with poor performance on tests of memory, whereas whipworm (Trichuris trichiura) infection was associated with poor performance on tests of verbal fluency among Filipino children. To our knowledge, there is a paucity of studies investigating whether soil-transmitted helminth infections are associated with selective attention.
Children from families with low SES are also less likely to have access to health care or health insurance, resulting in a greater risk of illness and school absenteeism and consequently a lack of academic input compared to better-off peers [16]. Recent reviews and meta-analyses have shown that physical activity elicits short- and long-term benefits for children’s executive function [26], attention [27], and other academic outcomes [28]. Yet, physical activity levels are often low among poor children and youth, also in South Africa [29]. For instance, a study by Walter [30], which focused on primary school children in disadvantaged schools, observed that most children do not achieve the recommended 60 min of daily moderate-to-vigorous physical activity (MVPA). These results are not surprising given that sport and recreation facilities are often inadequate, inaccessible, or in poor condition, while qualified teachers are scarce and physical education and extramural sport programs are rare [31].
The purpose of the present study was to find out how children’s selective attention and academic achievement relate to age, sex, SES, helminth infection status, stunting, food security, and physical fitness. In a first step, we looked at bivariate associations and compared children with or without helminth infection, and stunted or non-stunted children. In a second step, we examined multivariate associations to find out how age, sex, SES, helminth infection status, stunting, food security, and physical fitness relate to selective attention and academic achievement if all these variables are considered simultaneously.
The “Disease, Activity and School children’s Health” (DASH) cohort study was approved by the ethical review board of Northwestern and Central Switzerland (EKNZ; reference no. 2014–179, approval date: 17 June 2014), the Nelson Mandela Metropolitan University (NMMU) Human Ethics Committee (study number H14-HEA-HMS002, approval date: 4 July 2014), the Eastern Cape Department of Education (approval date: 3 August 2014), and the Eastern Cape Department of Health (approval date: 7 November 2014) in Port Elizabeth, South Africa. The study is registered at ISRCTN registry under controlled-trials.com (unique identifier: ISRCTN68411960, registration date: 1 October 2014).
Details regarding the information of potential study participants, exclusions due to medical reasons, management of helminth infections, and referrals, are provided in a previously published study protocol [32]. In brief, oral assent from each participating child was sought and individual written informed consent was obtained from parents/guardians. Participation was voluntary and children could withdraw from the study at any time without further obligations. Children were eligible for this study if they met the following inclusion criteria: (i) are willing to participate in the study; (ii) have a written informed consent by a parent/guardian; (iii) are not participating in other clinical trials during the study period; and (iv) do not suffer from medical conditions, which will prevent participation in the study, as determined by qualified medical personnel. To ensure confidentiality, each study participant was given a unique identification number. All tests were available in English, Xhosa, and Afrikaans. To ensure optimal translation of the tests, we collaborated with independent professional translators and followed the procedure set out by Brislin [33]. Thus, test instructions and items were translated from English into Xhosa and Afrikaans, and pilot-tested with a small sample of Xhosa and Afrikaans speaking students and school children of the same age as the study cohort. Schools were recruited from 2014 to 2015. Data assessment took place between February 2015 and March 2015.
The study involved 8- to 12-year-old children attending grade 4 from eight schools located in socioeconomically disadvantaged neighborhoods in Port Elizabeth, South Africa. South African public schools are classified into five groups, with quintile one standing for the poorest and quintile five for the least poor [34]. Study schools belonged to quintile three. The sample size calculation for the study was based on achieving sufficient precision in estimating the prevalence of soil-transmitted helminth infections, with a targeted sample size of approximately 1000 grade 4 school children (for more details regarding power calculation see Yap et al. [32]).
All children were asked to remove their shoes and jerseys/jackets before standing on a digital weighing scale (Micro T7E electronic platform scale, Optima Electronics; George, South Africa). Body weight was measured once and recorded to the nearest 0.1 kg. With the shoes removed, each child then stood against a Seca stadiometer (Surgical SA; Johannesburg, South Africa) with their back erect and shoulders relaxed. Body height was measured once and recorded to the nearest 0.1 cm.
Upper body strength was determined by the grip strength test [35]. The Saehan hydraulic hand dynamometer (MSD Europe BVBA; Tisselt, Belgium) was employed. The field investigator demonstrated how to hold the hand dynamometer and instructed the child to sit relaxed, spine erect, and arm position at a 90° angle. Each child had six trials, alternating between the right and left hand with a 30 sec resting period between trials, griping the hand dynamometer as hard as possible. All six trials were recorded to the nearest 1 kg and averaged.
To measure children’s aerobic fitness, the 20 m shuttle run test was utilized, following the test protocol described by Léger et al. [36]. A premeasured running course was laid out on a flat terrain and marked with color-coded cones. Children who felt sick or voiced discomfort were excluded. The test procedures were explained and a researcher demonstrated two trial runs. Once children were familiar with the test procedures, they started with a running speed of 8.5 km/h, following a researcher who set the pace according to the acoustic signal. The frequency of the sound signal gradually increased every min by 0.5 km/h. If a child was unable to cross the marked 2 m line before each end of the course at the moment of the sound signal for two successive intervals, the individual maximum was reached. Children were then asked to stop running and the fully completed laps were noted.
To estimate children’s SES, they were asked to answer nine items, covering household-level living standards, such as infrastructure and housing characteristics (house type, number of bedrooms, type of toilet and access to indoor water, indoor toilet/bathroom, and electricity) and questions related to ownership of three durable assets (presence of a working refrigerator, washing machine, and car). The dichotomized items (0 = poor quality, not available; 1 = higher quality, available) were summed up to build an overall SES index, with higher scores reflecting higher SES. The validity of similar measures has been established in previous research [37].
Food insecurity was measured with four questions about hunger, portion size, and meal frequency (e.g., “did you go to bed hungry last night?”). The items were adapted from the Household Hunger Scale [38]. Response options were summed up to obtain a score for each participant ranging from 0 (food insecure/hungry) to 4 (food secure/not hungry). This score was used to obtain an overall index of food security, with higher scores reflecting higher food security.
To diagnose helminth infections, stool containers with unique identifiers were handed out to school children with the instruction to return them with a small portion of their own morning stool. The diagnostic work-up was done on the same day. Duplicate 41.7 mg Kato-Katz thick smears were prepared from each stool sample [39]. Slides were independently read under a microscope by experienced laboratory technicians who counted the number of helminth eggs and recorded them for each species separately. For quality control, a random sample of 10% of all Kato-Katz thick smears was re-examined by a senior technician. In case of discordant results, the slides were re-read a third time and results discussed among the technicians until agreement was reached. Soil-transmitted helminth egg counts were multiplied by a factor of 24 to obtain a proxy for helminth infection intensity, as expressed by the number of eggs per 1 g of stool (EPG) [40]. Subsequently, a single 400 mg oral dose of albendazole (INRESA; Bartenheim, France) was administered to all children participating in the study, according to WHO and national treatment guidelines. Otherwise, to our knowledge, no further helminthiasis control interventions took place in recent years in the study community where the cohort group stems from.
Children’s selective attention capacity was measured with the d2 attention test, developed by Brickenkamp et al. [41]. The d2 test determines the capacity to focus on one stimulus/fact, while suppressing awareness to competing distractors. The d2 attention test is a paper and pencil letter-cancellation test that consists of 14 lines of 47 randomly mixed letters d and p. Participants were instructed to identify and mark all d letters with two dashes arranged either as single dashes (i.e., one above and one below the ds), or in pairs above or below the ds. After 20 sec, the researcher signaled to continue on the next line. Altogether, the test lasted 4 min and 40 sec. The test was performed in groups of 20–25 students and conducted during the first school lesson in a quiet room, with an average room temperature of 24°C. Pencils were distributed and the test procedure was explained to the children in their native language. Additionally, a practice line was provided on the blackboard to ensure that all participants understood the test procedures. Furthermore, children were encouraged to practice on the test line prior to launching the test.
As shown in Table 1, several different parameters can be calculated after completion of the d2 test. For instance, the total number of items processed is a measure of processing speed (TN), while the number of all errors relative to the total number of items processed is a measure of precision and thoroughness, referred to as accuracy in the present text (E%). By contrast, the number of correctly marked characters minus the number of incorrectly marked characters is a measure of concentration ability and performance (CP). E% and CP are not inflated by excessive skipping as they are based on the number of target and non-target characters cancelled as opposed to processing speed which can be influenced by test strategies [42]. In our study, we therefore used E% and CP as dependent variables due to their resistance to falsification. Processes of selective attention are required for successful completion since not only the letter d is orthographically similar to the letter p, but there are many distracting letters with more or less than two dashes [41].
As an indicator of academic achievement, we collected from each school the children’s end-of-year results which are based on the mean of four subjects: (i) home language (Xhosa or Afrikaans, in this case); (ii) first additional language (English, in this case); (iii) mathematics; and (iv) life skills. Learner achievement in each subject is graded on a scale of 1 to 7, whereby a rating of 1 (0–29%) indicates “not achieved” and one of 7 (80–100%) indicates “outstanding achievement”. A rating of 4 (50–59%) indicates “adequate achievement”.
Data were double-entered, validated using EpiData version 3.1 (EpiData Association; Odense, Denmark), and merged into a single database. Statistical analyses were performed with SPSS version 23 (IBM Corporation; Armonk, United States of America) for Windows and STATA version 13.0 (STATA; College Station, United States of America). Anthropometric indicators and fitness performance scores were expressed as means (M) and standard deviations (SD). To describe the anthropometry of the children, body weight and height values were utilized to calculate the body mass index (BMI), defined as weight (in kg)/height2 (in m2). BMI-for-age and height-for-age (HAZ) were thus available for every participant [43]. The BMI and height-for-age z-scores (HAZ) were calculated using the World Health Organization (WHO) growth reference [43]. The sex-adjusted HAZ z-scores were used as an indicator for stunting [44]. The level at which the child stopped running during the 20 m shuttle run test was used to calculate an estimate of maximal oxygen uptake (VO2 max), readily adjusted for age [36]. The parasitological status was expressed in terms of prevalence of helminth infection. Selective attention was expressed as raw values. Statistical significance was set at p<0.05.
In a first step, separate mixed linear and mixed logistic regression models with random intercepts for school classes were calculated to compare selective attention and physical fitness among (i) stunted and normally grown children; and (ii) soil-transmitted helminth infected and non-infected children. In a second step, SES, age, sex, soil-transmitted helminth infection status, stunting, food insecurity, grip strength, and VO2 max were analyzed simultaneously in multiple linear regression models, with random intercepts for school classes, in order to determine the simultaneous impact of these variables on selective attention and academic achievement. To interpret the findings, the following statistical coefficients were displayed: for mixed linear and mixed logistic regression models the means and 95% confidence interval (CI), and for multiple linear regression models the unstandardized B coefficients in combination with the 95% CI.
As shown in the participant flow chart diagram (Fig 1), after receiving written informed consent from a parent or legal guardian, a total of 1,009 students agreed to take part in the study. Data of 970 children were available for further analyses. Complete data records were available for 835 children; 61.8% (n = 516) were black African (mostly Xhosa speaking), while the remaining 38.2% (n = 319) were colored African (mostly Afrikaans speaking). All analyses presented in this article refer to this final cohort, including 417 girls (49.9%) and 418 boys (50.1%).
An overview of the descriptive statistics and sex differences for all study variables is provided in Table 2. Boys were, on average, slightly older than girls and had a lower BMI. Overall, 31.0% of the children were infected with T. trichiura and/or A. lumbricoides, yet no hookworm infections were found. Stunting was observed in 12.3% of the children and the mean food insecurity score was 3.1. No significant sex differences were identified for height, weight, helminth infection status, stunting, food insecurity, and SES. Stratification by age (Table 3) revealed that older children were significantly taller, heavier, and more stunted, and had a higher prevalence of helminth infection.
As shown in Table 2, boys achieved significantly higher mean grip strength and had a higher mean VO2 max estimate than girls. As shown in Table 3, older children (aged 10–12 years) achieved higher mean grip strength scores than their younger peers (8- to 9-years old). The stratification by age also revealed that the younger group (8–9 years) reached a higher estimated VO2 max than the older group (10–12 years).
As displayed in Table 2, stratification by sex revealed that girls and boys did not differ with regard to their selective attention capacity. Younger children had a significantly lower percentage of errors (see Table 3 for mean scores).
Girls reached statistically significantly higher academic scores than boys (Table 2). Stratification by age revealed that older children’s academic achievement was lower than younger children’s academic achievement (see Table 3 for mean scores). A higher percentage of errors in the attention test was associated with poorer academic achievement (r = -0.33, p<0.05), whereas a positive association was observed between concentration performance (CP) and academic achievement (r = 0.33, p<0.05), as assessed by students’ academic achievement scores.
As shown in Table 4, children with no soil-transmitted helminth infection had higher mean grip strength test results compared to their infected counterparts. The comparison between stunted and non-stunted children revealed that children not classified as being stunted achieved significantly higher mean grip strength test results. The mean VO2 max results did not differ between the two groups.
Fig 2 shows the univariate comparisons between infected versus non-infected and stunted versus non-stunted children in selective attention and academic performance. As illustrated, in these uncontrolled analyses, children infected with soil-transmitted helminths performed weaker on the d2 test of attention, compared to their non-infected counterparts. Stunted children had a lower mean concentration performance and a higher mean percentage of errors, but only the latter was statistically significant. Children without a soil-transmitted helminth infection and non-stunted children achieved statistically significantly higher academic achievement scores compared to their infected and stunted peers. Additional analyses showed that infected children had a significantly higher risk of being stunted, and vice versa (infected children: 25% stunted; non-infected children: 7% stunted; stunted children: 62% infected; non-stunted children: 27% infected χ2[1,835] = 53.2, p < .001). Accordingly, multivariate analyses were performed in the next step to avoid problems associated with multi-collinearity.
In the multiple linear regression model presented in Table 5, soil-transmitted helminth infection was statistically significantly and negatively associated with the mean CP score. The mean CP score of children with soil-transmitted helminth infection was 7.99 points lower compared to their non-infected peers. Grip strength and the estimated mean VO2 max were statistically significantly and positively associated with the mean CP score. The mean CP score increased by 0.98 points per ml kg-1 min-1 VO2 max, whereas the mean CP score increased by 0.92 points per kg grip strength. Age and soil-transmitted helminth infection were negatively associated with the error percentage in the d2 test of attention. The mean error percentage increased by 1.6% per year of age whereas a soil-transmitted helminth infection was associated with a 3.3% higher error percentage compared to non-infected children. The mean VO2 max was statistically significantly and positively associated with the mean E% score. The mean E% score decreased by 0.24% per ml kg-1 min-1 VO2 max.
In the multiple linear regression model, lower SES, male sex, higher age, being infected with soil-transmitted helminths, and a lower cardiorespiratory fitness were statistically significantly and negatively associated with academic achievement. The mean academic achievement score increased by 0.06 per point in the SES score. By contrast, children’s academic achievement score decreased by 0.43 per additional year of age and was 0.45 lower among children classified as being infected with soil-transmitted helminths compared to non-infected peers. Boys had 0.42 lower academic achievement scores than girls and a higher VO2 max was associated with higher academic achievement, yet only with a marginal increase of 0.02 per ml kg-1 min-1 VO2 max. No significant associations were observed for stunting, food insecurity, and grip strength.
The most important findings of the present study are that, in the multivariate analyses, soil-transmitted helminth infections and lower physical fitness were negatively associated with selective attention, while lower SES, positive soil-transmitted helminth infection status, lower cardiorespiratory fitness, and higher age were associated with poorer academic achievement. Without implying causality, our data suggest that an infection with T. trichiura, A. lumbricoides, or both, is associated with lower selective attention capacity (in terms of attention capacity and accuracy) and reduced physical fitness among school-aged children in terms of muscular strength measured as grip strength. Moreover, children infected with soil-transmitted helminths had significantly lower academic achievement scores. It is conceivable that the general well-being of infected children, as expressed in abdominal pain, fatigue, and listlessness, negatively affects their cognitive performance [21,25]. In a study by Liu et al. [22] carried out in South-western China, children infected with either T. trichiura or A. lumbricoides were also lagging behind their non-infected peers. In the same study from China, infection with one or multiple species of soil-transmitted helminths was associated with reduced speed of processing and working memory performance and worse school performance (in terms of standardized mathematics test scores).
Heavy A. lumbricoides and T. trichiura infections have been associated with cognitive impairment and were both linked with significantly increased disability weight (DW) in the Global Burden of Disease (GBD) study [45]. Our finding that soil-transmitted helminths are associated with reduced attention capacity and accuracy is novel and warrants further investigation. Yet, to our knowledge, there is no conclusive evidence whether reduced physical fitness and strength are a direct consequence of soil-transmitted helminth infection. Our analyses did not reveal any associations between VO2 max and single or double species helminth infections. Müller et al. [46] found that 9-year-old boys infected with T. trichiura had a lower mean VO2 max estimate in a slightly different sample of children from the same cohort. Of note, another cross-sectional study by Müller et al. [47] did not find any correlation between VO2 max results and soil-transmitted helminth infections among school-aged children from Côte d’Ivoire, which is at odds with findings from China by Yap et al. [48] who reported reduced VO2 max estimates of school-aged children infected with T. trichiura. In our study, irrespective of age, children infected with A. lumbricoides, T. trichiura, or both species concurrently had a lower mean grip strength compared to non-infected children. Yap et al. [20] reported increased grip strength one month after albendazole treatment. Given these findings, further research is needed to deepen the understanding of whether and how soil-transmitted helminth infections are related to VO2 max and grip strength among school-aged children.
The univariate analyses also suggested that stunted children have deficits in selective attention and achieve lower academic performance compared to non-stunted children. However, these associations disappeared in the multiple regression analyses. Thus, while previous research suggested that the main causes of stunting include intrauterine growth retardation, inadequate nutrition, and poor dietary diversity to support the rapid growth and development of infants and young children [49], and that stunting can result in cognitive impairments [49,50], the association between stunting and the outcomes was no longer significant after all possible influences were taken into account. In the present study, multivariate analyses are warranted as some of the independent variables were associated. For instance, our findings confirmed that stunted children had a significantly higher risk of being infected with soil-transmitted helminth, which is in line with prior research showing that chronic soil-transmitted helminthiasis is a cause of stunting [49].
The univariate analyses further showed that stunted and non-stunted children differed significantly in grip strength, whereas they had similar mean VO2 max values. Our findings align with a study of Malina et al. [51] reporting that stunted children had lower grip strength than their non-stunted peers. Grip strength was shown to be a valid indicator for total muscle strength in children [52], and was associated with physical health outcomes in previous studies with children and adolescents [53]. While we did not observe a statistically significant difference, a recent study by Armstrong et al. [54] found that stunted South African primary school children also performed poorer in a 20 m shuttle run test as well as in other physical fitness tests, a finding corroborated by Yap et al. [48] who reported a lower mean VO2 max estimate of stunted school children in China.
With regard to selective attention and academic achievement and how they might be associated with soil-transmitted helminth infection status, stunting, food insecurity, and physical fitness, we found that attention capacity is associated with infection status and physical fitness. This confirms the notion of a negative relationship between T. trichiura and A. lumbricoides on the one hand and cognition on the other, as reported in prior research [21,22,25,55]. Furthermore, our findings suggest that after controlling for confounding factors, academic achievement is negatively associated with age and soil-transmitted helminth infection, and positively associated with SES.
Only few studies have looked at the relationship between children’s physical fitness and their selective attention in low socioeconomic settings. A study by Tine and Butler [56] reported improvements in selective attention after a 12 min session of aerobic exercise in both lower- and higher-income children. Lower-income children exhibited greater improvements in selective attention compared to their higher income peers. The fact that aerobic fitness was associated with selective attention in our sample of disadvantaged school children, combined with the finding of Tine et al. [56] is highly encouraging since (i) primary school children’s aerobic fitness can be improved through regular training [57], and (ii) selective attention is associated with academic and cognitive outcomes [2]. As highlighted by Armstrong et al. [58], there is a particularly pronounced need for encouraging fitness in South African primary schools. However, the multifactorial nature of physical fitness and attention capacity of children growing up in socioeconomically deprived environments requires that health conditions such as asthma, fetal alcohol syndrome, and human immunodeficiency virus (HIV) infection status, which were not assessed in the present study, must also be considered [59].
Stratification by age revealed that 8- and 9-year-old children achieved better academic achievement scores than their 10- to 12-year-old peers. This may be explained by the fact that disadvantaged communities do not have the financial means to promote children with special needs or learning disabilities [59]. Children suffering from reading difficulties, attention deficit hyperactivity disorder (ADHD), fetal alcohol syndrome or neglect do not get the required academic support and as a consequence are not able to keep up with their peers. Students failing to achieve adequate grades are retained up to 3 years until they get too old and automatically progress to the next grade [59,60], which explains the wide age range of the participants in the current study. Girls seemed to achieve better academic results compared to boys, while there was no statistically significant difference between sex in the test of attention. A meta-analysis by Voyer and Voyer [61] found a consistent female advantage in school marks for all course content areas.
The present study expands previous research in several important ways; to our knowledge, associations between selective attention and soil-transmitted helminth infection status as well as stunting has not previously been investigated. It also contributes to the finding that chronic soil-transmitted helminth infections and cognitive impairment are associated [62]. Furthermore, this study provides new evidence that physical fitness might be associated with increased selective attention in children from a low socioeconomic environment, even after controlling for major covariates.
Our study has several limitations. First, our results are derived from a cross-sectional study and causal inferences cannot be drawn. Second, academic achievement was measured with the average end-of-year mark (achieved at the end of grade 3), which corresponds to the summary of four subjects (mathematics, home language, additional language, and life skills). While the objectivity of school grades can be questioned as a reliable outcome in empirical research (e.g., due to attributions or stereotypes of the teachers, different standards between classes/schools), this measure has a high ecological validity because sufficiently high grades are needed for academic promotion. Moreover, the influence of class was controlled for, and our study showed that selective attention and the academic achievement scores were moderately correlated (r>0.30). Third, we used an indirect measurement of VO2 max to assess aerobic fitness and it is still debated whether the maximal oxygen uptake is receptive enough for change [63] due to varying personal living conditions. However, this test was chosen because it seemed well suited for a resource-constrained setting due to its ease of application [36]. Furthermore, the 20 m shuttle run test proved to be a valid measure of children’s physical fitness in previous studies [64], and could be related to various health outcomes in school-aged children [65]. Fourth, anthropometric measurements were taken only once, which could be a source of increased measurement error. Fifth, only a single stool sample was obtained from each child. Hence, some of the helminth infections, particularly those of light intensity, were missed. Finally, we acknowledge that our study took place in disadvantaged communities (quintile three schools). As a consequence, variation in SES was limited, which might have resulted in an underestimation of SES as a predictor of selective attention and academic achievement.
In conclusion, our study provides new insights into the relative importance of different determinants of school children’s selective attention in a disadvantaged setting of South Africa. We found that soil-transmitted helminth infection and lower physical fitness may hamper children’s capacity to pay attention during cognitive tasks, and directly or indirectly impede their academic performance. It is conceivable that poor academic achievement will hinder children from realizing their full potential and disrupt the vicious cycle of poverty and ill health.
|
10.1371/journal.ppat.1002594 | TGF-β1 Down-Regulation of NKG2D/DAP10 and 2B4/SAP Expression on Human NK Cells Contributes to HBV Persistence | The mechanism underlying persistent hepatitis B virus (HBV) infection remains unclear. We investigated the role of innate immune responses to persistent HBV infection in 154 HBV-infected patients and 95 healthy controls. The expression of NKG2D- and 2B4-activating receptors on NK cells was significantly decreased, and moreover, the expression of DAP10 and SAP, the intracellular adaptor proteins of NKG2D and 2B4 (respectively), were lower, which then impaired NK cell-mediated cytotoxic capacity and interferon-γ production. Higher concentrations of transforming growth factor-beta 1 (TGF-β1) were found in sera from persistently infected HBV patients. TGF-β1 down-regulated the expression of NKG2D and 2B4 on NK cells in our in vitro study, leading to an impairment of their effector functions. Anti-TGF-β1 antibodies could restore the expression of NKG2D and 2B4 on NK cells in vitro. Furthermore, TGF-β1 induced cell-cycle arrest in NK cells by up-regulating the expression of p15 and p21 in NK cells from immunotolerant (IT) patients. We conclude that TGF-β1 may reduce the expression of NKG2D/DAP10 and 2B4/SAP, and those IT patients who are deficient in these double-activating signals have impaired NK cell function, which is correlated with persistent HBV infection.
| NK cells have been viewed as the most important effectors of the initial antiviral innate immune response. Their activation depends on the integration of signals from “co-activation” receptors, and the cytotoxic effects of NK cells on target cells are tempered by a need for combined signals from multiple activating receptors, such as NKG2D and 2B4. In this study, we showed that NKG2D and 2B4 expression levels were decreased on NK cells from patients in the IT phase of HBV infection. We further demonstrated that lower levels of intracellular adaptor proteins (DAP10 and SAP) were associated with lower surface expression of NKG2D and 2B4. As a result, the synergistically co-activated signalling pathway initiated by NKG2D and 2B4 did not operate properly in IT-phase patients. We demonstrated that high levels of soluble TGF-β1 were associated with the reduction of NKG2D and 2B4 in patients. In addition, we showed that TGF-β1 causes the cell-cycle arrest of NK cells by up-regulating the levels of p15 and p21 in NK cells from IT patients. Collectively, these findings may contribute to our understanding of the immune tolerance mechanism and aid in the development of novel therapeutic methods to clear HBV infection during the initial phase.
| Hepatitis B virus (HBV) infects more than 350 million people worldwide, accounting for over 1 million deaths annually due to immune-mediated chronic liver damage [1]–[3]. The course of HBV infection is complicated. Three phases of chronic HBV (CHB) infection are now widely accepted: 1) an immune tolerant (IT) phase, characterised by an HBV DNA concentration >200,000 IU/mL, normal alanine aminotransferase (ALT) levels, normal liver biopsy or only minimal inflammation and perinatal infection of infants born to HBsAg/HBeAg-positive mothers; 2) an immune active (IA) phase, which is also referred to as the “chronic hepatitis B phase” or the “immune clearance phase”, characterised by an HBV DNA concentration >20,000 IU/mL, elevated ALT levels and active hepatic inflammation on biopsy; and 3) an inactive (IN) phase, characterised by HBV DNA levels <2000 IU/mL, normal ALT levels and minimal or absent hepatic inflammation [4]. During the IA phase, the immune system of the host recognises the virus as foreign and initiates the immune clearance response, which results in hepatocyte damage. After one or more episodes of reactivation to the IA phase, patients begin the IN phase [5]–[7]. Patients in the IT phase have normal ALT levels and elevated levels of HBV DNA, commonly well above 1 million IU/mL; this phase can last anywhere from a few to >30 years [8]. The IT phase has been suggested to occur most frequently in patients who were infected via perinatal transmission from HBeAg-positive mothers [9]. In China, approximately 2 million infants are infected with HBV via perinatal transmission from HBsAg/HBeAg-positive mothers annually, and many of them are at high risk of developing chronic liver inflammation resulting in cirrhosis and hepatocellular carcinoma (HCC) in later life [10]–[13]. The risk of developing HCC with HBV infection is higher in East Asian countries than in Western countries, possibly due to the frequency of earlier viral infection [14]. During the IT phase of CHB infection, the virus evolves strategies to evade immune clearance in the majority of patients. However, the tolerance mechanism of the IT phase has not been widely studied. Due to their normal ALT levels, there is no available treatment to reduce the very high HBV DNA levels or alleviate psychological pressure in IT-phase patients. The development of an anti-HBV therapy for such patients will require insight into the mechanisms of HBV persistence.
The innate immune system provides the first line of defence in antiviral responses and activates adaptive immune responses. NK cells have been viewed as the most important effectors of the initial antiviral innate immune system [15]–[16]. Previous investigations have demonstrated that NK cells may be particularly important in patients with CHB. NK cells are highly enriched in the liver, and the site of HBV replication, and they are partially functionally tolerant in CHB [13], [17]. The substantial quantity of NK cells in the liver suggests that they act as “watcher cells”, surveying the liver for indications of cellular stress and implying that HBV has to evade NK cell-mediated immune responses to establish a persistent viral infection. The evidence shows that the cytotoxic capacity of NK cells is retained. Moreover, the activation of NK cells and the secretion of IFN-γ are strongly inhibited during CHB infection [18]. Blockage of IL-10 with or without TGF-β1 can restore the capacity of NK cells to produce the antiviral cytokine IFN-γ in CHB patients [19]. NKG2D, the activating receptor of NK cells, is constitutively expressed on human NK cells and CD8+ T cells [20]. The importance of the NKG2D pathway is highlighted by evidence that tumours and viruses have developed distinct escape mechanisms to avoid NKG2D-mediated recognition [21]–[26]. The signalling lymphocyte activation molecule (SLAM)-related receptor 2B4 is predominantly expressed on human NK cells and CD8+ T cells. The immunoregulatory role of 2B4 as an activating or inhibitory receptor depends on three factors: 1) surface expression, because costimulatory qualities are associated with low expression and inhibitory qualities are associated with high expression; 2) the coexpression of additional inhibitory molecules; and 3) the presence of the intracellular adaptor protein SLAM-associated protein (SAP) [18], [27]–[30]. NKG2D and 2B4 are the main triggering receptors of NK cells [31]. Many studies have provided evidence for a functional dichotomy in patients with chronic HBV that may contribute to virus persistence [32]. NK cell-mediated cytotoxicity is efficiently initiated by the NKG2D activation signal on NK cells. NKG2D recognises pressure-induced antigen signals on a target cell, whereas 2B4 receives the costimulatory signal. Therefore, these two molecules play a key role in NK cell activation and function. The tolerance mechanism of HBV persistence and the contribution of the NKG2D/DAP10 and 2B4/SAP pathways to the control of persistent HBV infection are unclear. Here, we show that NKG2D/DAP10 and 2B4/SAP are down-regulated on circulating NK cells and are associated with the impaired functionally of NK cells in IT-phase patients. Moreover, this defect is mediated by TGF-β1, which causes NK cell-cycle arrest by inducing high expression of p15 and p21. These findings may contribute to our understanding of immune tolerance mechanisms and may aid in the development of novel therapeutic methods to clear the viral infection during the initial phase.
The multiple functions of NK cells, such as cytotoxicity and cytokine secretion, can be induced through interactions between inhibitory and activating NK receptors and their respective ligands [15]. NKG2D is constitutively expressed on NK cells and is one of the main triggering receptors of NK cells. NK cell-mediated cytotoxicity is efficiently initiated by engaging the NKG2D-DAP10 receptor complex on NK cells [31]. To explore the effector potential of NK cells during persistent HBV infection, we first analysed the frequency of NKG2D expression on NK cells in 93 patients with HBV infection compared to 63 healthy gender- and age-matched controls. Due to the restricted availability of fresh, persistently HBV-infected liver samples from which to isolate infiltrating NK cells, we examined the expression of NKG2D on circulating NK cells. HBV-infected patients were classified into three phases based on their natural histories: the IT phase, the IA phase and the IN hepatitis B phase [4], [33]–[34]. The proportion of circulating NKG2D+ NK cells was significantly decreased in patients in the IT phase relative to healthy controls (P<0.0001) and patients in either the IA phase (P = 0.04) or the IN (P = 0.001) phase (Figure 1.A, B). The proportion of total NK cells in IT-phase patients was also lower than in CHB patients in other phases or in healthy controls (Figure S3.B). More importantly, we observed that both the percentage of circulating NKG2D-expressing NK cells and the absolute count of NKG2D+ NK cells were significantly lower in IT-phase patients than in healthy controls (P<0.0001) or patients in the IN (P = 0.019) but not patients in the IA (P = 0.303) (Figure 1.D).
To explore whether other activating NK receptors were expressed at low levels in the IT patients, we quantified another NK cell-activating receptor, 2B4, using flow cytometry. The proportion of circulating 2B4+ NK cells displayed significantly lower levels in the IT patients than in healthy controls (P<0.0001) and in patients in the IN (P = 0.022) (Figure 1.A, C). Similarly, the absolute number of 2B4-expressing NK cells in patients in the IT phase was lower than in healthy controls (P<0.0001) (Figure 1.E). Upon further analysis, a linear relationship was observed between the percentage of 2B4 and NKG2D on NK cells (r = 0.7695, P<0.0001) (Figure 1.F). We also analysed the frequency of the expression of other NK cell activation receptors (NKp30, NKp44, NKp46, CD16, CD27 and CD226) on circulating NK cells from healthy controls and CHB patients, but there were no significant differences in their expression levels on NK cells between patients and healthy controls (Figure S2.A,B). The levels of NKG2D and 2B4 did not correlate with age, HBV viral load or ALT/AST levels (data not shown), and NKG2D (P = 0.0112) and 2B4 (P = 0.0101) expression levels were lower in females than males (Figure S3.A).
The activation of NK cells depends on the integration of signals from co-activation receptors, and the cytotoxic effects of NK cells on target cells are tempered by a request for combined signals from multiple activating receptors, such as NKG2D and 2B4 [16], [35]–[36]. Therefore, we hypothesised that these phenotypic changes might be paralleled by functional alterations in NK cells. To determine whether NK cells in IT-phase patients have an intrinsic defect in cytolytic activity, NK cell cytotoxicity was evaluated by measuring the lysis of 51Cr-labelled K562 cells. To ensure that there was no other cellular factor that could influence NK cell cytotoxicity, we purified NK cells from PBMCs by negative selection. IT-phase patients were deficient in K562 killing compared to healthy controls and patients in the IA or IN phase (Figure 2.A). Across multiple evaluations, the IT patients had a mean of 23.6±5.5% K562 lysis at a 5∶1 E∶T ratio compared with 51.4±7.7% for the controls (P = 0.0384). No significant difference in NK cell cytotoxicity was detectable between patients in the IA or IN phases and healthy controls (Figure 2.B). These data demonstrate that patients in the IT phase have a specific defect in NK cell cytotoxic activity. Furthermore, a positive correlation was observed between the cytotoxicity of NK cells and the percentage of NKG2D+NK cells (r = 0.7264) and the percentage of 2B4+NK cells (r = 0.4183) (Figure 2.C). To further evaluate the potential function of NK cells in IT-phase patients, we examined their capacity to produce cytokines by measuring IFN-γ expression following stimulation with IL-12. There was a significant reduction in the production of IFN-γ produced by NK cells from patients in the IT phase compared with healthy controls and IA- and IN-phase patients (Figure 2.D, E). There was also no change in T-bet expression in NK cells from patients and healthy controls (Figure S4.A, B).
To further determine whether the defect in NK cell function could be reversed during antiviral therapy, PBMCs from patients at the onset of inflammation and after treatment were stimulated with IL-12 for 16 h, and IFN-γ, and CD107a production was determined. As shown in Figure 2, the production of IFN-γ (F) and CD107a (G) was higher in NK cells from patients after treatment compared with before treatment, suggesting that NK cell function could be reversed by antiviral therapy. To further validate this hypothesis, ex vivo NK cell cytotoxicity against K562 cells was measured (Figure 2.H). Our results demonstrated that NK cell cytotoxic activity was strengthened after antiviral therapy. These results suggest that in vivo administration of nucleoside analogues partially overcomes the deficit in NK cell function in CHB patients.
We observed that both the percentage and the absolute count of NKG2D and 2B4 were decreased on circulating NK cells during IT-phase HBV infection. We then asked whether the decreased expression of activating receptors on NK cells was paralleled by functional alterations. There is a consensus that the NKG2D-DAP10 receptor complex on human NK cells efficiently initiates cell-mediated cytotoxicity [31]. When NK cells encounter target cells expressing CD48, the 2B4-SAP–Fyn complex is responsible for the activation of NK cells. To test this hypothesis, we measured the expression of the intracellular adaptor protein DAP10 and the two members of the SAP family of adaptors that are expressed in humans: SAP (also known as SH2D1A or DSHP) and Ewing's sarcoma-activated transcript-2 (EAT-2, also known as SH2D1B). The mRNA expression levels of DAP10, SAP and EAT-2 in NK cells were investigated in HBV patients and compared to healthy controls. As shown in Figure 3.A., our results demonstrated that the mRNA expression levels of DAP10 and SAP were lower in NK cells from IT patients than in those from IA patients and healthy controls; there were no differences between IA patients and healthy controls. Moreover, there were no significant differences in EAT-2 mRNA expression levels in NK cells between patients and healthy controls. To further investigate the signalling potential of activating NKRs within NK cells, we assessed the protein expression levels of DAP10 and SAP in NK cells. Due to having access to limited numbers of NK cells from blood samples, we investigated the expression of DAP10 and SAP in NK cells by immunofluorescence. As shown in Figure 3.B, we found fewer DAP10+ NK cells in samples from IT-phase patients than in those from patients in other phases and healthy controls. The proportion and absolute numbers of circulating DAP10+ NK cells were also lower in the IT patients compared with the patients in other phases and healthy controls (Figure 3.C, D). No significant differences in DAP10+ NK cells were detectable between patients in the IA and IN phases (Figure 3.C, D). Upon further analysis, the percentages of SAP-expressing NK cells were similar in HBV patients and healthy controls (>95% of NK cells in all four groups). However, a decreased MFI was detected in SAP+ NK cells from IT patients compared with healthy controls and IA or IN patients (Figure 3.G). No significant difference in SAP+ NK cells was detectable between patients in the IA and IN phases (Figure 3.F, G). Furthermore, to verify whether SAP proteins were reduced or not, we examined the expression of SAP in NK cells using western blotting (Figure S5). Our new data demonstrated that considerably low levels of SAP on NK cells from IT patient samples and relatively low levels of SAP in IA, IN patients compared with healthy controls. These data suggest that the loss of NKG2D+ and 2B4+ NK cells in IT patients may be accompanied by the loss of DAP10+ and SAP+ NK cells. To verify whether this phenomenon was associated with functional consequences, we specifically silenced DAP10 (si-DAP10) or SAP (si-SAP) in NK92 cells using RNA interference (RNAi). As shown in Figure 4A, B, the mRNA expression levels of DAP10 and SAP on NK92 cells were significantly lower after transfection with either siRNA. Furthermore, we chose siR-SAP-214 and siR-DAP-501 to investigate the functional consequences. As expected, compared with NK92 cells transfected with the negative control siRNA (siR.NC), NK92 cells transfected with either DAP10 siRNA (siR-DAP) or SAP siRNA (siR-SAP) resulted in a much stronger deficiency in NK cell cytotoxicity (Figure 4.D). These results demonstrate that DAP10 and SAP, at least in part, mediate the effects of NK92 cell cytotoxicity.
To evaluate the functional consequences of the reduction of the double-activating signals (NKG2D and 2B4), secondary cross-linking goat anti-mouse F(ab')2 antibody was added to NK cells obtained from fresh peripheral blood samples and preincubated with activated mAbs for NKG2D and 2B4 to investigate the mobilisation of Ca2+ triggered by NKG2D and 2B4 synergy [37]. Because the supply of freshly isolated NK cells from patients and healthy controls is quite limited and because in vitro expansion of NK cells in the presence of cytokines changes their signalling properties, we performed confocal microscopy experiments to follow intracellular Ca2+ responses in 500 NK cells after the receptors were co-crosslinked. Recordings of NK cells revealed that NKG2D and 2B4 cross-linking elicited distinct responses in cells from different sources (Figure 5). In the NK cells from healthy controls and IA-phase patients, cross-linking induced a sharp and sustained increase in intracellular Ca2+ (Figure 5.A, B). Notably, in NK cells from IT-phase patients, there was no significant increase in intracellular Ca2+ flow (Figure 5.C). The Ca2+ mobilisation induced by NKG2D and 2B4 synergy in >100 NK cells from representative healthy controls or patients at 200 (Figure 5.D) and 500 s (Figure 5.E) was analysed. Altogether, these results suggest that the synergistic co-activation signalling pathway activated by NKG2D and 2B4 may not operate properly in IT-phase patients due to the reduction in NKG2D and 2B4 expression in NK cells.
Surface expression of NKG2D has been shown to be down-regulated by TGF-β and IL-21 [38]–[40]. During chronic HCV infection, TGF-β down-modulates the expression of NKG2D on NK cells, leading to the impairment of their function [26]. Therefore, to verify whether there are also elevated levels of TGF-β1 in patients with persistent HBV infection, we used a cytometric bead array (CBA) inflammation kit and ELISA technology to simultaneously quantify the presence of multiple cytokines in sera from HBV patients and healthy controls. The cytokines analysed were IL-1α, IL-1β, IL-2, IL-4, IL-6, IL-10, IL-12p70, IL-13, TNF, IFN-γ and TGF-β1. The peak concentrations of TGF-β1 in IT patients were far in excess of those observed for the other ten cytokines; this cytokine was also present at significantly higher concentrations than those measured in healthy controls and patients in other phases of infection (Figure 6.A). In addition, an inverse linear relationship was observed between TGF-β1 concentration and the percentage of circulating NKG2D-expressing NK cells (Figure 6.B, C). An identical relationship was identified with the proportion of 2B4-expressing NK cells (Figure 6.D, E). These data demonstrate a correlation between the down-regulation of activating NK receptors and high levels of TGF-β1 and suggest a role for the influence of soluble TGF-β1 on NK cells in IT-phase HBV patients.
Down-modulation of expression is associated with elevated levels of TGF-β1 and has also been observed in cancer patients [39], [41]–[45]. Therefore, we hypothesised that TGF-β1 might modulate NKG2D surface expression. We investigated whether the altered expression of activating NK cell receptors was induced by TGF-β1. To test this possibility, we used NK cells from healthy control peripheral blood preincubated with TGF-β1 for 3 days. The percentage of NKG2D and 2B4 levels on NK cells, as monitored by flow cytometry, were markedly down-regulated (Figure 7A, B). Fresh NK cells were also cultured with sera from healthy controls and patients in different immune states. After 3 days, the surface expression of NKG2D on NK cells preincubated with sera from IT patients (Figure 7.C, red lines) was markedly lower than on NK cells preincubated with sera from healthy controls (black lines). Furthermore, anti-TGF-β1 partially restored the expression of NKG2D on NK cells co-cultured with sera from IT-phase patients (Figure 7.D, blue lines). In contrast, there were no significant changes in NK cells preincubated with isotype control Abs (green lines) (Figure 7.D). In the same cell culture experiments, NK cells isolated from healthy controls were stimulated with TGF-β1 or sera from various sources. The proportion of 2B4 levels on NK cells decreased when co-cultured with TGF-β1 or sera from IT patients (Figure 7.B, E, red lines). Similarly, the expression levels of 2B4 recovered when the incubation was performed in the presence of a neutralising anti-TGF-β1 antibody (Figure 7.F, blue lines). Cumulative data shown in Figure 7.G, H indicate that the percentage of NKG2D and 2B4 levels on NK cells decreased significantly after co-culture with sera from IT patients, and minor reductions were observed when NK cells were cultured with sera from IA- or IN-phase patients. The expression of NKG2D was partially restored and 2B4 expression recovered when NK cells were incubated with neutralising anti-TGF-β1 antibody. There were no significant changes in NK cells preincubated with isotype control Abs. Taken together, these data indirectly suggest that the expression levels of NKG2D and 2B4 on NK cells were suppressed by high levels of TGF-β1 in IT-phase patients.
During chronic HCV infection, TGF-β1 down-modulates the expression of NKG2D on NK cells, leading to the impairment of their function [26]. Inhibition of TGF-β1 restores the ability of NK cells from both the peripheries and livers of patients with CHB infection to produce antiviral IFN-γ [19]. Anti-TGF-β1 partially restored NKG2D and 2B4 surface expression on NK cells from IT patients. We then postulated that high levels of TGF-β1 might affect the functions of NK cells in IT patients. To investigate this hypothesis, we used in vitro experimental models. First, NK cells obtained from healthy controls were stimulated with sera from IT patients plus anti-TGF-β1 antibodies or isotype control Abs for 72 h to imitate a physiological state. The Ca2+ mobilisation induced by synergism between NKG2D and 2B4 on NK cells was then assessed. As observed in Figure 8.A, sera from IT-phase patients abrogated the Ca2+ mobilisation potential of NK cells from healthy controls. Additionally, anti-TGF-β1 partially restored Ca2+ flux in NK cells from healthy controls that had been incubated with sera from IT patients (Figure 8.C). Recordings of NK cells incubated with sera from different patients and anti-TGF-β1 Abs revealed distinct responses. In some patients (Pt508), NKG2D+2B4 cross-linking induced a sharp, oscillating rise in intracellular Ca2+. In other patients (Pt487, 489, and others) lower, transient peaks were observed (Figure 8.C). The Ca2+ mobilisation induced by NKG2D and 2B4 synergy in >100 NK cells from representative groups at 200 (Figure 8.D) and 500 s (Figure 8.E) was analysed. Overall, a rise in Ca2+ flux induced by NKG2D and 2B4 synergism was observed when anti-TGF-β1 Ab was added.
Interestingly, four activating NK cell molecules (NKG2D/DAP10 and 2B4/SAP) have been found to be expressed at low levels in IT patients, and this phenotype was associated with high levels of TGF-β1, reminiscent of the cell-cycle arrest induced by TGF-β1. Therefore, we next examined whether high levels of TGF-β1 in IT patients could facilitate cell-cycle arrest in NK cells. An important role of TGF-β1 involves restricting the growth of neuronal, epithelial and hematopoietic cells. Key elements of this mechanism are the expression of the cyclin-dependent kinase inhibitors p15INK4b and p21 (also called WAF1/CIP1) [46]–[47]. These proteins play an important role in restraining cell-cycle progression. Substantial evidence indicates that TGF-β1 participates in blocking the development of T-lineage acute lymphoblastic leukaemia by suppressing T-cell proliferation [48]. It has also been shown that the effect of TGF-β1-mediated depression of NKG2D surface expression may only depend on the transcription levels of NKG2D [39]. Due to the limited supply of fresh NK cells from patients, NK cells from healthy controls were stimulated with TGF-β1 or serum from various sources (IA, IT and IN patients) in the presence or absence of anti-TGF-β1Ab. Strikingly, NK cells preincubated with serum from IT patients exhibited both significantly decelerated cell proliferation and G1 cell-cycle arrest (Figure 9). Conversely, preincubation of NK cells with sera from IT patients plus anti-TGF-β1Ab resulted in the restoration of cell proliferation and a decrease in the cell population in G1 phase (Figure 9).
The TGF-β-Smad signal-transduction pathway is an important tumour growth-suppressor pathway. Binding of TGF-β to the functional TGF-β1RII in NK cells induces the phosphorylation of the signalling molecule Smad2 and initiates signalling in the TGF-β-Smad pathway [46]–[47], [49]. The cyclin-dependent kinase inhibitors p15 and p21 play important roles in restraining cell cycle progression [46]. To further substantiate our finding that high levels of TGF-β1 inhibited the NK cell cycle in IT patients, we examined the expression of Smad-2, phosphorylation of Smad2 (Smad-2P) and expression of p15 and p21 in NK cells using western blotting. First, NK cells were stimulated for 30 min, 12 h and 72 h with TGF-β1. As shown in Figure 10.A, TGF-β1 was able to induce Smad-2P in NK cells from controls at 30 min and 72 h. p21 was undetectable in NK cell extracts, and p15 was able to be induced only after a 72-h incubation with TGF-β1. Next, NK cells were preincubated with sera from IT patients or with hepatitis ascites with or without anti-TGF-β1Ab for 72 h. As expected, sera and ascites were able to induce the expression of P-Smad-2 and p15 but had no effect on the expression of p21. In comparison, our analysis of NK cells preincubated with anti-TGF-β1 Ab confirmed the absence of P-Smad-2 and the reduced levels of Smad2 and p15 in the absence of TGF-β1 signalling (Figure 10.B). We then assessed the endogenous status of Smad2, P-Smad-2, p15 and p21 in patients. Fresh NK cells obtained from the peripheral blood of three IT patients were analysed using western blotting (Figure 10.C). Relatively high levels of Smad-2P, Smad2 and p15 on NK cells were observed in IT patient samples. In particular, the levels of p21 exhibited considerable enhancement in IT patients (Figure 10.C).
Altogether, these data demonstrate that TGF-β1 signalling is responsible for the defects in NK cell phenotypes and performance in IT-phase patients, and the defective NK cells found in patients facilitate persistent HBV infection.
There are over 350 million persistent HBV carriers worldwide, and approximately 90% of children become chronic carriers after HBV infection [1], [3]. Immune tolerance is a serious problem in CHB carriers, who are at high risk of developing cirrhosis and HCC later in life [4], [10]. HBV persistence is thought to result from inefficiencies of innate and adaptive immune responses. NK cells are a major component of innate immunity. Accumulating evidence has suggested a role for NK cells in the fight to control persistent virus infection [15], [50]–[52]. However, the tolerance mechanisms of HBV persistence have not been well explored. For the first time (to our knowledge), we demonstrated that NKG2D/DAP10 and 2B4/SAP were down-regulated on circulating NK cells. Consequently, these NK cells were functionally impaired in IT-phase patients. The loss of these molecules was mediated by TGF-β1, which resulted in cell-cycle arrest due to the induction of p15 and p21. Our results indicated that NKG2D and 2B4 expression were decreased on circulating NK cells from IT-phase patients but not CHB patients in other phases or healthy controls. Furthermore, DAP10 and SAP, the intracellular adaptor proteins of NKG2D and 2B4 in humans, were also significantly reduced in NK cells from IT patients. It has been reported that NK cell cytotoxicity towards target cells is tempered by a request for combined signals from multiple activating receptors, such as NKG2D and 2B4 [16], [35]–[36]. To evaluate the functional consequences of the observed reduction in the proportion of NKG2D and 2B4, Ca2+ mobilisation triggered by the double-activating signals was analysed. Our data revealed that the down-regulation of Ca2+ flux was induced by synergism between NKG2D and 2B4 in NK cells from IT patients but occurred at normal levels in IA patients and healthy controls. In addition, NK cell cytotoxicity and IFN-γ production were decreased in IT patients compared to healthy controls and IA patients. Anti-TGF-β1 Abs could partially restore Ca2+ flux in NK cells from healthy controls incubated with sera from IT patients. Moreover, anti-TGF-β1 also restored NKG2D and 2B4 surface expression on NK cells incubated with sera from IT patients. p21 and p15 were elevated in IT patients and induced the arrest of the NK cell cycle. Taken together, these results suggest that TGF-β1 reduces NKG2D/DAP10 and 2B4/SAP expression on NK cells during persistent HBV infection and suppresses innate antiviral immunity by blocking the cell cycle, which would eventually provide an additional HBV strategy to avoid NK cell-mediated recognition.
Our results suggest that NK cells may be functionally impaired in IT patients. This conclusion is supported by at least four important findings. First, the percentage of NKG2D and 2B4 levels was lower on NK cells from patients in the IT phase compared to patients in other phases and healthy controls, which indicates an activation defect in circulating NK cells. Second, lower levels of intracellular adaptor proteins were associated with lower surface expression levels of NKG2D and 2B4, which implies that the signalling pathways leading to NK-cell activation might be impeded. Specific silencing of DAP10 or SAP led to deficient NK-cell cytotoxicity. Third, patients in the IT phase had a specific defect in NK cell cytotoxic activity. Moreover, there was a significant reduction in the production of IFN-γ by NK cells from patients in the IT phase compared to healthy controls and IA- and IN-phase patients. Fourth, NKG2D and 2B4 receptor synergy down-regulated the mobilisation of Ca2+ in primary NK cells from IT patients. These observations indicate that NK cells are not completely functional during the IT phase, which may contribute to the persistence of HBV infections. In our study, IT-phase HBV patients were characterised by lower levels of NKG2D and 2B4 compared with healthy individuals. To our knowledge, this is the first report of reduced expression of NKG2D/DAP10 and 2B4/SAP in IT patients, thus indicating that the function of NK cells was impaired due to deficiencies in the double-activating signals. In this context, abundant data in cancer patients has shown that impaired NK function can be attributed to the down-modulation of activating receptors, such as NKG2D, which can be inhibited via TGF-β1 [42]–[44]. Notably, NKG2D-dependent NK cell functions are also modulated during chronic HCV infection [26]. These findings provide further evidence for our observation that NKG2D was down-regulated during the IT phase of HBV infection. The increased expression of 2B4 on virus-specific CD8+ T cells, both in the peripheral blood and in the liver, is believed to mediate inhibitory signalling in the absence of SAP during CHB infection [53], but this has not been evaluated in NK cells in the presence of HBV infection. Interestingly, the absence of 2B4 resulted in diminished LCMV-specific CD8+ T cell responses and prolonged viral persistence in mice persistently infected with LCMV. Additionally, long-lasting viral persistence was regulated by 2B4-deficient NK cells acting early in infection. These observations illustrate the value of NK cell self-tolerance to activated CD8+ T cells in early infection, similar to the IT phase of HBV infection; these results also demonstrate how NK cells can regulate a persistent infection that appears to be dependent on T cell responses [54]. We also analysed the frequency of NKp30, NKp44, NKp46, CD16, CD27 and CD226 expression in circulating NK cells from healthy controls and CHB patients. We observed that the frequency of NKp30 was slightly decreased in patients in the IT phase relative to the healthy controls. No statistical differences existed on other receptors between patients and healthy controls. It has been previously shown that TGF-β1 can down-regulate NKp30 and NKG2D in vitro, suggesting that slightly decrease of NKp30 in IT-phase patients may related to the high levels of TGF-β1 [39]. Evidence from previous studies suggested that SAP deficiency could lead to the inhibition of NK cytotoxicity in humans [28], [55]–[56]. It has also been shown that in the absence of SAP, lymphoma development would normally be eliminated by apoptosis in patients with X-linked lymphoproliferative disease [57]. However, the analysis of potential apoptosis caused by reducing expression of SAP indicated that the deficiency of SAP would not lead to the significant apoptosis of NK92 cells (Data not show). Here, we show that the expression of SAP is deficient in NK cells from patients in the IT phase, suggesting that 2B4 receptors may become inhibitory, rather than activating, during the IT phase of HBV infection due to the absence of SAP. 2B4 molecules with inhibitory functions are responsible for the inability of NK cells to kill virus-infected cells. This feature may further compromise NK cell function in IT patients in the absence of SAP, which implies a positive regulatory role for SAP during NK cell activation. Furthermore, TGF-β down-modulates NKG2D expression on NK cells, leading to the impairment of their cytolytic activity and ability to produce IFN-γ during HCV infection [26]. Therefore, we suggest that there may be a functional impairment in NK cells in patients in the IT phase of HBV infection.
T and B cells each possess a single antigen receptor, which regulates their development and activation. Signals initiated through antigen receptors need co-stimulatory molecules to augment the signals. In contrast, NK cells do not possess one dominant receptor but instead depend on synergistic co-activation by NK cell receptors to initiate effector functions. In our studies, we have observed that Ca2+ mobilisation cannot be induced by NKG2D and 2B4 synergy in NK cells from IT patients, providing an explanation for the dominance of the double-activating signals in controlling NK cell activation. Strikingly, the deficiencies of the double-activating signals may be significant with respect to NK-cell tolerance. TGF-β1 inhibits NK cell activity and cytotoxicity by down-regulating NKG2D, as originally proposed in various cancer fields [42]–[44]. It has also been observed that TGF-β can impair NK cell cytolytic activity and IFN-γ production during HCV infection [26]. However, the role of TGF-β1 in blocking NK-cell function during the IT phase of HBV infection has not yet been well defined. Our data demonstrate that TGF-β1 can down-regulate NKG2D and 2B4 surface expression and restrain the Ca2+ mobilisation triggered by NKG2D and 2B4 receptor synergy. Moreover, anti-TGF-β1 Abs can restore NKG2D and 2B4 expression and also partially restore Ca2+ flux. HCV infection can induce TGF-β1, which can regulate HCV RNA replication [58]. HCV-specific Th17 cells are suppressed by virus-induced TGF-β [59]–[60]. Furthermore, the HBV X protein significantly up-regulates the expression of TGF-β1 and TGF-βRII in the LX-2 human hepatic stellate cell line [61]. The HBV-encoded pX oncoprotein amplifies TGF-β signalling [62], and HBV X antigen promotes TGF-β1 activity [63]. These findings suggest that during the IT phase of HBV infection, there may be a similar mechanism by which the virus evades host-protective immune responses.
NK cell function can be regulated by TGF-β [49] and by CD4+ CD25+ regulatory T cells (Tregs) through a TGF-β-dependent mechanism in both humans and mice [64]–[65]. We surmised that in IT patients, Treg cells might be the source of high levels of TGF-β. In a similar mechanism, Treg cells expressing membrane-bound TGF-β directly down-regulated the expression of NKG2D/2B4 on NK cell surfaces and inhibited NK cell effector functions [64]. Previous studies have reported that cancer-expanded myeloid-derived suppressor cells can directly contact NK cells and induce anergy via membrane-bound TGF-β1 [66], though other studies have found that Treg cells are the main negative regulators of NK cell function in tumours [64]. In our study, anti-TGF-β1 did not completely restore Ca2+ flux and cell-cycle arrest, suggesting that some other inhibitory molecules such as IL-10, which is also produced by regulatory CD8+ T cells [67], may also play a role in NK impairment during the IT phase of chronic HBV infection. Moreover, recent studies have shown that the partial functional tolerance induced in NK cells by the immunosuppressive cytokine environment in CHB can be corrected in vitro by the specific blockage of IL-10 and TGF-β [19]. These findings support our results for IT phase patients and suggest that regulatory CD8+ T cells may also influence NK cells in IT-phase patients by producing IL-10. Furthermore, our results demonstrate that the level of IL-10 in IT-phase serum was higher than in sera from healthy controls and IN-phase patients but lower than in sera from IA-phase patients. However, the levels of IL-10 in the sera were significantly lower than the levels of TGF-β1, which suggests that TGF-β1 was the predominant suppressive factor in IT-phase patients.
Our results suggest that the in vivo administration of nucleoside analogues partially overcomes the defect in NK cell function in patients. In addition, we observed that NK cell function was significantly lower at the onset of inflammation in IT patients, suggesting that some other inhibitory factors may restrain NK cell function during the initial phase of liver inflammation. Some studies have found that reduced IFN-γ production by NK cells in HBV infection independent of the IT phase [32], we believe this is due to different methods of classifying patient groups. Another interesting finding of our study was the observation that p21 and p15 were obviously elevated on NK cells from IT patients. The cyclin-dependent kinase inhibitors p15 and p21Waf1/Cip1 acted as inhibitors of cell-cycle progression [46]. Relatively high levels of Smad-2P and Smad2 were observed in NK cells from IT-phase patients, implying that TGF-β1 had the potential to arrest the NK cell cycle. High rates of p15 methylation in HCCs have been shown to be associated with HCV infection. TGF-β-dependent inhibition of HCV replicons was also associated with cell-cycle arrest in a Smad-dependent manner [68]. Furthermore, the expression of p21 during liver cirrhosis is related to the persistence of infection with HBV [69], which suggests that p21 plays an important role in the progression of HBV. However, one study has shown that the HBV X protein overcomes cellular senescence by down-regulating levels of p16 and p21 via DNA methylation [70]. In that study, HepG2 cells were used, which may explain why their findings differed from our own. Some previous studies reporting persistent HBV infection have focused on genetic variants [71]–[74]. Unfortunately, this study was limited by the supply of freshly isolated NK cells from patients and healthy controls, and therefore, our analysis was restricted to the circulating compartment only. A more detailed investigation of the frequency and function of intrahepatic NK cells in IT patients should be performed.
In summary, our findings demonstrate that high levels of TGF-β1 are associated with reduced NKG2D and 2B4 expression, the functional impairment of NK cell function, and consequently, with the development of persistent HBV. Our study provides a basis for improving current therapies for IT patients by blocking TGF-β1 inhibitory pathways, which could result in additive efficacy at eliminating the virus during the initial phase of CHB infection.
Peripheral blood samples and clinical assessments were obtained during routine hepatitis follow-ups. In accordance with the Declaration of Helsinki and the Ethical Board of the Institutional Review Board of the University of Science and Technology of China, all participants provided written informed consent, which was obtained before enrolment in the study.
Peripheral blood samples were obtained from the Department of Infectious Diseases of the First Affiliated Hospital of Anhui Medical University. The patient characteristics are listed in Table 1. There were no significant differences in demographic variables (gender/age) between the patient and the healthy control groups (Table 1). Blood samples were processed within 4 h of collection. Among the 249 clinical samples, 65 were diagnosed as IT phase, 48 were diagnosed as IA phase, and 41 were diagnosed as IN phase. The diagnostic criteria for the clinical terms of HBV infection were adopted from the National Institutes of Health (NIH) conferences on the management of hepatitis B in 2000 and 2006 [4], [10]. Detailed characteristics of the patients are shown in Table S1. Peripheral blood samples obtained from 95 healthy donors at Hefei Blood Bank and Anhui Provincial Hospital served as controls. All patients were negative for anti-hepatitis C and anti-human immunodeficiency virus antibodies and were treatment naïve. All patients were treated with anti-viral therapy consisting of nucleoside analogues (lamivudine, adefovir dipivoxil, and entecavir) without immunomodulators.
Fresh human peripheral blood mononuclear cells (PBMCs) were isolated from peripheral blood by Ficoll-Isopaque (Solarbio, China) gradient centrifugation. NK cells were freshly purified from PBMCs by negative selection using magnetic microbead separation kits (Miltenyi Biotec), resulting in >90% purity. Human PBMCs and NK cells were prepared and stained with monoclonal antibodies (mAbs). Fc receptors (FcRs) were blocked using normal mouse serum or rat serum. Abs against the following proteins were used for staining human PBMCs: CD3 (SK7), CD56 (B159), CD16 (3G8), CD27 (M-T271), CD226 (DX11), NKp30 (p30-15), NKp44 (p44-8.1), NKp46 (9E2/NKP46), NKG2D (1D11) (BD PharMingen) and 2B4 (eBioC1.7) (eBioscience). For intracellular cytokine assays, PBMCs were fixed in 2% formaldehyde after surface protein staining. After washing, PBMCs were permeabilised with 0.5% saponin, and the cells were then incubated with FITC–anti-IFN-γ (BD PharMingen) or PE–anti-T-bet (eBioscience). For CD107a analysis, PBMCs were cultured with IL-12 (10 ng/ml, R&D systems) and phycoerythrin-conjugated anti-CD107a (BD Bioscience) for 16 hr. Then, the cells were washed, blocked and stained with surface antibodies for ordinary FACS staining. For direct immunofluorescence staining of human whole blood, the appropriate volume of fluorochrome-conjugated monoclonal antibody was added to 100 µL of whole blood, which was then incubated in the dark at room temperature (20° to 25°C) for 30 min. Next, 2 mL of 1× RBC Lysis Buffer (Biolegend) was added, and the solutions were incubated for 10 min in the dark at room temperature. They were then centrifuged at 500× g for 5 min, and the supernatants were removed. After washing once, 0.5 mL of 1% paraformaldehyde was added to the solution, which was mixed thoroughly. The solutions were stored at 2 to 8°C until analysis. Appropriate isotype controls were used in all experiments to estimate background fluorescence. Stained cells were analysed using a FACSCalibur flow cytometer (Becton Dickinson), and the data were analysed using FlowJo analysis software 7.6.1 (Treestar). For the cell-cycle analysis, cells were seeded in 12- or 24-well plates at 60 to 70% confluency. After a 72-h incubation with patient serum, NK cells were harvested, washed in PBS and fixed overnight in 70% ethanol. The next day, the cells were collected, washed in PBS, incubated at 37°C for 10 min with PBS containing 100 mg/mL RNase A and 25 mg/mL propidium iodide (PI) (Sigma-Aldrich) and analysed using flow cytometry (BD Biosciences, San Jose, CA, USA). Data analysis was performed using ModFit LT software for Win32 3.1 (Verity Software House, PO Box 247, Topsham, ME 04086, USA).
NK cell populations were enriched from whole blood by negative selection using an NK Cell Isolation Kit (Miltenyi Biotec). A total of 1×106 human NK cells were cultured in 24-well plates at 37°C in a 5% CO2 incubator. The cells were incubated in medium alone or with TGF-β1 (1 ng/ml; R&D systems) and then with IL-15 (10 ng/ml; PeproTech) plus IL-2 (100 U/ml) for 72 h. In another culture model, NK cells were cultured in the presence of 20% healthy control serum, IT patient serum or IA/IN patient serum with or without anti-TGF-β-neutralising Ab (10 µg/ml; R&D systems) for 72 h. For the detection of T-bet expression, PBMCs were cultured in the presence of IFN-γ (10 ng/ml, R&D systems) and IL-12 (5 ng/ml, R&D systems) for 0, 12, or 72 h. For intracellular IFN-γ detection, PBMCs were cultured with RPMI 1640 supplemented with 10% FBS in the presence of IL-12 (10 ng/ml, R&D systems) for 12 h. And then monensin (10 µg/ml, Sigma) was added to prevent the secretion of the induced cytokines for 4 h. Transfection of NK92 cells was performed using an Amaxa Cell Line Nucleofector Kit R VCA-1001(Lonza Swiss). The transfection procedure was performed according to the manufacturer's instructions. Unless otherwise indicated, cells were transfected with 200 nM siRNAs. siRNA duplex homologous in sequence to HCST-119,214,356 and SH2D1A-459,501,674, as well as scrambled negative control constructs, were synthesised and purified by Shanghai Gene-Pharma Co. (Shanghai, China). The primers been used are as follows: HCST-homo-119: (5′- UCC AUG UGG GUC ACA UCC UTT AGG AUG UGA CCC AGA UGG ATT-3′); HCST-homo-214: (5′- GGC ACU UCA GGC UCU UGU UTT AAC AAG AGC CUG AAG UGC CTT-3′); HCST-homo-356: (5′- GGC AAA GUC UAC AUC AAC ATT UGU UGA UGU AGA CUU UGC CTT-3′); SH2D1A-homo-459: (5′- AGG CGU GUA CUG CCU AUG UTT ACA UAG GCA GUA CAC GCC UTT-3′); SH2D1A-homo-501: (5′- CAC GGU UAC AUU UAU ACA UTT AUG UAU AAA UGU AAC CGU GTT-3′) and SH2D1A-homo-674: (5′- GUC CUC AGC UAG AAG UAC ATT UGU ACU UCU AGC UGA GGA CTT-3′).
Serum cytokine concentrations were analysed using a CBA Inflammation Kit (BD Biosciences) according to the manufacturer's protocols. In brief, 50 µl of patient serum or a standard recombinant protein dilution was added to a mixture of capture beads coated with mAb to a group of cytokines (IL-1α, IL-1β, IL-2, IL-4, IL-6, IL-10, IL-12p70, IL-13, TNF and IFN-γ) and a PE-conjugated detection reagent. After 3 h, the capture beads were washed and acquired by a flow cytometer (BD Biosciences, San Jose, CA, USA). Using the recombinant standards and FlowJo analysis software 7.6.1 (Treestar), cytokine concentrations were quantified for each serum sample. Serum TGF-β1 was assayed using a standard sandwich ELISA Kit (CSB-E04725h) (Cusabio), in which 80 µl of patient serum was analysed according to the manufacturer's high sensitivity protocol. To activate latent TGF-β1 to the immuno-reactive form, serum was incubated with 1N HCL for 10 min at room temperature.
Total RNA was prepared from NK cells from healthy controls and HBV patients using TRIzol Reagent (Invitrogen, CA) according to the manufacturer's instructions. Cellular RNA was used for cDNA synthesis. Semi-quantitative real-time RT-PCR was performed using a SYBR Premix Ex Taq Perfect Real Time Kit (TaKaRa) and a sequence detector (Rotor Gene 3000, Corbett Research). The oligonucleotide primers used to amplify DAP10 were DAP10-F (5′-GGC TGC AGC TCA GAC GAC-3′) and DAP10-R (5′- AGG AGC GGC AGA GAG AGG-3′). The primers used to amplify SAP were SAP-F (5′-GAC GCA GTG GCT GTG TAT-3′) and SAP-R (5′-TCA TGG GGC TTT CAT TTC AGG CAG ACA TCA GG-3′). The primers used to amplify EAT-2 were EAT-2-F (5′-AGA CAG CGA GTC GAT ACC AG-3′) and EAT-2-R (5′-CCG TGT TTC TCT CTG AAG ATT CG-3′). A negative control without cDNA template was performed with every assay. The transcript copy number per subject was calculated by normalisation to GAPDH expression.
Primary NK cells (5×104 cells) were allowed to adhere to poly-L-lysine-coated slides before fixation in 4% paraformaldehyde for 20 min. Slides were then washed with TBST buffer (10 mM Tris-HCl, 0.1% Tween-20, 150 mM NaCl, pH 7.5) three times for 5 min each and permeabilised with 0.5% saponin (Sigma-Aldrich). After blocking with 1% bovine serum albumin (BSA) in TBS for 1 h, the slides were then washed three times in TBST for 5 min and incubated with a primary antibody (anti-DAP10 (FL-93) sc-25623; anti-SAP (C-18) sc-8639, Santa Cruz Biotechnology) diluted 1∶100 for 1 h. Afterwards, the slides were washed three times again in TBST for 5 min, incubated for 30 min with the secondary antibody (donkey anti-goat IgG-FITC, sc-2024 or goat anti-rabbit IgG-FITC, sc-2012, Santa Cruz Biotechnology) at a dilution of 1∶1,000 and stained with DAPI (Santa Cruz Biotechnology) for 1 min. A final wash in PBS was performed, and the slides were then mounted on coverslips in VECTASHIELD (Vector Laboratories) anti-fade solution. All procedures were performed at room temperature. Images were acquired on a Zeiss 510 Meta multi-photon confocal microscope (Zeiss, Oberkochen, Germany). Analysis was performed using LSM510 META software (Zeiss) and a JD801 Morphological Image Analysis System (JSJD Tech Inc., China). The analysis was performed on five confocal images from 200 cells per experiment in images using channels that showed the co-stained proteins of interest.
NK cells were resuspended (1×106 cells/mL) in culture medium containing 0.2 M Fluo-4 AM (Invitrogen, USA) and 1 M Fura Red AM (Invitrogen, USA) and incubated at 37°C for 30 min. The cells were then washed once with HBSS supplemented with 1% FBS and resuspended at 1×106 cells/mL in 0.5 mL HBSS plus 1% FBS. They were then added to poly-L-lysine-coated (Sigma) wells attached to Lab-Tek II chambered coverslips (Thermo Fisher Scientific Inc.). The NK cells were allowed to settle onto the coverslips at 37°C for 15 min. Then, the medium was replaced with 300 µL cold HBSS plus 1% FBS containing 3 g each of the mAbs specific for NK receptors. The cells were incubated on ice for 30 min. The cells were then washed once in 1 mL HBSS plus 1% FBS, and 300 µL cold HBSS with 1% FBS was placed into each chamber. The chamber was warmed in a 37°C, 5% CO2 incubator for 5 min. Data were captured on a Zeiss LSM510 confocal microscope using a 20× objective (Zeiss, Oberkochen, Germany). The microscope was set to scan mode to acquire wavelengths from 499 to 670 nm in a single pass, and the scan speed was set to ensure that one scan was completed every 4 s. Sixty seconds after the beginning of the acquisition process, secondary cross-linking goat anti-mouse F(ab')2 antibody (Jackson Immuno Research, West Grove, PA) was added at a final concentration of 13 µg/mL. Scanning continued for 10 min. The data were analysed by linear unmixing, using control spectra acquired from cells loaded with either Fluo-4 or Fura-Red. After unmixing, the cells were processed to yield ratios of Fluo-4 fluorescence to Fura Red fluorescence for the entire time series.
The cytotoxicity of primary NK cells or NK92 cells against target cells was measured using a standard 4-h 51Cr release assay, as previously described [75]. Primary NK cells or NK92 cells were used as effector cells. 51Cr-labelled K562 cells (an NK cell–sensitive erythroleukaemia cell line) were used as target cells at effector-to-target cell (E∶T) ratios of 20∶1, 10∶1, and 5∶1. Target cells (1×106) were labelled with 200 µCi sodium [51Cr] chromate (PerkinElmer) for 1 h at 37°C and then washed 3–4 times with PBS. Individual assays were performed in triplicate with 1×104 target cells in 96-well V-bottom plates. Maximum 51Cr release was determined by incubating target cells with 2% Triton X-100. For spontaneous 51Cr release, the targets were incubated without effector cells in assay medium alone. All samples were assayed in triplicate. Plates were briefly spun down and incubated at 37°C for 4 h. After 4 h, the cells were pelleted, and 100 µl of supernatant was counted by a gamma counter. The percentage of specific 51Cr release was calculated using the formula ([experimental release-spontaneous release]/[maximum release-spontaneous release])×100%. Spontaneous release did not exceed 10% of the maximum release in all experiments.
After NK cell purification using MACS (magnetic-activated cell sorting), NK cell lysates were separated by SDS-PAGE under non-reducing conditions on a 12% polyacrylamide gel. After electrophoresis, the proteins were transferred onto PVDF membranes (Millipore) by electroblotting using vertical buffer tanks. The membranes were blocked with blocking buffer (5% non-fat dry milk in wash buffer (50 mM Tris-HCl, pH 7.4, 0.9% NaCl, and 0.1% Tween 20)) overnight at 4°C and then incubated with the primary antibodies (at the recommended dilutions) for 1.5 h at room temperature. The membranes were then washed with wash buffer six times and incubated with HRP-conjugated secondary antibodies for 1 h. After washing six times with wash buffer, protein bands were visualised using an enhanced chemiluminescent system (Pierce). The primary antibodies used [(Smad2 (D43B4), Phospho-Smad2 (Ser465/476), p15INK4B and p21Waf1/Cip1 (DCS60)] were all obtained from Cell Signaling Technology.
The data are expressed as the mean ± SEM. One-way ANOVA was used to compare the differences among three or more groups, followed by the Bonferroni post hoc test. The analysis was completed using SPSS for Windows (version 10.1; SPSS). In all analyses, P-values<0.05 were considered statistically significant.
NKG2D, P26718 (NKG2D_HUMAN); CD244, Q9BZW8 (CD244_HUMAN); DAP10, Q9UBK5 (HCST_HUMAN); SAP, O60880 (SH21A_HUMAN); EAT-2, O14796 (SH21B_HUMAN); Smad2, Q15796 (SMAD2_HUMAN); P15, P42772 (CDN2B_HUMAN); P21, P38936 (CDN1A_HUMAN).
|
10.1371/journal.pgen.1002361 | Short Day–Mediated Cessation of Growth Requires the Downregulation of AINTEGUMENTALIKE1 Transcription Factor in Hybrid Aspen | Day length is a key environmental cue regulating the timing of major developmental transitions in plants. For example, in perennial plants such as the long-lived trees of the boreal forest, exposure to short days (SD) leads to the termination of meristem activity and bud set (referred to as growth cessation). The mechanism underlying SD–mediated induction of growth cessation is poorly understood. Here we show that the AIL1-AIL4 (AINTEGUMENTALIKE) transcription factors of the AP2 family are the downstream targets of the SD signal in the regulation of growth cessation response in hybrid aspen trees. AIL1 is expressed in the shoot apical meristem and leaf primordia, and exposure to SD signal downregulates AIL1 expression. Downregulation of AIL gene expression by SDs is altered in transgenic hybrid aspen plants that are defective in SD perception and/or response, e.g. PHYA or FT overexpressors. Importantly, SD–mediated regulation of growth cessation response is also affected by overexpression or downregulation of AIL gene expression. AIL1 protein can interact with the promoter of the key cell cycle genes, e.g. CYCD3.2, and downregulation of the expression of D-type cyclins after SD treatment is prevented by AIL1 overexpression. These data reveal that execution of SD–mediated growth cessation response requires the downregulation of AIL gene expression. Thus, while early acting components like PHYA and the CO/FT regulon are conserved in day-length regulation of flowering time and growth cessation between annual and perennial plants, signaling pathways downstream of SD perception diverge, with AIL transcription factors being novel targets of the CO/FT regulon connecting the perception of SD signal to the regulation of meristem activity.
| Day length is a critical and robust environmental cue utilised by plants to modulate their patterns of growth to adapt to changing seasonal conditions. In perennial plants such as long-lived trees of the boreal forest, reduction in day length (short day signal/SD) induces the cessation of growth prior to the advent of winter. This developmental switch is of major adaptive significance as inability to undergo growth cessation leads to mortality in these plants. Our knowledge of how SD signal induces growth cessation is rudimentary. Here we show that AIL1 (AINTEGUMENTALIKE 1), a plant-specific transcription factor, is a downstream target of the SD signal and can regulate the expression of key cell proliferation related genes. Intriguingly, the early acting components in day length–regulated processes such as flowering and growth cessation are conserved between annual and perennial plants. However our results show that the pathways downstream of short day perception diverge between these day length–controlled developmental transitions. These results have important implications for the evolution of the perennial life cycle and demonstrate how the same signal, namely day length, can regulate diverse developmental switches in annual and perennial plants.
| The ability to adapt to changes in the environment is crucial to the survival of both animals and plants. Plants, unlike animals, are sessile organisms and have therefore evolved highly sophisticated mechanisms to anticipate seasonal changes and modulate their patterns of growth and development. Day length is one of the key environmental cues utilised by plants to anticipate seasonal changes and regulates several key developmental transitions associated with plant adaptation and reproduction. One of the most fascinating examples of this is provided by perennial plants, e.g. the long-lived trees of the boreal forest, in which the day length signal regulates the developmental transition from active growth to a more resilient dormant state prior to the onset of winter [1]. These perennial plants anticipate the approach of winter by detecting the reduction in day length (i.e. the short day signal, or SD signal) in the autumn and when the day length falls below the critical day length required for the promotion of growth, cell division in the meristems ceases [2]. The most visible indicator of short day–induced growth cessation is the formation of a bud that encloses the apical meristem and leaf primordia [3]. The importance of day length sensing for the survival of perennial plants is illustrated by the increased mortality due to delayed growth cessation in transgenic hybrid aspen plants that are unable to sense reductions in day length [4].
Intriguingly, there are numerous similarities at the regulatory level between day length mediated control of growth cessation in perennial plants and one of the most well studied developmental transitions in plants - the transition from vegetative growth to floral development. For example, key flowering time regulators such as the CONSTANS (CO), FLOWERING LOCUS T (FT) and the group of photoreceptors known as PHYTOCHROMES (PHYs) that are involved in day length mediated regulation of flowering time regulation in Arabidopsis [5], [6], [7], [8], are all also involved in SD–induced growth cessation in trees [4], [9], [10], [11]. In poplar species two closely related orthologs of FT (FT1 and FT2) have been found and recent analysis in hybrid aspen clone 353 indicates that FT2 could be primarily involved in SD–mediated growth cessation whereas FT1 is primarily involved in flowering [11]. In hybrid aspen (clone T89, used in this study), it has been shown that short day mediated downregulation of FT gene expression (FT1 and FT2) triggers the induction of growth cessation whereas overexpression of FT1 eliminates the plants' ability to respond to the SD signal and thus prevents timely growth cessation [9]. Both CO and PHYTOCHROME A (PHYA) act upstream of the FT genes in SD-mediated induction of growth cessation in much the same manner as in the flowering transition in Arabidopsis [9]. Subjecting aspen trees to conditions in which the peak of CO expression occurs in the dark (e.g. under SD conditions) brings about rapid downregulation of FT2 expression leading to the induction of growth cessation response [9]. These findings indicate evolutionary conservation of the day length response pathway between annual plants such as Arabidopsis and perennial trees such as hybrid aspen.
Despite the evolutionary conservation of early acting components involved in day length regulated growth cessation and flowering time, considerable lacunae remain in our understanding of SD mediated regulation of growth cessation at the molecular level. Particularly, the factors targeted by SD signal downstream of the early acting components such as PHYA and the CO/FT regulon in regulating growth cessation responses remain unknown. The critical role of these hitherto unknown downstream targets of SD signal has become evident from the analysis of growth cessation response in hybrid poplar where they have been shown to be important for regulating the variation in timing of growth cessation responses [12]. Thus a key question that remains unanswered is; How does SD mediated downregulation of FT2 expression lead to the induction of growth cessation response? Answering this question would require the identification of targets of SD signal downstream of the CO/FT regulon in trees and elucidating their role in the regulation of growth cessation responses. The network of genes involved in day length regulation of the floral transition is well defined, and downstream targets of the CO/FT pathway like SUPPRESSOR OF OVEREXPRESSION OF CONSTANS (SOC1) and floral meristem identity genes FRUITFUL (FUL) and APETALA1 (AP1) are known (reviewed in [13]). In contrast the targets of the SD signal downstream of the CO/FT module in the growth cessation response can not simply be deduced from extrapolation of knowledge of floral transition related genes given the difference between the growth cessation process and floral transition.
To identify the downstream targets of the SD signal in growth cessation we have previously analysed global transcriptional changes associated with this process [14], [15]. One of the genes whose transcript is strongly downregulated during growth cessation is a Populus homolog of the Arabidopsis gene AINTEGUMENTA (ANT)[16]; the Populus homolog is henceforth referred to as AINTEGUMENTALIKE1 (AIL1). The expression data along with a proposed role of ANT in the regulation of cell cycle [17], [18], [19] suggested that AIL1 (and most likely the other closely related members AIL2-AIL4 of this sub-family) is a potential downstream target of the SD signal transduced via the CO/FT module in the regulation of growth cessation response in perennial trees. We tested this hypothesis by investigating the regulation of AIL1 expression by SD signal in transgenic hybrid aspen plants that are perturbed in the SD response. In a complementary approach we investigated the short day mediated regulation of growth cessation response in transgenic hybrid aspen plants that either maintain high levels of AIL1 or AIL3 expression even after SD treatment or have reduced expression of AIL1. Finally we identified downstream targets of the AIL1 transcription factor in the apex. Taken together, these analyses show that AIL genes are targets of the SD signal downstream of the CO/FT module and their down-regulation is necessary for short day regulated growth cessation in hybrid aspen plants.
The Populus genome contains 13 genes belonging to the ANT-subgroup of the AP2 transcription factor family [20]. Four of these genes are here designated as AIL1-AIL4 (AINTEGUMENTALIKE 1-4) as they belong to the same clade as the Arabidopsis ANT transcription factor (Figure S1). We investigated the expression of AIL1 as well as the expression of the related genes AIL2-AIL4 in the apex of hybrid aspen plants after SD treatment (Figure 1A and Figure S2). RT-PCR data indicates that AIL1 (Figure 1A) as well as AIL2-AIL4 expression (Figure S2) are downregulated along with that of cell cycle markers CYCD3:2 and CYCD6:1 after SD treatment (Figure 1B, panels B and C) and this downregulation coincides with the cessation of growth and bud set in the apex of hybrid aspen T89 trees [21].
We then investigated the effect of perturbed SD perception or response on the regulation of AIL gene expression after SD treatment. For this we used transgenic hybrid aspen that are unable to respond to the SD signal due to overexpression of PHYA or FT1 cDNA as well as plants that are hypersensitive to the SD signal and undergo premature growth cessation due to the downregulation of FT expression (FTRNAi) [4], [9]. Since all 4 AIL genes are highly similar and displayed similar expression pattern after SD treatment, we chose to perform detailed analysis of AIL1 regulation. RT-PCR analysis indicated that the downregulation of AIL1 expression after SD treatment is severely attenuated in the apex of PHYA and FT1 overexpressors in contrast with the wild type (Figure 2A). In contrast the FTRNAi plants that respond more rapidly to SD treatment than wild type [9] display a stronger and earlier reduction in the expression of AIL1 (Figure 2B). These results strongly suggest that AIL1 expression is a potential downstream target of the SD signal transduced via the CO/FT module in cessation of growth and bud set in the apex of hybrid aspen.
We further investigated expression of AIL1 in different tissues and found that AIL1 is primarily expressed in the apical region of hybrid aspen (Figure 3A). Since the downregulation of AIL1 gene expression was closely associated with cessation of growth and bud set, we analysed the domain of AIL1 gene expression in the apex. For this analysis we generated transgenic hybrid aspen expressing a transcriptional fusion between 2.5 Kb upstream sequence of AIL1 gene from Populus trichocarpa and the uidA (b-glucoronidase/GUS) reporter gene. In the transgenic hybrid aspen, the reporter gene expression was mostly confined to the zone of dividing cells in the apex, the provascular tissues and the leaf primordia (Figure 3B). The expression pattern of pAIL1:UidA reporter construct correlates well with the previously described expression pattern of CYCA1 which serves as a marker of dividing cells in the apex of hybrid poplar plants [22] indicating that AIL1 expression is associated primarily with cell proliferation.
Analysis of AIL1 gene expression suggested that its downregulation could be important for the SD mediated cessation of growth and bud set. We tested this hypothesis by generating transgenic hybrid aspen that would maintain high levels of AIL1 expression even after SD treatment in contrast with the wild type by expressing AIL1 cDNA under the control of the 35S promoter (these transgenic lines are henceforth referred to as AIL1oe lines). Several independent lines were obtained and tested for high expression of AIL1 (Figure S3A shows data for two chosen lines); two lines were chosen for detailed analysis of their response to SD treatment. Unlike the wild type plants that undergo growth cessation and form an apical bud after 6 week of SD treatment, the apices of AIL1oe fail to undergo proper growth cessation and bud set after 6 weeks of SD treatment (Figure 4A–4C). We also generated transgenic hybrid aspen overexpressing AIL3 (Figure S3) to investigate whether other members of the gene family share the same function (these transgenic lines are henceforth referred to as AIL3oe lines). AIL3oe display a similar phenotype to AIL1oe during SD treatment (Figure 4D–4F).
We also investigated the effect of downregulation of AIL gene expression on SD mediated bud set. The functional redundancy between AIL genes suggested by similar regulation and effect on bud set in AIL1oe and AIL3oe plants lead us to generate transgenic hybrid aspen plants in which the expression of all 4 AIL genes was targeted for downregulation using artificial microRNA (amiRNA). Two amiRNA constructs (255 and 256) were expressed in hybrid aspen and two lines (255-6 and 256-23) with reduced expression of AIL1 gene expression (Figure S4) were selected for further analysis of growth cessation response. The transition from active growth to bud set after SD treatment in the wild type and lines 255-6 and 256-23 was investigated using a method separating different stages of bud development [23]. Compared to the wild type, the lines 255-6 and 256-23 displayed a more rapid transition from active growth to bud set and a majority of the plants in the two transgenic lines made the transition to intermediate and late stage of bud set at least a week (or more) earlier than the wild type plants after SD treatment (Figure S5). This data along with the perturbed growth cessation response in AIL1oe and AIL3oe plants indicate that downregulation of AIL gene expression is necessary for SD mediated growth cessation response.
The altered growth cessation responses in AIL1oe plants could either be due to the failure of these plants to perceive the short day signal or to a failure in properly responding to it. To distinguish between these two possibilities we compared the response of FT2 expression to SD treatment in the leaves of wild type and AIL1oe. Downregulation of FT2 expression in the leaves is the earliest known marker for the detection of the SD signal and recent results have implicated its downregulation in SD mediated growth cessation [9], [11], [12]. Our data show that both the wild type (Figure 5A) and AIL1oe lines (Figure 5B) exhibit similar decreases in their levels of FT2 transcripts following SD treatment. This result indicates that unlike FT genes, AIL1 does not act early in SD response but is rather a downstream target of the SD signal.
The expression of several cell proliferation related genes, e. g. D-type cyclins, that are key cell cycle regulators [24], [25], [26], [27] is downregulated in a similar manner to AIL genes during SD mediated cessation of growth in hybrid aspen [15], Figure 1B, 1C). We therefore investigated whether the AIL1 transcription factor could be involved in the regulation of the D-type cyclin genes and if so whether their expression is perturbed in AIL1oe lines after SD treatment. Therefore we analysed the expression of two D-type cyclins, CYCD3:2 and CYCD6:1 after SD treatment in AIL1oe plants after 6 weeks of SD treatment. Our RT-PCR data (Figure 6) showed that while the expression of CYCD3:2 and CYCD6:1 is downregulated in the wild type after 6 weeks of SD treatment, this was not the case in the AIL1oe plants. This result indicates that AIL1 could be involved in the regulation of D-type cyclins and the failure to downregulate AIL1 expression after SD treatment leads to a corresponding failure to downregulate the expression of these key cell cycle regulators.
The observation that CYCD3:2 and CYCD6:1 expression after SD treatment is perturbed in AIL1oe plants prompted us to investigate whether the AIL1 transcription factor can interact with CYCD promoters from hybrid aspen using electrophoretic mobility shift assays (EMSA). We expressed HA-tagged AIL1 protein in Arabidopsis protoplasts and used the extracts in gel shift assays using 3 different different fragments from a hybrid aspen CYCD3:2 promoter (results from two fragments are shown here). Our data show that extracts containing AIL1 protein specifically display a gel shift with the promoter fragment consisting of 200 bp of sequence situated upstream of the start codon of CYCD3:2 (Figure 7). Together with the CYCD3:2 and CYCD6:1 gene expression data, the gel-shift analysis strongly suggests these cyclin genes might be potential downstream targets of the AIL1 transcription factor in hybrid aspen.
Short day mediated cessation of growth and budset prior to the onset of winter is a key developmental transition that is critical to the survival of perennial plants in boreal forest [1]. In this work, we identify AIL genes belonging to the AP2 transcription factor family as downstream targets of the SD signal transduced via the CO/FT module and that downregulation of their expression is necessary for cessation of growth and bud set in hybrid aspen.
In poplar, there are 4 closely related AIL genes and our data indicates that all four AIL genes could have similar function at least in SD mediated growth cessation response as suggested by the similar phenotypes of AIL1 and AIL3 overexpressing plants as well as in plants in which these genes are targeted for downregulation. However we cannot exclude the possibility that there could be functional differences between the different AIL genes with respect to other biological processes as we have not been able to specifically downregulate individual genes of the AIL family and study the effect on growth and development so far. It is particularly important to note in this respect that even closely related genes can diverge both in expression profiles and at a functional level as suggested by the careful analysis of FT1 and FT2 in poplar species which indicate that FT1 could be primarily involved in reproductive growth whereas FT2 controls growth cessation [11].
Several observations suggest that the downregulation of AIL gene expression following SD treatment is necessary for the activation of growth cessation responses. AIL1 (and most likely other genes of this family as well) is primarily expressed in dividing and meristematic cells in hybrid aspen (Figure 3) and the downregulation of their expression coincides temporally with the SD-mediated induction of growth cessation responses in hybrid aspen (Figure 1 and Figure 4), including the termination of elongation growth, bud set and the downregulation of core cell cycle genes such as the D-type cyclins (Figure 1). Furthermore, AIL1 downregulation after SD treatment is attenuated in FT or PHYA overexpressors (Figure 2) that fail to respond properly to SD treatment [4], [9]. Importantly, growth cessation response is perturbed when AIL1 or AIL3 expression is maintained at high levels even after SD treatment and earlier bud set is observed in transgenic hybrid aspen with reduced expression of AIL1 (Figure 4 and Figure S5). All of these results are consistent with the AIL genes being downstream targets of the SD signal in the control of the growth cessation response.
Three hypotheses can be proposed to explain the role of AIL genes in SD mediated control of growth cessation response and why growth cessation response is perturbed when the AIL1 or AIL3 expression is maintained at high level even after SD treatment as in AIL1oe and AIL3oe lines. Firstly, the AIL genes could act upstream of FT2, in which case the increased expression of AIL1 as in AIL1oe could counteract the downregulation of FT2 by the SD signal. However, this hypothesis is incompatible with the observation that the downregulation of FT2 subsequent to SD treatment proceeds as normal in AIL1oe lines (Figure 5). Alternatively, the AIL genes could act independently of FT2 and their increased expression in AIL1oe and AIL3oe could prevent the downregulation of the targets of FT2 following SD treatment. Alternatively, the AIL genes are the targets of SD signal downstream of CO/FT regulon leading to their downregulation after SD treatment. Our results support the latter hypothesis because SD treatment results in the downregulation of the AIL gene expression and this downregulation of AIL gene expression is severely attenuated in plants that overexpress FT1. Further evidence for the connection between the CO/FT regulon and AIL1 expression was obtained by analysis of FTRNAi lines that respond more rapidly than the wild type to SDs and in these lines, AIL1 expression is significantly reduced compared to wild type after 2 weeks of SD treatment (Figure 2B). Finally the downregulation of the AIL1 expression leads to earlier transition from active growth to bud set strongly suggesting that the AIL genes are the downstream targets of the SD signal. Thus our results suggest a mechanism in which AIL genes act downstream of the CO/FT regulon and that downregulation of AIL gene expression culminates in growth cessation and bud set after SD treatment.
FT has been shown to act as a transcriptional co-regulator in Arabidopsis [28]. In poplar species, two FT genes are present of which FT2 is rapidly downregulated after SD treatment; thus FT2 could either directly regulate AIL at the transcriptional level in hybrid aspen or alternatively downstream targets of FT2 could regulate AIL gene expression. Our data supports the latter suggestion because the kinetics of downregulation of FT2 and AIL gene expression subsequent to SD treatment is not consistent with direct regulation of AIL gene expression by FT2. While FT2 is typically downregulated within 3–7 days in the leaves after the commencement of SD treatment ([9], [12], Figure 5), it takes 2–3 weeks until downregulation of the AIL genes becomes apparent in the apex (Figure 1). Moreover, induction of FT2 in the leaves has little effect on expression of most of AIL genes [11], which might again suggest an indirect regulation of AIL expression by FT2. Thus these results suggest that there may be one or more genes that are direct targets of FT2 and act upstream of the AIL genes regulating their expression in the apex. Determining the identity of these targets of FT2 and regulators of AIL expression in the apex is an important objective for future research in this area. The downstream targets of FT in daylength mediated regulation of flowering time such as SOC1 and the floral meristem identity genes FUL and AP1 [13] are well known. However these are unlikely to be the targets of the CO/FT regulon in the regulation of the AIL genes in the apex unless the tree homologs of these genes have acquired novel functions and have been recruited to regulate meristem activity by controlling AIL gene expression.
AIL1 is expressed in dividing cells (Figure 2), can potentially interact with the promoters of D-type cyclins (Figure 7) and maintaining high level of AIL expression prevents the downregulation of D-type cyclin expression after SD treatment (Figure 6) suggesting that AIL1 has a role in regulation of key cell cycle regulators. Indeed, data from Arabidopsis also shows that the putative AIL1 ortholog ANT can positively regulate cell division as its overexpression leads to increased duration of cell division [17]. We therefore propose that the downregulation of AIL gene expression after SD treatment leads to the downregulation of a subset of D-type cyclins such as CYCD3:2 and CYCD6:1. The downregulation of the expression of core cell cycle regulators such as the abovementioned cyclins would then culminate in cessation of growth and bud set. However it is unlikely that the D-type cyclins are the only targets of AIL1 because the expression of several other cell cycle genes is also downregulated after SD treatment [15]. Additionally transcriptional network analysis indicates that several other cell cycle genes might be regulated by the AIL1 transcription factor [29]. Moreover, preliminary investigations suggest that altering CYCD3:2 expression alone is not sufficient to activate the growth cessation response.
Substantial progress has recently been made in understanding how SD signal is perceived, and downregulation of FT2 expression after SD treatment has been identified as a key early event in the induction of growth cessation response [9], [11]. However, the components targeted by SD signal downstream of the CO/FT regulon in the induction of growth cessation response have remained elusive, especially factors that would link the downregulation of FT2 expression to cessation of growth. Indeed analyses of hybrid poplar clones that differ in timing of bud set have suggested an important role for such factors in differential growth cessation [12]. Our finding that the AIL genes are the targets of the SD signal that is transduced via the CO/FT module in growth cessation response and bud set therefore represents an important step in elucidating the mechanism underlying this key developmental transition in perennial plants as this links the CO/FT module to the regulation of cell cycle through the AIL genes in SD mediated cessation of growth and bud set.
The CO/FT module is an important component of the molecular machinery that allows plants to respond to changes in day length, and its role in day length mediated control of flowering time is well established [30]. Therefore it was not surprising that the same CO/FT module is also involved in controlling the timing of SD-mediated growth cessation in perennial trees, as this is another key developmental transition that is regulated by the day length signal. However, given that flowering and growth cessation processes are distinct morphologically, it appears unlikely that the downstream targets of this module in the regulation of flowering would be the same as those involved in the growth cessation response. Our findings suggest that the AIL transcription factors, which have the potential to regulate the expression of cell cycle genes, were co-opted at some point in evolutionary history to serve as mediators of the day length signal. This co-option would have allowed the versatile CO/FT module to regulate a novel developmental transition. These results demonstrate that an evolutionary “mix and match” strategy involving combining different regulatory modules can allow a small number of regulatory modules to control a wide range of diverse biological processes. In conclusion, our data demonstrates the divergence of the regulatory pathway downstream of the conserved CO/FT module between day length controlled floral transition and growth cessation response and identifies AIL1 as a potential regulator of cell cycle related genes and a novel target of the short day signal downstream of the CO/FT module in regulation of growth cessation in perennial trees.
Cuttings of hybrid aspen (Populus tremula x tremuloides) clone T89 (wild type) and the transgenic lines were grown in half-strength Murashige/Skoog medium (½ MS) under sterile conditions for approximately 4 weeks and then transferred to soil. After four weeks in greenhouse the plants were moved to growth chambers (18 hour light/6 hour night, 20°C). After one week the chamber settings were shifted to short day conditions (8 hour light/16 hour night, 20°C or 14 hour light/10 hour night, 20°C). Growth cessation was determined by measurement of elongation growth and/or bud set. Pictures of apices to assess bud formation were taken using Canon EOS digital camera. For tissue specific expression analysis of AIL genes, samples were taken from tissue culture grown plants 4 weeks after cuttings were transferred to new media.
AIL-genes were identified by blasting the Arabidopsis AINTEGUMENTA gene (AT4G37750) against the Populus genome. Gene models (http://genome.jgi-psf.org/Poptr1_1/Poptr1_1.home.html) were manually chosen based on intron-exon structure (JGI protein ID for each model can be found in Figure S1). Sequences were aligned and a bootstrapped phylogenetic tree generated using ClustalX [31]. The phylogenetic tree was visualised using TreeView (http://darwin.zoology.gla.ac.uk/~rpage/treeviewx/).
The full length cDNA for AIL1 transcription factor was cloned into the donor vector pDONR201 (Invitrogen.com) before transfer into the destination vector pK2GW7 [32]. The resulting vectors were introduced into agrobacterium GV3101pmp90RK [33] followed by the transformation of hybrid aspen clone T89 [34]. The same strategy was used to generate AIL3oe lines with the exception of entry clone construction that in this case was performed using the pENTR/D-TOPO cloning kit (Invitrogen.com).
The AIL1 promoter was amplified using the primers: FW: CACCCGGGGAATGATAGGCTGACAA and RP:CCCAAAATCTTGCCTACTTCC and cloned into the pENTR/D-TOPO vector (Invitrogen.com). The fragment was transferred into the pK2GWFS7 binary vector [32]. The construct was transformed into hybrid aspen using Agrobacterium mediated transformation as described before [34]. Apices from transgenic lines expressing the reporter gene were collected from greenhouse grown trees approx. 5 weeks after potting. The apices were incubated approx. 3 h at 37°C in GUS-solution (1 mm X-gluc, 1 mm K3Fe(CN)6, 1 mm K4Fe(CN)6, 50 mm sodium phosphate buffer (pH 7.0), and 0.1% (v/v) Triton X-100). The samples were then rinsed with water, dehydrated to 50% (v/v) ethanol, fixed for 10 min in FAA (5% (v/v) formaldehyde, 5% (v/v) acetic acid, and 50% (v/v) ethanol), and cleared in 100% (v/v) ethanol. Once cleared, the samples were embedded in LR-White/10% PEG 400 resin in polypropylene capsules (TAAB) The apices were then sectioned on a Microm HM350 microtome (Microm International GmbH, Germany) at approx 20 µm, floated on water, heat-fixed to glass slides, mounted in Entellan neu (Merck, Germany) Sections were visualized with Zeiss Axioplan light microscope and captured with a digital camera, AxioCam together with the Axiovision 4.5 software (Zeiss, Germany).
Total RNA from poplar apices was extracted using the Aurum Total RNA kit (Bio-Rad). Care was taken to collect tissue samples for RNA isolation at the same time of the day (usually between 13–16 PM) for each experiment. 100–500 ng of RNA was DNase treated with RNase free DNaseI (Fermenta) and used for cDNA synthesis using iScript cDNA synthesis kit (BioRad) or qScript cDNA synthesis kit (Quanta BioSciences). Reference genes were validated using GeNorm Software [35]. The reference gene chosen was UBQ in all experiments except for the analysis of the overexpression of AIL1 and AIL3 in the AIL1oe and AIL3oe lines, where 18S rRNA was used as the reference gene. Analysis of expression in FTRNAi used two reference genes, UBQ and TIP-41 like. SYBR green (Bio-Rad or Quanta BioSciences) was used as non-specific probe in all reactions and relative expression values were calculated using the Δ-ct-method [9]. A complete list of primers used in RT-PCR analysis can be found in Table S1.
To downregulate the expression of AIL genes, artificial microRNAs were designed using the online tools at http://wmd.weigelworld.org/cgi-bin/mirnatools. Briefly primers (Table S3) were used to generate artificial microRNAs directed against all the 4 AIL genes and cloned into the plant transformation vector pK2GW7 according to the cloning protocol at http://wmd3.weigelworld.org/. Two different miRNA contructs (named 255 and 256) were made and transformed into hybrid aspen clone T89 as described earlier. Following transformation several hybrid aspen lines with reduced expression of AIL1 were obtained and one line each for the two constructs 255 and 256 were selected for the analysis of bud set after SD treatment (lines 255-6 and 256-23).
Bud set was scored using the method described by [23]. We used a score of 3 to indicate active growth (complete lack of bud set) and 0 to indicate a completely closed bud and score of 2 or 1 to indicate intermediate stages. For this analysis, bud set was scored every 7 days in a minimum of 5 or more plants for a period of 7 weeks.
AIL1 full length cDNA was amplified using the following primers: pttAIL1(EcoRI) FW- CATGGAATTCATGAAATCTACGGGTGATAA and pttAIL1(SalI) RP-CATGGTCGACTTCTCCTTTTCCTTGGTTCATGC. The resulting fragments were cloned into pRT104-3xHA [36]. Transfection into Arabidopsis protoplasts were performed as described [36], [37] using 8 µg of purified plasmid. Cells were lysed in a lysis buffer containing 25 mM Tris-HCL (pH 7.5), 50 mM KCl, 1 mM EDTA, 10% Glycerol 1 mM DTT, 0.1% Igepal and 1X PIC (Protease Inhibitor Cocktail). After centrifugation the supernatant was collected and immediately frozen in liquid nitrogen. The expression of the HA-tagged AIL1 protein was confirmed with western blot and resulting cell extracts were used for subsequent analysis.
CycD3:2 promoter sequences were identified using the JGI populus genome database (http://genome.jgi-psf.org/Poptr1_1/Poptr1_1.home.html). Approx. 200 base pair fragments were amplified using primers specified in Table S2. The fragments were gel-purified using E.Z.N.A. Gel Purification Kit (Omega Bio-Tek) followed by phenol-chloroform extraction and ethanol precipitation prior to use in gel-shift assays. Five pmol of purified fragments were biotin labeled using the Biotin 3′ End DNA Labeling Kit (Pierce). Labeling and labeling efficiency determination was performed according to the manufacturers recommendation.
The biotin-labelled promoter fragments were mixed with protoplast cell extracts containing AIL1-HA or control extracts from non-transfected protoplasts. For the binding reaction the following conditions were used: 10 µl protoplast cell extract, 0.5 µl biotin-labelled DNA (10 fmol/µl), 0.4 µl non-specific competitor (poly (dI:dC), 1 mg/ml), 0.5 µl BSA (20 mg/ml) and lysis buffer to a total of 20 µl. For specific competition, 500 fmol non-labelled fragment was added to the reaction. Binding was performed on ice for 10 min followed by 30 min in room temperature. The samples were run on a non-denaturing polyacrylamide gel (5%-0.5xTBE) and transferred to a Hybond N+ membrane (GE Healthcare, Sweden). Crosslinking and detection was performed using the LightShift Chemiluminescent EMSA kit (Pierce.com).
|
10.1371/journal.ppat.1004674 | Molecular and Functional Analyses of a Maize Autoactive NB-LRR Protein Identify Precise Structural Requirements for Activity | Plant disease resistance is often mediated by nucleotide binding-leucine rich repeat (NLR) proteins which remain auto-inhibited until recognition of specific pathogen-derived molecules causes their activation, triggering a rapid, localized cell death called a hypersensitive response (HR). Three domains are recognized in one of the major classes of NLR proteins: a coiled-coil (CC), a nucleotide binding (NB-ARC) and a leucine rich repeat (LRR) domains. The maize NLR gene Rp1-D21 derives from an intergenic recombination event between two NLR genes, Rp1-D and Rp1-dp2 and confers an autoactive HR. We report systematic structural and functional analyses of Rp1 proteins in maize and N. benthamiana to characterize the molecular mechanism of NLR activation/auto-inhibition. We derive a model comprising the following three main features: Rp1 proteins appear to self-associate to become competent for activity. The CC domain is signaling-competent and is sufficient to induce HR. This can be suppressed by the NB-ARC domain through direct interaction. In autoactive proteins, the interaction of the LRR domain with the NB-ARC domain causes de-repression and thus disrupts the inhibition of HR. Further, we identify specific amino acids and combinations thereof that are important for the auto-inhibition/activity of Rp1 proteins. We also provide evidence for the function of MHD2, a previously uncharacterized, though widely conserved NLR motif. This work reports several novel insights into the precise structural requirement for NLR function and informs efforts towards utilizing these proteins for engineering disease resistance.
| The plant hypersensitive defense response (HR) is a rapid, localized cell death, usually occurring upon the recognition of specific pathogen-encoded molecules and consequent activation of nucleotide binding-leucine rich repeat (NLR) proteins. Rp1-D21, a naturally-occurring mutant caused by the recombination of two NLR genes, confers a ‘lesion mimic’, HR-like phenotype in the absence of pathogen infection and provides a powerful tool to investigate the molecular mechanisms of NLR regulation. Here we report the results of a genetic screen in maize that identified novel mutations abrogating Rp1-D21-induced HR. To characterize the function of Rp1-D21, we transiently expressed Rp1-D21 and various derivatives in Nicotiana benthamiana to observe the resulting levels of HR. We furthermore examined the protein-protein interactions between and within different Rp1-D21 derivatives. We report novel insights into the precise structural requirements for NLR function and determine the function of a previously undefined motif. These insights enable a better understanding of how NLRs regulate the switch between the resting and the active states.
| One of the main plant disease resistance mechanisms is mediated by dominant resistance (R) genes with a major effect [1]. Many R genes have been cloned from diverse plant species, most of which encode nucleotide binding leucine-rich-repeat (NB-LRR or NLR) proteins [2,3]. Based on the secondary structure of their N-termini, NLRs can largely be subdivided into two classes: one containing a Toll-interleukin 1 receptor (TIR) domain (TIR-NB-LRR, TNL hereafter), and the other a putative coiled-coil (CC) domain (CC-NB-LRR, CNL hereafter). Both TNL and CNL proteins have been identified in dicots, while only CNL proteins have been found in monocots [4].
Each NLR is capable of directly or indirectly recognizing the presence of at least one effector protein, usually produced by a subset of isolates of a single pathogen species [5]. Recognition generally leads to activation of the NLR protein and initiation of signal transduction, resulting in the hypersensitive response (HR), which is often accompanied by the induction of a rapid localized cell death at the point of pathogen penetration [6]. HR can contribute to the halting of pathogen growth. When the corresponding effector is not present, wild-type NLRs are held in an inactive state, largely through self-inhibitory intra-molecular interactions, which may also be facilitated or reinforced by additional host proteins [5,7,8].
In general, plant NLR proteins contain three separable structural domains, an N-terminal CC or TIR domain, a C-terminal LRR domain and a central NB-ARC (ARC: APAF1, R gene products and CED-4) domain. The ARC subdomain in plant NLRs is further divided into two separated structural units: ARC1 and ARC2 [9,10]. In the NB-ARC domain, several conserved motifs are recognized; in linear order: P-loop/Kinase-1a/Walker A (henceforce called the P-loop), RNBS (Resistance Nucleotide Binding Site)-A, Kinase-2/Walker B, RNBS-B, RNBS-C, GLPL, RNBS-D, MHD [11]. Among the different motifs, the P-loop and MHD motifs are known to be very important for NLR function.
In their inactive form, NLRs likely have ADP bound to the NB-ARC domain. Activation of NLRs involves the exchange of ADP for ATP through a structural change that is thought to result in an open structure of the proteins [12,13]. The P-loop motif within the NB-ARC domain is thought to be involved in binding ATP/ADP [13]. P-loop mutations in the NLR resistance genes I-2 from tomato, M from flax and RPM1 from Arabidopsis impair ATP binding and ability to confer HR-based resistance [14,15,16]. Mutations in the conserved MHD-motif lead to autoactivation of NLRs, which is thought to be due to weakened ADP-binding and resulting structural change into an open conformation, favoring nucleotide exchange [15,17].
Autoactivate NLRs, i.e. NLRs that can be activated without the need for a recognition event, can be generated in vivo or experimentally via recombination between different NLRs, or through specific point mutations. One example of an in vivo recombination event leading to an autoactive NLR is from the maize Rp1 locus. Rp1 is a complex locus that carries multiple NLR paralogs. The number of these paralogs varies widely between haplotypes, some carrying more than 50 [18]. The Rp1-D haplotype consists of 9 NLR paralogs; Rp1-dp1 to Rp1-dp8 which have no known function, and Rp1-D itself, which confers resistance against the biotrophic fungal pathogen Puccinia sorghi, the causal agent of maize common rust [19,20]. These Rp1 paralogs are more than 90% identical in nucleotide sequence, allowing them to undergo unequal crossing over and occasional intragenic recombination. An intragenic recombination between paralogs Rp1-D and Rp1-dp2 produced the chimeric gene Rp1-D21, which was identified on the basis of its ‘lesion mimic’ phenotype in the absence of pathogen infection [20,21].
We previously demonstrated that many of the hallmarks of pathogen-induced immune response, such as H2O2 accumulation, increased expression of the defense-related genes PR1, PRms and WIP1, are associated with Rp1-D21-mediated lesion phenotype in maize. We also demonstrated that the Rp1-D21 lesion phenotype is genetic background-, temperature- and light-dependent [22,23]. Recently, we deployed Rp1-D21 as a tool in a novel enhancer/suppressor screen to identify natural modifiers of the HR [23,24,25,26].
Different NLR R-genes appear to have a variety of different structural requirements for proper activation and functioning [27,28,29,30,31,32]. The autoactive nature of Rp1-D21 and the fact that its ‘parental’ proteins, Rp1-D and Rp1-dp2, are not autoactive makes it a very useful tool to explore the molecular requirements for NLR regulation, specifically the switch between inactive and active states.
Here we report the characterization of the activity of Rp1 proteins in maize and in Nicotiana benthamiana, a model system widely used for characterization of R-gene mediated HR for both dicot and monocot NLRs [27,29,33,34]. Using a combination of genetic, molecular biological, biochemical and computational techniques we derive a model for Rp1 activity which provides a better understanding of how NLRs regulate the switch between the resting and the activation states.
Rp1-D21 is derived from the recombination of two NLR paralogs at the Rp1 locus, Rp1-D, and Rp1-dp2 that are 90% identical at the amino acid level [19,20,21,35]. Sequencing of Rp1-D21 showed that it encodes a protein of 1290 amino acids (AAs) and is a typical CNL with an N-terminal coiled-coil (CC) domain (AAs 1–189), an NB-ARC domain (AAs 190–527) and an LRR domain at the C-terminus (AAs 528–1290). The N-terminus of Rp1-D21 derives from Rp1-dp2 (up to AA 770–778), with the remainder deriving from Rp1-D (the precise breakpoint is impossible to define since nucleotides 2310–2333, corresponding to AAs 771–777 are identical in the two progenitor genes, Fig. 1). Thus, Rp1-D21 is comprised of the CC, NB-ARC and the N-terminus of the LRR domain from Rp1-dp2 and the C-terminus of the LRR domain from Rp1-D. A conserved EDVID motif (EDLLD in Rp1-D21, Fig. 1) can be identified in the CC domain of Rp1-D21. All of the important motifs present in the NB-ARC domain of typical NLR proteins can be found in Rp1 proteins (Fig. 1; [11]).
We performed a targeted ethyl methanesulfonate (EMS) mutagenesis screen in a maize line harboring Rp1-D21 to identify mutations that lost the autoactive HR phenotype conferred by Rp1-D21. Putative suppressor mutants were easily identified due to their robust growth compared to the stunted, lesion-mimic siblings heterozygous for Rp1-D21 (Fig. 2A). From about 23,000 EMS-mutated M1 plants, 12 missense and 2 nonsense intragenic mutants were identified (Fig. 2B; Table 1). Among the 12 missense mutants, five had mutations in the CC domain, three in the NB-ARC domain and four in the LRR domain (Fig. 2B). Notably, one mutant (nucleotide G244A, thus D82N in amino acid) was in the conserved EDVID motif of the CC domain, and one (C1193T, P398L) in the conserved GLPL motif of the NB-ARC domain (Table 1; Figs. 1 and 2B). The mutants T260I and E312K were adjacent to the conserved RNBS-A and Walker B motifs, respectively (Table 1; Figs. 1 and 2B).
We used Agrobacterium-mediated transient expression in N. benthamiana to investigate the structure/function of Rp1-D21. Rp1-D, Rp1-dp2 and Rp1-D21 were either fused to the N-terminus of enhanced green fluorescent protein (EGFP), a 3×HA (influenza hemagglutinin) tag or not tagged for subsequent functional analysis. When the three fusion proteins were transiently expressed using the cauliflower mosaic virus 35S promoter, a HR phenotype was observed 3 days post-infiltration (dpi) only with Rp1-D21, but not with Rp1-D, Rp1-dp2 or the empty vector (EV) control. The same phenotypes were obtained regardless of the tags used (EGFP, 3×HA or no tag) and regardless of the differing levels of protein accumulation observed for the three proteins (Fig. 3; S1A Fig.).
To further confirm that the phenotype conferred by Rp1-D21 in our N. benthamiana system conformed to the phenotypes observed in maize, we constructed Rp1-D21 expression vectors carrying the missense Rp1-D21 suppressor mutations identified in maize. When transiently expressed in N. benthamiana, Rp1-D21 produced a strong HR phenotype, rating 4 based on a 0–5 scale rating with 0 being no cell death or chlorosis at all and 5 being confluent cell death [36]. The HR rating is based on the average results from at least 10 individual leaves. All the mutants we identified as non-functional in maize, except for H59Y and G850D, abolished or greatly reduced the level of Rp1-D21-induced HR (Table 1; S2A Fig.). All constructs conferred high levels of protein expression (S2B Fig.). The concordance between the maize and N. benthamiana systems is very good and we expect that our conclusions based on the transient expression N. benthamiana system are generally relevant to the endogenous maize system. Thus, we used the transient expression of Rp1-D21 and its derivatives in N. benthamiana to further analyze the molecular mechanism of activation/auto-inhibition of Rp1 proteins.
The MHD motif is a highly conserved region of the NB-ARC domain involved in nucleotide binding [15]. Mutations in the MHD domain confer autoactivity to a number of NLRs. Two MHD motifs, here termed MHD1 and MHD2 (actually LHD in Rp1-D and Rp1-dp2; Fig. 4A), separated by a single amino acid are apparent in Rp1-D, Rp1-dp2 and Rp1-D21, and in a number of other CNLs [37]. To investigate whether mutations in either of the MHD motifs could cause autoactivity in the non-autoactive parental proteins, Rp1-D and Rp1-dp2, we generated the following mutations: Rp1-D(H517A), Rp1-D(D518V), Rp1-dp2(D513V) and Rp1-dp2(H512A/D513V) in the MHD1 motif, and Rp1-D(H521A), Rp1-D(D522V), Rp1-dp2(H516A), Rp1-dp2(D517V) in the MHD2 motif. Transient expression of the Rp1-dp2(D517V) MHD2 mutant conferred an obvious HR, with symptoms appearing at 3.5 days, about 12 h later than that observed with Rp1-D21 (Fig. 4B). HR was not induced by transient expression of any of the other MHD mutants, including the Rp1-D(D522V) MHD2 mutation (S3 Fig.). However, Rp1-D(D522V) expression was not detectable by western blot analysis using anti-HA antibody. Protein expressed from all the other constructs could be detected (Fig. 4B; S3 Fig.). We also generated the MHD2 mutations V1(D517V) and V16(D522V) in V1 and V16, two recombinant constructs that were not autoactive (see below) and found both of them induced HR (Fig. 4C).
The P-loop is a highly conserved motif in NLRs and mutations in this domain often result in loss of function and loss of autoactivity [14,15,27,34,38,39]. P-loop mutations (K225R) introduced into Rp1-D21 and the autoactive Rp1-dp2(D517V) mutant abrogated their ability to confer HR (Fig. 4D). This indicated that the Rp1-D21 autoactivity and presumably also the activity of the Rp1 proteins are P-loop dependent.
To examine which region of Rp1-D21 was required for triggering HR, different domains or domain combinations of Rp1-D21, including CC, CC-NB-ARC, NB-ARC, NB-ARC-LRR and LRR domains (hereafter CCD21, CC-NB-ARCD21, NB-ARCD21, NB-ARC-LRRD21 and LRRD21; Fig. 5A), were fused to the N-terminus of EGFP. CCD21 and NB-ARCD21 were derived from (and are therefore identical to) the corresponding domains of Rp1-dp2 while LRRD21 is recombinant between the LRRs of Rp1-D and Rp1-dp2. The transient expression of CCD21, but of no other Rp1-D21 domains or domain combinations, conferred HR (Fig. 5B). CCD21, NB-ARCD21 and CC-NB-ARCD21 produced higher protein accumulation than NB-ARC-LRRD21 and LRRD21 (Fig. 5B), excluding the possibility that lack of HR phenotype seen with most of the domain constructs was solely due to low protein accumulation. We also tested the HR phenotype conferred by different Rp1-D domains and found that the CC domain, but no others, induced HR when fused with EGFP (Fig. 5A). Surprisingly, no untagged or HA-tagged domains from either Rp1-D or Rp1-D21 induced HR (Fig. 5; S1B Fig.).
CCD21, but not CC-NB-ARCD21, induced HR, which suggested that NB-ARCD21 can inhibit CCD21-induced HR in cis (when the two domains were fused in the same molecule). Consistent with this result, no HR was observed when EGFP-tagged CCD21 and NB-ARCD21 were transiently co-expressed in trans (when the two domains were co-expressed as separate molecules) in N. benthamiana (Table 2). LRRD21 did not restore HR when co-expressed with CCD21 and NB-ARCD21 separately, or with CC-NB-ARCD21 in trans. However, co-expressing CCD21 with NB-ARC-LRRD21 retained HR in trans, suggesting that LRRD21 might interact with NB-ARCD21 to regulate CCD21–induced HR in cis but not in trans (Table 2). However, since LRRD21 or NB-ARC-LRRD21 are expressed at substantially lower levels than CCD21 or CC-NB-ARCD21 (Figs. 5B and 5C), we cannot exclude the possibility that the concentration of LRRD21 or NB-ARC-LRRD21 is simply too low to affect the NB-ARC suppression of CCD21–induced HR or CCD21–induced HR itself when co-expressed in trans. In contrast to the in cis result of CC-NB-ARCD, we found that NB-ARCD did not suppress CCD-induced HR in trans (Table 2), suggesting NB-ARCD might interact with CCD in cis but not in trans.
To further investigate the inhibition region of NB-ARC in CC-induced HR, we generated a series of deletion constructs from CC-NB-ARC (S4 Fig.). We found that extension including AAs 190–260 from the NB domain was sufficient to suppress CC-induced HR. Consistent with the data, the nonsense EMS mutant (Q346*) suppressed the Rp1-D21 lesion-mimic and stunted growth phenotype in maize (Fig. 2B; Table 1).
The C-terminus of the LRR domain has been demonstrated to be important for NLR activity [37]. We therefore investigated whether C-terminal deletions of the LRR domain from Rp1-D21 affected the HR phenotype. None of the six C-terminal deletion constructs (ranging from 34 to 639 AAs) induced HR after transient expression, even D21-LRR27 that lacked only the last 34 AAs of the acidic tail in C-terminus (S5 Fig.). This is despite the fact that all the constructs conferred high levels of protein expression (S5B Fig.).
To delineate the functional regions that cause Rp1-D21 to be autoactive and keep its progenitor proteins auto-inhibited, we generated a series of chimeric constructs recombinant at different positions between Rp1-D and Rp1-dp2 (Fig. 6). These proteins, both with and without a 3×HA C-terminal tag, were tested in the N. benthamiana transient expression system. Previously, the chimeric construct Rp1-dp2-D2, with the N-terminus deriving from Rp1-dp2 and the C-terminus from Rp1-D, with a recombination point at 980 AA (Fig. 6A), had been shown to cause HR in transgenic maize [21]. The identical chimeric construct, hereafter named Hd2, induced a strong HR when transiently expressed in N. benthamiana (Fig. 6A), consistent with the transgenic maize result.
The results of the transient expression of a series of chimeric proteins are summarized in Fig. 6. The HR phenotype conferred by the constructs fused with 3×HA was largely similar to that conferred by the constructs without any tag (Fig. 6A). All the tagged proteins accumulated to high and broadly comparable levels (Figs. 7 and 8; S6 Fig.).
Constructs V1 through V7 were generated with recombination points between Rp1-dp2 and Rp1-D ranging from AAs 1200 to 488 (Fig. 6A). V1 with a recombination point at AA 1200 did not confer HR while V2 with a recombination point just 30 additional AAs N-terminal to AA 1170 produced a strong HR, indicating that AAs 1170–1200 from Rp1-dp2 are important for auto-inhibition of V1. V4 and V5 (recombination point at 690 and 651, respectively) also produced strong HR phenotype, while V6 and V7 (recombination point 575, 488 respectively) conferred very weak or no HR (rating 1.5 and 0, respectively, Fig. 6A). These results suggest that the AAs 488–651 from Rp1-dp2 are important for the autoactivity of Rp1-D21. Together these data suggest that proteins with a combination of AAs 488–651 from Rp1-dp2 and AAs 1170–1292 from Rp1-D replacing the corresponding sequences of either parental protein are unable to maintain a self-inhibited state.
To test this hypothesis, we investigated whether swapping the N-terminal region (CC, NB-ARC or AAs 1–488) in Rp1-D21 affected its autoactivity. We generated 8 more chimeric constructs in which we exchanged the Rp1-dp2-derived N-terminal parts of Rp1-D21 with the corresponding parts from Rp1-D (Fig. 6B). Construct V8, in which the CC-NB-ARC (AAs 1–527) of Rp1-dp2 was replaced with the corresponding Rp1-D sequence, did not confer an HR. Similar results were obtained by exchanging the NB-ARC domain in V9. However, V10, exchanging the CC domain of Rp1-D into Rp1-D21 conferred a strong HR (Fig. 6B). These results suggest that the ‘parental origin’ of the CC domain (AAs 1–189) in Rp1-D21 plays only a minor or no role in its autoactive phenotype. V11 and V12 were generated by replacing AAs 190–370 or 190–458 (NB-ARC1 plus part of ARC2, respectively) of V10 with the corresponding Rp1-D sequence. They differed from V10 by only 2 and 5 single amino acid polymorphisms (SAAPs), respectively. V11 conferred a slightly weaker HR than V10 and the HR conferred by V12 was weaker still (Fig. 6B). Construct V13, of which the CC-NB-ARC domain was ‘reciprocal’ of V12, did not confer HR, confirming that AAs 458–527 (ARC2 region) from Rp1-dp2 were important for the autoactivity of Rp1-D21 (comparing V13, V12 and V8). Constructs V14 and V15 were generated by exchanging AAs regions 690–775 and 651–775 respectively of V10 with the corresponding Rp1-D sequence. Both constructs showed strong HR (Fig. 6B), similar with the corresponding AAs exchanges in Rp1-D21 (constructs V4 and V5 in Fig. 6A). V16 and V17 were ‘reciprocals’ of Rp1-D21 and V3, respectively and did not confer an HR phenotype (Fig. 6C).
In summary, this set of experiments indicated that the combination of AAs 458–651 from Rp1-dp2 and AAs 1170–1200 from Rp1-D is central to the deficiency of self-inhibition of Rp1-D21.
There are only 17 single amino acid polymorphisms (SAAPs) between AAs 1170–1292 of Rp1-D and Rp1-dp2, and only 5 SAAPs in the AAs 1170–1200 region that differentiate constructs V1 and V2 (Figs. 6A and 7A). However, V2 was autoactive, while V1 was not (Fig. 6). To analyze which SAAP is important for the self-inhibition of Rp1 proteins, we generated several mutants based on these 5 SAAPs in the autoinhibited construct V1, including V1(V1181A), V1(K1184N), V1(V1181A/K1184N) and V1(V1186T/F1188L/C1189D). When transiently expressed in N. benthamiana, V1(V1181A) was not autoactive, while V1(K1184N) and V1(V1181A/K1184N) produced a strong HR (rating 4), and V1(V1186T/F1188L/C1189D) a weak HR (rating 2; Fig. 7A). The results indicated that residue K1184 from Rp1-dp2 plays an important role in self-inhibition, and the residues V1186/F1188/C1189 play a minor role. Interestingly, one of the Rp1-D21 suppressor EMS mutants had a mutation at residue 1180 (P to S), only 2 AAs away from N1182 in Rp1-D21 (equivalent to N1184 in Rp1-D), emphasizing the importance of this region for regulating the activity of the Rp1 proteins.
To investigate whether mutating K1184 was sufficient to convert Rp1-dp2 into an autoactive NLR, we generated Rp1-dp2(K1184N). However, this single point mutation was not able to ‘activate’ Rp1-dp2 (Fig. 7B). This result suggests that additional SAAPs in the region of AAs 1200–1292 are also important for self-inhibition of Rp1-dp2. There are 12 additional SAAPs in the last 92 AAs between Rp1-D and Rp1-dp2 and of these six are found in the last 16 AAs (Fig. 7B). Therefore, we constructed the chimeric protein V18 by replacing the C-terminal 16 AAs in V1(K1184N), the autoactive V1 protein, with the equivalent AAs from Rp1-dp2. Interestingly, this chimeric protein V18 lost the ability to induce HR (Fig. 7B), demonstrating that the very C-terminal 16 AAs are very important for the self-inhibition of the Rp1 proteins.
In conclusion, these experiments further defined the combination of AAs required for the autoactivity of Rp1-D21: a combination of AAs 458–651 from Rp1-dp2 and AAs 1184–1292 (especially N1184 and the last 16 AAs) from Rp1-D are central to the autoactivity (see model in Fig. 7C).
As shown in Fig. 6B (construct V13), AAs 458–527 from Rp1-dp2 were very important for the autoactivity of Rp1-D21. There are two major regions of polymorphism, which we termed Patch 1 and 2, between Rp1-D and Rp1-D21 within the ARC2 sub-domain, from AAs 458–527 (Figs. 1 and 8). Homology modeling based on the crystal structure of the NB-ARC domain of the ADP-bound human apoptosis regulator APAF1 (Protein data bank entry 1z6t; [40]) predicted that Patch 1 and 2 were surface-exposed (Figs. 8A and 8B). Surface electropotential predictions indicated that the surfaces of both patches in Rp1-D were mostly negatively charged, while for Rp1-D21 Patch 1 was slightly positively charged and Patch 2 was mostly positively charged (Fig. 8C). To investigate whether Patch 1 and 2 played roles in the autoactivity of Rp1-D21, we constructed another two chimeric proteins, V19 and V20, by replacing AAs 458–463 (Patch 1) and 488–527 (Patch 2) in Rp1-D21 with the corresponding AAs from Rp1-D, respectively (Fig. 8D). Both V19 and V20 conferred a greatly reduced HR compared to Rp1-D21, while V13 (containing Patch 1+2 from Rp1-D) did not cause HR (Fig. 8D), suggesting that these patches have an additive effect on the autoactivity of Rp1-D21.
Self-association has been observed in both CNL (eg. barley MLA, Arabidopsis RPS5) and TNL (eg. tobacco N, flax L6) proteins and is important for the activity of NLRs [38,41,42,43]. To test whether Rp1-D21, Rp1-D or Rp1-dp2 could self-associate, we transiently co-expressed EGFP- and 3×HA-tagged proteins in N. benthamiana and performed co-immunoprecipitation (Co-IP) analyses. We observed that full-length Rp1-D21, Rp1-D and Rp1-dp2 all self-associated (Fig. 9A). We further showed that all three of the domains (CCD21, NB-ARCD21 and LRRD21) of Rp1-D21, were able to self-associate (Fig. 9B).
To further investigate whether Rp1-D or Rp1-dp2 could form heteromers with Rp1-D21, we transiently co-expressed Rp1-D21:EGFP and 3×HA-tagged Rp1-D or Rp1-dp2 proteins in N. benthamiana and performed Co-IP analyses. We found that Rp1-dp2 interacted with Rp1-D21 (Fig. 9C). We did not detect interaction between Rp1-D and Rp1-D21, which might have been due to the low expression of Rp1-D (Fig. 3B). Consistent with the interaction data, we observed that Rp1-dp2 could partially suppress Rp1-D21-induced HR (Fig. 9C).
To investigate whether different intra-molecular interactions are correlated with the different activities of Rp1-D21 and Rp1-D, we co-expressed pair-wise combinations of different domains in N. benthamiana and analyzed by Co-IP. We observed that NB-ARCD21 interacted with CCD21, but not with LRRD21 or LRRdp2 (Fig. 10; S7A–S7B Fig.). Additionally, we did not observe interactions between NB-ARCD and CCD or between NB-ARCD and LRRD under our conditions (Fig. 10; S7A–S7B Fig.).
Since the CC and the NB-ARC domains in Rp1-D21 and Rp1-D have different interactions, we investigated the regions required for this interaction. As noted above, the ARC2 domain (including Patch 1 and 2) is the major difference in the NB-ARC domain between Rp1-D21 and Rp1-D. In order to test whether this region affects the interaction between the CC and NB-ARC domains, we performed a Co-IP assay with transiently co-expressed CCD21/CCD and NB-ARCV12/NB-ARCV13, and observed that NB-ARCV12 interacted with both CCD and CCD21, while NB-ARCV13 did not interact with CCD21 (Fig. 10; S7A Fig.). The results suggested that ARC2D21 (AAs 458–527 from Rp1-dp2) is important for the interaction between CC and NB-ARC domains. In other words, ARC2D can suppress the interaction.
We also observed that LRRD21 interacted with NB-ARCV12, but not NB-ARCV13 (Fig. 10; S7B Fig.), suggesting that NB-ARC1D is required for the interaction between NB-ARC and LRRD21, which is consistent with previous report that ARC1 is required for binding to LRR of potato CNL protein, Rx [44].
To determine whether, as suggested by the recombination studies, N1184 and the C-terminal 16 AAs are involved in the intra-molecular interaction, we performed Co-IP assays with NB-ARCD21 and LRRV1, LRRV1(K1184N) or LRRV18 and observed that NB-ARCD21 interacted with LRRV1(K1184N), but not with LRRV1 and LRRV18 (Fig. 10; S7C Fig.), indicating that N1184 and the last 16 AAs from Rp1-D are required for the interaction between NB-ARCD21 and LRRV1(K1184N).
We investigated the effect of the five CC domain missense EMS mutations on CCD21-induced HR. Consistent with the results from full length proteins, CCR125W abolished HR induction and CCL89F and CCS108F reduced HR, while CCH59Y had no obvious effect compared to CCD21 (Table 1). Surprisingly, CCD82N induced stronger HR than CCD21, while CCD82N in the full length Rp1-D21 did not induce HR (Table 1).
Co-IP experiments were performed to explore whether the CC domain mutations affect the self-association of CCD21. CCD82N but not the other four point mutants reduced the strength of CCD21 self-association (S8A Fig.). To further test whether the mutations affect the interaction between CCD21 and NB-ARCD21, we chose CCD82N and CCR125W, two mutants that completely abolished HR in full length, to test the inter-domain interaction. We found that both CCD82N and CCR125W greatly reduced the interaction between CCD21 and NB-ARCD21 (S8B Fig.).
While 10 of the 12 Rp1-D21 loss of function EMS mutants isolated from maize also reduced or abolished the HR phenotype when transiently expressed in N. benthamiana, two, H59Y and G850D, did not. It is possible that these two mutations reduced the activity of Rp1-D21 to a level below the threshold for signaling in maize, but not in N. benthamiana due to the higher expression.
One mutation (D82N) was in the conserved EDVID motif. Previous reports have indicated an important role for the EDVID motif in NLR function. In Rx, the EDVID motif mediates the intra-molecular interaction between the CC and the NB-ARC-LRR domains [30,45] and the residues flanking the EDVID motif affect inter-molecular interactions with RanGAP2, which is required for Rx function [30,46,47]. In this study, we found that D82N abolished HR induced by full length Rp1-D21, but not by CCD21 (Tables 1 and 2), similar to what was observed for a mutation in the EDVID motif of MLA10 [27]. We further showed that CCD82N and CCR125W reduced the interaction with NB-ARCD21 (S8B Fig.). The “mousetrap model” for NLR activation [48,49] suggests that, like a mousetrap, the NLR protein must be ‘set’ in a primed state, ready to be activated. The apparent paradox that the D82N mutation causes loss of autoactivity in full length Rp1-D21 but not in the CC domain alone may be explained by the reduced association of CCD21 and NB-ARCD21 and the consequent loss of ability to set up this initial primed state in the full-length protein. It is also likely that the mutations in the CC domain might disrupt the inter-molecular interactions of Rp1-D21 with other co-factors.
Three of the missense mutations were within or next to the ADP binding pocket of the NB-ARC domain when the NB-ARC was modeled onto the APAF-1 structure; T260I was close to the RNBS-A motif, E312K was next to Walker B motif and P398L was in the GLPL motif (Table 1; Figs. 1 and 8B), suggesting that these mutations might change the ADP binding state. Consistent with our results, important roles for the RNBS-A, Walker B and GLPL motifs on modulating HR have been reported in other NLRs [14,42,44]. Mutations in the Walker B and GLPL motifs can also affect the intra-molecular interaction between CC and NB-ARC-LRR domains as evidenced in Rx [30]. Thus, we infer that the three mutations in the NB-ARC domain might also affect the interaction between CCD21 and NB-ARCD21.
Finally, four of the suppressor mutations (S737L, S794F, G850D and P1180S) were in the LRR domain (Fig. 2B). Loss-of-function mutations in the LRR domain have been reported in several NLRs [16,50,51,52,53,54]. According to our model (see below), the mutations likely abolish the ability of the LRRD21 domain to destabilize the interaction between NB-ARCD21 and CCD21.
Mutations from H (histidine) to A (alanine) or D (aspartate) to V (valine) in the highly conserved MHD domain result in constitutive activity of multiple NLRs from multiple species [17,27,37,41,55,56]. In the Flax NLR, M, ADP is bound to wild-type NB-ARC domain, while ATP is bound to the autoactive MHD mutant M(D555V) [15], providing direct evidence that the inactive “off” NLR binds ADP while the active “on” NLR binds ATP. It appears therefore that the MHD motif is important for maintaining the NLR protein in its appropriate state of activity.
Rp1-dp2 and Rp1-D as well as several other CNLs, but not TNLs, have two MHD motifs [17,37]. We have termed these motifs MHD1 and MHD2 (Fig. 4A). MHD1 is more conserved and is the functional MHD motif for many NLRs as defined in previous studies [15,17,27,38,56]. The aspartate (D) in MHD2 is quite widely conserved throughout most CNLs [37]. Of the CNLs that contained two MHD motifs [37], the effects of mutations in the MHD1 domain only have been reported for tomato NLR Mi-1.2(H840A) and Mi-1.2(D841V). Both mutations activate these proteins [34]. The possible function of the MHD2 domain has recently been examined in rice CNL-RGA5, but no functional effect was observed [57]. Transient expression of Rp1-dp2(D517V) and of V1(D517V) and V16(D522V), in which the MHD2 was mutated, conferred autoactive HR, while no MHD1 mutation had this effect (Figs. 4B and 4C; S3 Fig.). It is possible that the MHD2 rather than the MHD1 domain is functional in Rp1-dp2, or that the MHD1 and MHD2 domains coordinate the activity. This is the first report showing the functional significance of the MHD2 domain of a CNL for its activity.
The P-loop motif in the NB-ARC domain regulates nucleotide binding. P-loop mutations abolish the ability to confer disease resistance or HR induction in multiple NLR proteins [14,15,27,34,38,39,56]. As expected, the HR induced by Rp1-D21 and the MHD2 mutant Rp1-dp2(D517V) was abrogated by the introduction of a P-loop mutation (Fig. 4D), indicating that the activity of the Rp1 proteins is P-loop dependent.
There are several examples in the literature of recombinations between the TIR-NB-ARC or CC-NB-ARC and LRR domains of different NLRs resulting in autoactive proteins. The recombination of CC-NB-ARC from Gpa2 with LRR from Rx1 produces a gene conferring autoactive HR; The combination of Gpa2-ARC2 and the first two repeats of Rx-LRR region is essential for autoactivity [36,44]. Domain swaps between RPS5 and RPS2 [29] and between Mi-1.1 and Mi-1.2 [58] also gave rise to genes conferring an autoactive phenotype. In contrast to these artificially-constructed autoactive NLRs characterized in transient assays, Rp1-D21 is an autoactive protein that occurred via recombination and was identified in its endogenous genetic background. To identify the precise structural requirement for its activity, we performed a systematic structural analysis of Rp1-D21 using a set of artificial recombinants between the two ‘parental’ alleles (Fig. 6) and showed that the combination of AAs 458–651 (the ARC2 and N-terminus of the LRR region) from Rp1-dp2 and the C-terminal LRR region (especially N1184 and the C-terminal 16 AAs) from Rp1-D was critical for the autoactivity of Rp1-D21. This combination either destabilized the Rp1-D21 intra-molecular interactions that cause the inactive resting state, or stabilized interactions resulting in the active state (see below). In other words, these two regions appear to be important for maintaining the parental proteins in an inactive ‘resting’ state. Thus, in light of the “mousetrap” model [48], this region of the Rp1 family NB-ARC and LRR domains is involved in a precarious autoinhibiting conformation that is easily broken by alterations including exchange of a very small C-terminal region of the LRR domain, or presumably, in a wild-type context by the action of the cognate effector protein. It seems evident that the NB-ARC and LRR domains within each NLR must co-evolve to maintain the NLRs in a suitably auto-inhibited resting state in the absence of pathogen infection while maintaining the ability to respond to the cognate effector via intra-molecular interactions [36,38,55], but that precise mechanisms of auto-inhibition and activation may vary between NLRs.
The ARC2 domain is important for function in several NLRs [36]. We identified two major polymorphic regions localized in ARC2 that differentiated the NB-ARCD21 and NB-ARCD domains, Patch 1 and Patch 2, and showed that they are important for the autoactivity of Rp1-D21 (Figs. 1 and 8). Replacing either Patch 1 or Patch 2 of Rp1-D21 with the region from Rp1-D almost completely suppressed the autoactive phenotype (constructs V13, V19 and V20; Fig. 8D). In agreement with our structural modeling, these two patches were surface-exposed and carried largely opposite charges in NB-ARCD21 compared to NB-ARCD (Fig. 8B). The prevailing model for the activation of NLRs is that the off state binds ADP while the on state binds ATP [36,59]. Patch 2 is located three AAs N-terminal to the MHD1 motif and is localized in the exposed surface next to the nucleotide binding pocket [40], suggesting it might affect the state of nucleotide binding. Patch 1 is adjacent to a conserved RNBS-D motif in the ARC2 domain (Figs. 1 and 3A). Mutations in or next to the RNBS-D motif of PM3, RPM1 and Rx affected their function [16,36,45,55]. Thus, the sequence differences of Patch 1 between NB-ARCD21 and NB-ARCD might also affect the protein activity through disturbing the function of the RNBS-D motif.
Interestingly, we found that ARC2D21 is critical for the interaction between NB-ARCD21 and CCD21 or CCD (Fig. 10, compare V13 with V1 and Rp1-D with V12, and see S7 Fig.). This is, to our knowledge, the first demonstration of the role of ARC2 in determining NB-ARC and CC interaction in plant NLRs. Our modeling data also suggests why we detected interaction between NB-ARCD21 and CCD21 but not between NB-ARCD and CCD (Fig. 10; S7 Fig.). In the CC domain, the side chain of the EDVID motif of CNLs is largely negatively charged, and this motif is thought to mediate intra-molecular interactions of the CC domain with the NB-ARC-LRR domain of Rx [30,60]. In Rp1 proteins, the negatively charged EDVID and the positively-charged ARC2D21 are apparently very important for the interaction between CCD21 and NB-ARCD21 while the ARC2D is more negatively charged (Fig. 8C) and thus may not interact as strongly with the CC. Interestingly, we observed that AAs 190–260 from the NB domain were sufficient to suppress CCD21-induced HR (S4 Fig.), suggesting that ARC2D21 is not the only region which can regulate Rp1-D21 autoactivity.
CCD21 and CCD domains alone were sufficient to induce HR when fused with EGFP, but not with 3×HA or on their own (Fig. 5; S1 Fig.). This phenomenon of “tag-dependent activity” has been observed in other NLR studies [30,61]. The functional domains required for HR induction vary in different NLR proteins. The CC domain alone is sufficient for the HR phenotype triggered by MLA10, and also by NRG1 and ADR1, which contain atypical CC domains [28,41], while the NB-ARC domain of Rx can trigger HR [30], and for RPS5 the CC-NB-ARC is sufficient [29].
The fact that transient expression of CC-NB-ARCD21 or CC-NB-ARCD did not induce HR suggested that their respective NB-ARC domains repressed the signaling by the CC domain. Since full length Rp1-D21 induced HR and full length Rp1-D did not (Fig. 3), we inferred that the LRRD21 or LRRD domains are structurally incompatible with, respectively, repression of CCD21 autoactivity or with CCD autoactivity (Fig. 5). While interaction between the LRR and NB-ARC domains was detected in some combinations (e.g., see proteins V12, V1(K1184N) in Fig. 10), we did not detect any interaction between the LRR and NB-ARC domains of Rp1-D, Rp1-dp2 or Rp1-D21 in trans (Fig. 10; S7 Fig.). It is possible though that these domains interact in the full length context, or that their interactions are weak or transient and cannot be detected in our Co-IP conditions. Consistent with this, LRRD21 was unable to alter the suppression of CCD21 autoactivity by NB-ARCD21 in trans (Table 2). However, we cannot exclude the possibility that the lack of suppression in trans was due to the relatively low expression of LRRD21 (Table 2; Fig. 5B). A similar finding of cis but not trans interaction/autoactivity was reported for potato Rx and autoactive MHD mutants of tomato Mi-1.2 [34,44,62]. It is notable also that the CC-NB-ARC fusion from MLA10 and Rx can still trigger HR [27,30], but not in Rp1-D and Rp1-D21.
LRR domains have multiple reported roles in NLR activation. They are essential for the activation of the NLRs Rx and Mi-1.2 [34,62]. In the case of RPS5, the first four LRRs are the minimum region sufficient to inhibit the autoactive phenotype conferred by the CC-NB-ARC domain [29]. Conversely, the activation of RPS5 in response to disease requires the entire LRR domain [29]. Our analyses of Rp1-D21 C-terminal deletion constructs and LRR domain swap data (Fig. 7B; S5 Fig.) further confirmed the importance of the C-terminal LRR domain in regulating the activity of the Rp1 proteins. Similarly, the importance of C-terminal LRR domain has also been observed in flax TNL, L proteins [37].
The interaction patterns observed between NB-ARC and LRR from V1, V1(K1184N) and V18 suggested that K1184N and the last 16 AAs from Rp1-D are required for the inter-domain interaction (Fig. 10; S7 Fig.). No interaction was detected between the NB-ARC and LRR domains from V1 or between those from V18 when they were expressed in trans, but a trans-interaction was detected between these domains using the V1(K1184N) construct. Assuming that these interactions are maintained in the full length proteins, this result implies that the ability of V1(K1184N) to induce HR is due to the interaction of the LRR with the NB-ARC, which apparently is then locked into an activated state allowing CC-dependent HR. This also requires the C-terminal 16 amino acids of the Rp1-D21 LRR domain (Fig. 10; S7 Fig.). A further inference is that the C-terminal 16 AAs and N1182 in Rp1-D21 (corresponding to N1184 in Rp1-D) are crucial for the inhibition of NB-ARC activity that is required for CC-dependent HR.
Alternate “on” or “off” states mediated through intra-molecular interactions are thought to be one of the major mechanisms regulating NLR activity [51,62]. For example, intra-molecular interactions were detected between CC and NB-ARC-LRR, and CC-NB-ARC and LRR domains of Rx in the absence but not in the presence of its cognate effector, the PVX coat protein [62]. Similarly the differentially-activated states of Rp1-D, Rp1-dp2, Rp1-D21 and our additional recombinant proteins are likely due to the specific intra-molecular interactions found within each protein. While Rp1-D21 and V3 confer HR, their reciprocal constructs, V16 and V17, do not (Fig. 6C), indicating that the autoactivity of the recombinant protein is triggered by specific combination of sequences from Rp1-dp2 and Rp1-D. Our model (Fig. 11) explaining the intra-molecular interactions underlying the activity of the Rp1 proteins is as follows:
While we have not identified a cognate effector of Rp1-D and do not know how it is detected, we hypothesize that its presence is necessary for activating Rp1-D via direct or indirect interaction with the LRR domain. This model is based both upon the data presented here and a review of the related literature. While plant NLR resistance genes share similar structures and appear to perform very similar functions, it is clear that the molecular mechanisms underlying their function and their appropriate activation, while sharing certain similarities, vary substantially. This variation likely reflects the intimate and unique co-evolutionary processes each NLR has undergone both with respect to the interactions among their different domains and with their cognate pathogen effector proteins, and/or the host targets or decoys of those effectors. The molecular mechanisms underlying the auto-inhibition of the maize NLR common rust resistance protein Rp1-D and its autoactive derivative Rp1-D21 show both broad similarities to and distinct differences from what is known of other NLRs. Furthermore, we have identified several unique, or previously un-observed features, including: the AAs required for HR induction identified through EMS mutagenesis screening; the functional characterization of MHD2 motif; the involvement of the ARC2 domain in the interaction of the NB-ARC domain with the CC domain; the observation that N1184 and the C-terminal 16 AAs are involved both in the LRR/NB-ARC physical interaction and in regulating activity.
Recently, several NLR genes that have been transferred between plant species were demonstrated to still confer expected disease resistance specificities without obvious fitness effects, eg. Arabidopsis RPS4 (Resistance to Pseudomonas syringae 4), RRS1 (Resistance to Ralstonia solanacearum 1) and barley MLA1 [64,65]. Genetic manipulation of Rx indicates stepwise artificial evolution can be used to reduce the costs associated with disease resistance of NLRs [66]. Rp1-D21 is an autoactive mutant conferring nonspecific response to multiple maize rust species, including P. sorghi and P. triticina [19,21]. Presumably the resistance conferred by Rp1-D21 might extend beyond rusts to other biotrophic and hemi-biotrophic pathogens, however, the severe growth penalties associated with the expression of this gene make its application in agricultural production impractical. In conjunction with suitable promoters, it may be possible to engineer maize or even other plants with some of the chimeric constructs characterized here that confer a weaker HR phenotype to achieve an elevated disease resistance with fewer fitness consequences.
Wild type N. benthamiana plants were grown at 23°C with a cycle of 16 hrs light/8 hrs dark. Maize line Rp1-D21-H95 [25] was used to isolate the genomic DNA sequence of Rp1-D21. To generate material for EMS mutageneis, Rp1-D21 was first introgressed into A632 by 7 backcrosses (BC7). A few of these BC7 Rp1-D21 heterozygous plants were self-pollinated to generate Rp1-D21 homozygotes in the A632 background which were identified on the basis of their enhanced Rp1-D21 severity compared to the heterozygotes.
Pollen from the homozygous Rp1-D21 plants in an A632 background was collected and treated with EMS for 45 min before using it to fertilize the ears of an inbred line H95. The resulting M1 population was all heterozygous for Rp1-D21 and showed an HR phenotype with a relatively stunted growth stature. In this background, Rp1-D21-suppressed plants due to EMS mutagenesis were easily distinguished from the rest of the M1 siblings because of their non-lesioned and highly robust growth phenotype (Fig. 2A). About 23,000 M1 progenies were screened in field to identify intragenic suppressor mutants that had lost the Rp1-D21 phenotype. In total, 32 mutants which lacked the HR phenotype of Rp1-D21 were identified and 14 of them were characterized in detail. Gene specific primers for Rp1-D21 (S1 Table) were used to sequence the Rp1-D21 gene in these mutants.
All primers used in this study are listed in S1 Table. In Rp1-D and all its paralogs, no intron is found in the open reading frame (ORF) region [20,35], thus we amplified the ORFs of Rp1-D, Rp1-dp2 and Rp1-D21 from the plasmids gifted by Dr. Scot Hulbert (Washington State University), and cloned them into pENTR directional TOPO cloning vector (D-TOPO, Invitrogen). After sequencing, they were transferred into gateway vectors by LR reactions: pGWB2 (no tag), pGWB14 (with a 3×HA epitope tag in the C-terminus) or pSITEII-N1-EGFP (with EGFP epitopic tag in the C-terminus) [67,68]. Rp1-D21 was also isolated from maize line Rp1-D21-H95 using the primers listed in S1 Table.
The different domains (CC, CC-NB-ARC, NB-ARC, NB-ARC-LRR, and LRR) of Rp1-D21 or Rp1-D were amplified using primer pairs listed in S1 Table. The resulting PCR products were cloned into D-TOPO and sequenced, then transferred into pGWB2, pGWB14 or pSITEII-N1-EGFP by gateway LR reactions.
Overlapping extension PCR primers (S1 Table) were designed for generating the site-directed mutations: Rp1-D21(K225R), Rp1-D(H517A), Rp1-D(D518V), Rp1-D(H521A), Rp1-D(D522V), Rp1-dp2(D513V), Rp1-dp2(H512A/D513V), Rp1-dp2(H516A), Rp1-dp2(D517V), Rp1-dp2(K225R/D517V) and 12 missense EMS mutations listed in Table 2. The site-directed mutations were cloned into D-TOPO and verified by sequencing and sub-cloned into the gateway vector pGWB14 by LR reaction.
NLR sequences were aligned by ClustalW (www.ebi.ac.uk), and edited by BioEdit software. Homology modeling of the NB-ARC domain was performed with Phyre 2 [69] based on the crystal structure of human APAF1 (PDB: 1z6t). The three-dimensional structure and the surface electropotential were mapped using PyMOL (http://www.pymol.org/).
Agrobacterium tumefaciens strain GV3101 (pMP90) transformed with binary vector constructs was grown at 28°C overnight in 5ml L-broth medium supplemented with appropriate antibiotics. The bacteria were collected at 4,000g by centrifugation and resuspended in 2 ml resuspension buffer (10 mM MES pH5.6, 10 mM MgCl2 and 200 μM acetosyringone). The final concentration of the bacteria was diluted to the OD600 of 0.5 using the same resuspension buffer. To prevent the onset of post-transcriptional gene silencing and improve the efficiency of transient expression, a strain containing p19 protein was included at OD600 of 0.2 to all the bacteria strains [70]. The solution was left at room temperature for 3 hrs on bench before infiltration into the abaxial side of N. benthamiana leaves. After infiltration, plants were put at room temperature with 16h-light/8h-dark. At least 15 individual leaves were infiltrated by different constructs, and each experiment was repeated at least three times.
For protein expression analysis, three leaf discs (1.2 cm diameter) from different single plants were collected at 30 hours post infiltration (hpi). The samples were ground with prechilled plastic pestles in liquid nitrogen, and total protein was extracted in 160 μl extraction buffer [20 mM Tris·HCl (pH 8.0), 150 mM NaCl, 1 mM EDTA (pH 8.0), 1% Triton X-100, 0.1% SDS, 10 mM DTT, 40 μM MG132, and 1× plant protein protease inhibitor mixture (Sigma-Aldrich)]. Samples were centrifuged at 14,000 g for 15 min at 4°C, and 12 μl supernatants were mixed with 2× Laemmli buffer and loaded for SDS-PAGE. For Co-IP assay, EGFP- and 3×HA-tagged constructs were transiently co-expressed in N. benthamiana. Agrobacterium carrying each construct were diluted to a final concentration of OD600 = 0.4 plus p19 with OD600 = 0.2. Samples were collected at 30 hpi, and proteins were extracted by grinding 0.8 g of leaf tissues in 2.4 ml extraction buffer [50 mM HEPES (pH 7.5), 50 mM NaCl, 10 mM EDTA (pH 8.0), 0.5% Triton 100, 4 mM DTT and 1× plant protein protease inhibitor mixture (Sigma-Aldrich)]. Extracts were centrifuged at 14,000 rpm for 20 min at 4°C, and 2 ml supernatant was mixed with 30 μl anti-GFP microbeads (Miltenyi Biotec) and rotated for 3 hrs at 4°C. The samples were passed through pre-equilibrated MACS separation columns, and washed four times (1 ml, 1 ml, 500 μl and 500 μl) by washing buffer [50 mM HEPES (pH 7.5), 150 mM NaCl, 10 mM EDTA (pH 8.0), 0.2% Triton 100, 4 mM DTT and 1× plant protein protease inhibitor mixture (Sigma-Aldrich)]. The proteins were eluted by 100 μl pre-heated elution buffer and separated by SDS-PAGE. Proteins were transferred to nitrocellulose membrane (Fisher), and analyzed by western blot. HA detection was performed using a 1:350 dilution of anti-HA-HRP (horseradish peroxidase) (Cat# 12013819001, Roche). GFP detection was performed using a 1:8,000 dilution of primary mouse monoclonal anti-GFP (Cat# ab1218, Abcam), followed by hybridization with a 1:15,000 dilution of anti-mouse-HRP second antibody (Cat# A4426, Sigma). The HRP signal was detected by ECL substrate kit (Supersignal west femto chemiluminescent substrate, Thermo Scientific).
|
10.1371/journal.pgen.1002487 | Sequencing of Pooled DNA Samples (Pool-Seq) Uncovers Complex Dynamics of Transposable Element Insertions in Drosophila melanogaster | Transposable elements (TEs) are mobile genetic elements that parasitize genomes by semi-autonomously increasing their own copy number within the host genome. While TEs are important for genome evolution, appropriate methods for performing unbiased genome-wide surveys of TE variation in natural populations have been lacking. Here, we describe a novel and cost-effective approach for estimating population frequencies of TE insertions using paired-end Illumina reads from a pooled population sample. Importantly, the method treats insertions present in and absent from the reference genome identically, allowing unbiased TE population frequency estimates. We apply this method to data from a natural Drosophila melanogaster population from Portugal. Consistent with previous reports, we show that low recombining genomic regions harbor more TE insertions and maintain insertions at higher frequencies than do high recombining regions. We conservatively estimate that there are almost twice as many “novel” TE insertion sites as sites known from the reference sequence in our population sample (6,824 novel versus 3,639 reference sites, with on average a 31-fold coverage per insertion site). Different families of transposable elements show large differences in their insertion densities and population frequencies. Our analyses suggest that the history of TE activity significantly contributes to this pattern, with recently active families segregating at lower frequencies than those active in the more distant past. Finally, using our high-resolution TE abundance measurements, we identified 13 candidate positively selected TE insertions based on their high population frequencies and on low Tajima's D values in their neighborhoods.
| Transposable elements (TE's) are parasitic genetic elements that spread by replicating themselves within a host genome. Most organisms are burdened with transposable elements; in fact, up to 80% of some genomes can consist of TE–derived DNA. Here, we use new sequencing technology to examine variation in genomic TE composition within a population at a finer scale and in a more unbiased fashion than has been possible before. We study a Portuguese population of D. melanogaster and find a large number of TE insertions, most of which occur in few individuals. Our analysis confirms that TE insertions are subject to purifying selection that counteracts their spread, and it suggests that the genome records waves of past TE invasions, with recently active elements occurring at low population frequency. We also find indications that TE insertions may sometimes have beneficial effects.
| Transposable elements (TE's) are mobile genetic elements that parasitize genomes by semi-autonomously increasing their own copy number within the host genome. This evolutionary strategy has been remarkably successful: most organisms harbor TE's, and they can constitute anywhere from 3–80% of genomic DNA [1]. TE insertions may sometimes confer an adaptive advantage to the host organism [2], [3], [4], [5], [6], even performing essential functions, as in the classic example of Het-A elements, which comprise the telomeric DNA of Drosophila. In this case, the transposition machinery is used to regenerate telomeric DNA lost during DNA replication [7], [8]. In most cases, however, TE's are a liability for the host organism. Active TE's are an major source of deleterious mutations [9], [10], [11]—in extreme cases, resulting in a syndrome of chromosome breakage and sterility called hybrid dysgenesis [12], [13]. Even in less extreme cases, TE insertions can disrupt the coding or regulatory sequence of genes, impairing their function [14], [15], [16], [17]. TE's may also impose more subtle costs, such as a metabolic cost on the host due to the translation of TE-encoded proteins, and the replication of genomic DNA laden with both inactive and active elements [18], [19]. And lastly, similar TE sequences inserted into non-homologous regions of the genome can induce ectopic recombination between these regions, resulting in deleterious chromosomal rearrangements and aneuploid gametes [20], [21], [22]. Quite recently, it has become apparent that transposition is repressed by a special class of small RNAs devoted to this purpose [23], [24].
Thus, the primary forces affecting the spread and maintenance of TE's in populations —transposition, countered mainly by repression of transposition and purifying selection— are understood in broad outline. But, even in Drosophila, where the study of the population dynamics and forces affecting the patterns of transposable element insertion densities and frequencies has a long history (e.g, [25], [26], [27]), the dynamics of transposable element evolution remain controversial. Two patterns, and their conflicting interpretations, are of particular interest. First, low recombining regions such as the pericentric heterochromatin or the tiny fourth chromosome are highly enriched for TE insertions [20], [22], [28], [29], [30], suggesting that selection against new insertions is weaker in these regions than in regions with normal recombination rates. This is unlikely to be entirely caused by a general reduction in the efficacy of selection in low recombining regions due to Hill-Robertson effects [22], [31], [32], [33], as these effects only rarely lead to the fixation of TE insertions in non-recombining chromosomes [33]. Instead, the abundance of TE's in these regions is rather due to either a low rate of recombination yielding a low rate of ectopic recombination [21], [22], [33] or to the small fraction of functional DNA in these regions [15]. Second, insertions from the same TE family tend to segregate at similar population frequencies [34], [35]. This might be due to families experiencing bursts of activity, such that insertions from the same family tend to be approximately at the same age [18], [36], [37], [38], and thus also roughly at the same population frequency. Alternatively, families might differ in properties that determine their equilibrium population frequencies—in their transposition rate and in the strength of selection against individual insertions [34], [35]. In this scenario, high copy number elements are expected to experience high levels of purifying selection, due to an increased opportunity for ectopic recombination [34], [35]. Elements of these families will thus tend to segregate at low frequencies, but the family itself will be maintained by a high overall transposition rate.
Population level studies of different TE insertions provide the best opportunity for resolving these controversies, but these studies have been hampered by the lack of an unbiased and practicable method of characterizing the frequencies of at which TE insertions occur at individual insertion sites. In the past, unbiased estimates of TE insertion frequencies (except for those of small insertions which may be missed) have been obtained by in situ hybridization of DNA probes containing TE sequences to the polytene chromosomes of different individuals [20], [22], [39], [40], [41], [42], [43], [44], but this technique has limited resolution and finds only relatively complete insertions. More recent studies have used PCR to survey populations for known insertions (i.e, insertions identified from a reference genome) [34], [35], [45], [46], but these surveys are necessarily biased towards insertions with high population frequencies, as those insertions are most likely to be found in the reference genome. Methods to survey population frequencies of TE insertions in an unbiased fashion do exist [47], [48], [49], but genome-wide methods require separate sequencing of the genomes of multiple individuals, which is usually prohibitively expensive. Here, we use a novel and cost efficient approach to identify TE insertions, regardless of whether or not they occur in the reference genome. Using this method, we analyze TE insertion frequencies from a Portuguese population of Drosophila melanogaster. We find that this population harbors large numbers of TE insertions not present in the reference genome: a conservative estimate suggests that there are almost twice as many novel as known insertions. Using the frequency estimates from the Portuguese population, we investigate evidence for the different models of transposable element evolution outlined above.
We developed a method of identifying TE insertion sites, regardless of whether the insertion sites are known (present in the reference genome) or novel (not present in the reference genome). This method further provides estimates of the population frequencies of TE insertions without the large ascertainment bias that comes from sampling only TE insertions occurring in the reference genome. The method has three requirements: (i) an assembled reference genome (ii) a database of TE sequences, and (iii) paired-end (PE) sequences generated from the DNA of pooled individuals. The paired-end reads are mapped to a specially prepared reference genome, which consists of a repeat masked genome and the TE sequences used for repeat masking. A TE insertion is identified if one read of a PE fragment maps to a unique region of a reference chromosome and the other read maps to a TE (Figure 1A). We classified individual TE insertions using a nested hierarchy constructed from the information provided by FlyBase [50], with three primary orders (using the classification suggested by [1]) at the top level— one order of DNA-based elements, the terminal inverted repeat (TIR) elements, and two orders of RNA retrotransposons, the long-terminal repeat (LTR) elements and non-LTR elements. Within these orders, insertions are further classified into 115 families and 5,222 individual insertions (see Dataset S1). The use of a nested hierarchy allows us to operate at different hierarchical levels (mostly at the family level) thus facilitating identification of elements in spite of sequence divergence between the individual insertions (see Material and Methods).
Using this method, we characterized TE insertions in a D. melanogaster population from northern Portugal (Povoa de Varzim). To this end, we sequenced a sample of 113 isofemale lines and found that 11.4% of the aligned reads map to TE sequences, very similar to the proportion of sequences matching TE sequences (11.1%–13%) reported in a different study of a North American D. melanogaster population using low-coverage 454 shotgun sequencing [51]. In total, we identified 10,208 individual TE insertions (Table 1). These elements represent a broad taxonomic range, including 3,479 TIR insertions, 3,487 LTR insertions, and 2,975 non-LTR insertions (Dataset S2). To estimate the frequency of TE presence vs. absence at each insertion site, we first discarded low coverage sites (those having fewer than 10 reads) and overlapping TE insertions, as frequency estimates for these insertion sites are not reliable. We estimated the population frequency for the remaining 7,843 TE insertions (Table 1) as the ratio of the number of PE fragments showing the presence of the TE to total number of reads covering the site (Figure 1C; see Materials and Methods). As insertions present in the reference genome (“known” insertions) are expected to systematically differ in frequency from those that are not present (“novel” insertions), it is important that the method treats the two kinds of insertions equally.
We assessed the reliability of our method in three ways. First, we asked how well we were able to identify known insertion sites. We identify 3,384 of the 5,222 TE insertions that are present in the reference genome (Table 1), suggesting that we may have missed a large fraction of reference insertions. However, not all of the reference insertions necessarily occur in any given population. Using our data (see Material and Methods), we estimate that 3,639 (69.7%) of insertions present in the reference genome also occur in the Portuguese sample, (very similar to the estimate of 69.4% present in samples from an African and a North American population in another study [51]). We suspect that the remaining 255 (7% of the 3,639 reference TE insertions) missed by our approach were either nested within other TE insertions (as overlapping insertions can be difficult to detect), or segregating at low population frequency and so missed by our survey (see below).
Second, we assessed the reproducibility of our population frequency estimates using the 2,035 insertion sites identified by reads at both sides of the insertion site. Reads on either side of an insertion site represent independent assessments of the TE population frequency (Figure 1A), and, reassuringly, the resulting frequency estimates were highly correlated for insertions detected from both directions (Spearman's rank correlation, rS = 0.902, p<2.2e-16). From the empirical distribution of population TE insertion frequencies, we estimate that we cannot reliably detect TE insertions below a frequency of about 7% in this data set (see Text S1), causing a slight overestimate of average population frequencies. This effect can be reduced by increasing the depth of sequencing coverage of the population, though nested insertions will remain challenging to detect. Furthermore, this bias is small compared to the one introduced by estimating only the population frequency for inserts found in the reference sequence. In fact, the fraction of insertions fixed in the Portuguese population (those with a frequency >95% in our sample) is very different for the reference insertions than for the sample as a whole (83.1% for reference insertions vs. 34.5% for all insertions; Table 1; these proportions are similar across all TE orders).
Finally, we assessed how well our method was able to reproduce (and extend) two well-established results: 1) underrepresentation of TE's in functionally important regions, and 2) overrepresentation of TE's in low recombination regions [29], [30], [52], [53], [54]. Functional sequence is known to show a paucity of TE insertions [29], [30], [35], [45]. Here, we see a clear contrast between intergenic TE insertions and insertions into regions with annotated functions in both TE densities and population frequencies (Table 2). Insertions in exons, which are expected to have strong functional constraints, are both rarer and segregate at lower frequencies than those in intergenic regions (Table 2), suggesting that exonic insertions experience significant negative selection. Dividing exons into coding sequence (CDS), 3′-UTRs and 5′-UTRs reveals that all three categories show a deficit of insertions and fixed insertions compared to intergenic regions, though not to the same degree. Not surprisingly, the evidence for negative selection is strongest for coding sequence. We further found that 3′-UTRs consistently contain more TE insertions than 5′-UTRs (Table 2), which has been reported before [45], [55], [56]. This may indicate lower density of functional elements in 3′-UTRs [45], and that insertions in these regions have fewer or weaker deleterious effects than in the 5′-UTRs. Alternatively, TE insertions in 3′-UTRs may provide some important function, such as polyadenylation signals [56], and may therefore be beneficial. We did not, however, find a significantly higher fraction of fixed TE insertions in 3′-UTRs as compared to 5′-UTRs (Fisher's exact test; TIR: p = 0.61; LTR: p = 0.20; non-LTR p = 1), suggesting that the difference between 5′-UTRs vs. 3′-UTRs TE insertions is not due to positive selection. Insertions in introns are also underrepresented (Table 2; see also [29], [30], [45],but not [35]), possibly due to disruption of regulatory sequences. This finding should be treated with some caution, however, as the inexact positioning of insertion sites may cause us to misannotate some exonic and intronic insertions. While we expect very little contamination of the intronic insertions, as exonic insertions are rare, we cannot exclude them.
Next, we examine our data set for expected excess of TE insertions found in low recombination environments. We find the highest density of TE insertions among the different chromosome arms on the low-recombining fourth chromosome (Table 1). Within each of the major chromosome arms, TE densities increase near the low recombining regions of the centromere proximal regions [22], [29], [30], [35], [39], [41], [46], [57] a result which we also find in our data set (Table 3; Figure 2; Dataset S3; Figure S1). As our method cannot reliably detect nested or clustered TE insertions, the enrichment of insertions near the centromeres is likely to be underestimated. In contrast to the low recombination regions near centromeres, we find no enrichment of TE's in the telomere proximal regions, in spite of their low recombination rates (Table 3; Figure 2; see also [29], [30]), with the exception of INE-1, a very old and abundant TE element (Table 3; [58], [59]) Note that the assemblies of the major chromosome arms used here do not include the subtelomeric heterochromatin, in which the domesticated HeT-A and TART elements reside [29]. We also found, as expected, that both the total number of insertions and the fraction of fixed TE insertions were strongly negatively correlated with recombination rate (with both density and recombination analyzed in 100 kb windows, excluding windows with <10 insertions; rS = −0.36; p<2.2e-16 and rS = −0.73; p<2.2e-16 respectively; see also [30], [35], [39], [41], [46], [57]).
We estimate that, out of the 7,843 TE insertions for which we could obtain population frequency estimates, about one-third are fixed (34.5% at >95% population frequency), almost half are at low frequency (47.9% at <20% frequency), with the remainder segregating at intermediate frequencies (17.6% from 20 to 95%). That is, the distribution of insertion frequencies in the Portuguese population is U-shaped, with most insertions at either low or high frequencies. While TE insertions are well-represented in regions of normal recombination in our data set (just over half of the insertions with frequency estimates, or 3,985 of 7,843), most of the fixed insertions are in low recombination regions (83.1%, or 1,966 of 2,365).
Given that many TE insertions segregate at low frequencies [25], [35], [40], [41], [42], we might expect to find many insertions not present in the reference genome. In fact, this is the case: we detected 6,824 novel TE insertions, over twice the number of known insertions identified (3, 384 out of 5,222 present in the reference strain). These novel insertions have very different frequencies than those found in the reference sequence, with only 5% fixed, compared to 83.1% of the reference insertions. Consistent with the findings above, novel and known insertions tend to occur in different genomic regions. Most known insertions are in low recombination regions (78.8%, or 2,340 of 3,384; χ2 = 5257.3, p<2.2e-16; see also [51]), and most novel insertions are in normal recombination regions (63.5% or 4,151 of 6,824; χ2 = 566.5, p<2.2e-16).
The relative fraction of novel to known insertions is highly variable among orders and families (Figure 3). That is, families differ significantly in the typical population frequencies of their individual insertions (Figure 4; effect of family: Kruskal-Wallis χ2 = 4398.21, p<2.2e-16). Further, the families of the three TE orders differ in their median frequencies (Kruskal-Wallis χ2 = 6.122, p = 0.043, p-values obtained by permutation), probably due to the abundance of low-frequency elements in the LTR order (Figure 4).
Purifying selection has a strong potential to affect the density and population frequencies of TE insertions, as only TE insertions that do not disrupt important functions are free to drift to high frequencies. We confirmed above that TE insertions in functional sequence are rarer and, when they do occur, they have lower population frequencies than those in intergenic regions. But, in this data set, it is apparent that purifying selection on functional sequence cannot be the only evolutionary force affecting TE abundance. When we control for insertion into functional sequence, significant heterogeneity among insertions remains. Among intergenic insertions, there is still heterogeneity in population frequencies due to TE order and family [Kruskal-Wallis tests: family, χ2 = 2125.8, p<2.2e-16; order (using median family frequencies), χ2 = 10.013, p<0.002, p values obtained from permutation], suggesting that some property associated with these factors, such as its transposition rate or age (see below), affects insertion frequencies.
Importantly, the enrichment of TE insertions in low recombination regions does not appear to be solely attributable to a lack of functional sequence in these regions. That is, as low recombination regions have low gene density (the median number of exonic base-pairs per 100 kb window, excluding TE sequence, is 21,106 for low recombination regions and 25,569.5 for other regions; Mann-Whitney U-test, U = 168272, p = 2.4e-06), the argument could be made that the enrichment of TE insertions in these regions is due to the fact that TE insertions have fewer deleterious effects there [60]. But, this alone does not appear to explain the abundance of TE insertions in low recombination regions. Again restricting our analysis to intergenic insertions, we find that recombination still has a strong effect on TE density (number of insertions per 100 kb, rS = −0.311, p<2.2e-16) and on the population frequencies of insertions (rS = −0.504, p<2.2e-16). There is no evidence that this is because intergenic TE insertions are closer to genes in gene-dense high recombination regions, and therefore more constrained, as distance to the nearest gene has no effect on population frequencies of intergenic insertions (rS = −0.0005, p = 0.980). Moreover, even among exonic insertions, recombination plays a role in determining their population frequency (rS = −0.131, p = 0.008).
As the results above suggest that the concentration of TE insertions in regions of low recombination is not solely due to the lack of functional sequence there, we examined the role of recombination. It is generally assumed that the rate of meiotic recombination is positively correlated with the rate of ectopic recombination, although the exact relationship between meiotic recombination and ectopic recombination still needs to be determined. As ectopic recombination results in deleterious chromosomal rearrangements [21], [22], [61], it can cause purifying selection on insertions regardless of whether or not it disrupts functional sequence. We investigated the effect of recombination on polymorphic TE insertions. For this analysis, we excluded fixed TE insertions, the INE-1 family, and insertions on the fourth chromosome, as these insertions are potentially very old and thus unlikely to reflect ongoing purifying selection (see also [35]), and used only insertions in intergenic regions, to exclude the effect of purifying selection due to deleterious effects on genes. The remaining data set comprises frequency estimates for TE presence at 2,116 insertion sites.
We first confirmed that recombination rate also affects the frequency of these polymorphic elements, in addition to the effects on insertion density and fraction of fixed elements shown above. Insertions in low recombination regions are at higher frequencies than those in high recombination regions (rS = −0.101; p<0.0001; a negative correlation was found for all three orders, though the relationship is not significant for LTR elements; p = 0.059). We then examined our data for secondary factors affecting the rate of ectopic recombination. In addition to recombination rate, ectopic recombination between insertions is thought to be promoted by TE length, sequence similarity to other insertions, and the number of insertions from the same family [34], [35]. That is, insertions with more opportunity for mispairing —those with more extensive homology to paralogous sequence— should suffer correspondingly more from the effects of ectopic recombination than insertions with little similarity to other sequences. Some differences between TE families due to length could, instead, be caused by differential regulation of the piRNA pathway [23], [62], but stronger repression of transposition of a family should affect only the densities of new insertions, not their population frequencies.
We cannot explicitly explore sequence similarity or individual element length for most of the insertions in this data set, as we cannot recover sequence for insertions not present in the reference genome. However, we examine the effect of sequence length in two ways. First, we examine the effect of canonical length of a TE family on the median frequency of insertions from that family. We find a negative relationship between canonical family length and median family frequency (genome-wide, rS = −0.347, p = 0.0009; excluding regions of low recombination, rS = −0.064, p = 0.013). The interpretation of this result is complicated by the fact that the three major orders differ in both their typical lengths and in their median family frequencies (present study, [29]), but the relationship essentially holds within orders (though it is non-significant for the TIR order, which has very few families; Spearman rank correlation between median family frequency and length, LTR: 48 families, rS = −0.287, p = 0.048, non-LTR: 28 families, rS = −0.381, p = 0.050, TIR: 12 families, rS = −0.322, p = 0.308). Second, for the 123 polymorphic insertions that do occur in the reference sequence, we examine the effect of the length of individual insertions on population frequencies, and find that it is well-correlated with frequencies (rS = −0.33; p = 0.0002), even when only the 51 insertions in regions of normal recombination are considered (rS = −0.49; p = 0.0003; see also [35]).
As stated above, we expect TE insertions from families with large number of insertions to suffer more from the effects of ectopic recombination than those from families with few insertions. Consistent with this idea, we find a negative relationship between the size of a family (number of insertions genome-wide) and population frequency (rS = −0.178; p = <2.2e-16). There appears to be no additional effect of local element density, as a higher number of insertions from the same family in the immediate vicinity (within 1 MB) does not reduce population frequencies (i.e, there was no significant negative correlation between frequency and local density for any of the families with more than 30 insertions, rS = −0.153–0.362, p = 0.009–0.993), suggesting that ectopic recombination may not occur more often between nearby sequences.
Interestingly, the X chromosome does not have the lowest density of transposable elements, as might be expected given its overall higher rate of recombination ([22]; insertions per Mb for X = 73.1, 2L = 84.4, 2R = 99.3, 3L = 85.8, 3R = 69.4, 4 = 359.5; Dataset S4) or from direct effects of TE insertions in hemizygous males of D. melanogaster [20] (see also [63], [64]). However, these numbers cannot be compared directly, as chromosomes differ in the extent of low-recombining heterochromatin. Controlling for recombination rate, using an analysis of covariance, and ignoring the fourth chromosome, we find that there is no effect of X-linkage on the number of TE insertions per MB window (ANCOVA on ranks: median rank adjusted for X-linkage = 57.9; autosomal = 59.3; F1,120 = 3.53; p = 0.063; recombination rates not adjusted for X-linkage in this analysis only).
Finally, and perhaps surprisingly, we find that there is no correlation between recombination rate and frequency when we exclude regions of low recombination from the analysis (rS = −0.028, p = 0.274), suggesting that a small amount of recombination is sufficient to exert the whole effect of recombination on population frequencies. Accordingly, it might be the case that ectopic recombination is not directly related to recombination rates measured in genetic crosses, or that forces other than ectopic recombination are more important influences on population frequencies of TE insertions in the euchromatin.
To obtain an overall picture of the factors affecting TE frequencies, we used linear models to examine these factors in details, an approach previously used by [35], [37]. We examine several factors that should affect the rate of ectopic recombination: canonical family length, the recombination rate (adjusted for X-linkage in the standard way), the number of polymorphic insertions from the same family, and the number of polymorphic insertions of the same family in the neighborhood of the insert (within 1 MB). We also explore factors which might influence population frequencies by means other than ectopic recombination, such as chromosomal arm and distance to the nearest gene. Note that because of correlation between some explanatory variables in the model, we cannot use the model to make inferences about their strength of the effects (that is, the regression coefficients are not reliable); however, the overall fit of the model is valid [65], and is what will be assessed here.
We fit a linear model to the log-transformed frequencies, using the covariates listed above and their second order interactions. The insertion family and chromosome are added as factors. As we cannot simultaneously examine both family and order, we include only the family in the model, after confirming that using family is preferable to using order (Table 4). We then iteratively added terms to our model and used AIC as criterion for retaining the terms in the model. Consistent with the results above, the canonical length, number of polymorphic insertions in a family, recombination rate, and family are retained in the model (Table 4). Interestingly, the number of insertions in a MB region and the distance to the nearest gene, which had no significant effect above, are also retained, although dropping the distance to the nearest gene and its interactions had a minimal effect on the AIC (without distance to gene: AIC = −1110.1; Table S1). Not surprisingly, given the result above, dropping recombination rate as a continuous covariate and replacing it with a factor denoting whether the insertion is in a region of normal recombination, in the telomere proximal regions or in the centromere proximal regions improves the fit of the model (Table 4), again suggesting that a small amount of recombination is sufficient to reduce population frequencies of TE insertions. However, this picture may change with future improvements in estimates of the recombination rate, or a richer understanding of how homologous and ectopic recombination rates are related, particularly for telomeric regions [22], [66].
The patterns of population frequencies of insertions detailed above are attributable to ectopic recombination, but also have an alternative interpretation under a different model of TE evolution. That is, if TE families may have bursts of activity [18], [36], [37], [38] followed by long periods of inactivity, the different histories of different TE families will affect the characteristics of insertions examined here. Recently active families should show a large number of insertions segregating at low frequency, while recently inactive families should have fewer insertions, as many insertions will have been lost, while the remaining insertions are mostly fixed. For example insertions from the INE-1 family, which has not been active for >3 million years [58], [59], are mostly fixed (Figure 4). (Interestingly, we found that not all insertions of the INE-1 family are fixed (82% fixed; >0.95 population frequency). This could be due to either to a false absence reads leading to biased estimates of the population frequency (see Text S1), or, alternatively it is possible that not all insertions of INE-1 are fixed species-wide. In fact we find 124 INE-1 insertions that are not present in the reference sequence (45% of which are fixed in our sample), showing that at least some of these insertions are not completely fixed, as otherwise they would be present in the reference genome). Moreover, recent insertions will, on average, be longer than old insertions, as the well-documented deletion bias in Drosophila [67], [68], [69] will have had less opportunity to reduce the size of these insertions. In other words, the associations between frequency, length and number of TE insertions (although not the relationship with recombination rate) found above might be a consequence of recent activity of a family.
Assuming the burst model of family activity outlined above, we use the median age of 27 TE families (as estimated [37]) and ask how this affects population frequencies. We found that the estimated age of a family was well-correlated with its average population frequency (rS = 0.802; p = 4.72e-07; Dataset S5), suggesting that time of activity of a TE family has a strong impact on the population frequency spectra. There is a potentially confounding effect, however, as the age estimates of [37] are based on levels of sequence diversity in a TE family within the D. melanogaster genome. If sequence diversity affects the rate of ectopic recombination, such that insertions with many mutations enjoy a lower rate of ectopic recombination than those with very few, purifying selection alone might result in a negative correlation between diversity and frequency. To address this concern, we repeated the analysis, by using TE families thought to have recently invaded the D. melanogaster genome through horizontal gene transfer [70]. We find that these families show a significant enrichment of low population frequency insertions (Wilcox Rank Sum test; W = 234; p = 0.0042; Dataset S6).
To assess whether including the estimated age of a family improves the predictions of population frequencies, we also separately analyzed the subset of 794 insertions from the 11 element families for which we have age estimates and large numbers of insertions, allowing us to compare the effect of age and recombination rate. Because we are interested in comparing these family level effects, we drop family as a term in the model, but add order, and used age rank of the family to avoid any dependency on the specific age estimates. Using the add-drop procedure, age is retained in the model, regardless of the proxy used for recombination rate (Table 4). Removing age substantially worsens the model (with adjusted recombination rate: AIC = −380.1, with telomere proximal, normal recombination, and centromere proximal as proxies for recombination rate: AIC = −361.7; Table S1).
As our sample shows substantial number of polymorphic TE insertions, they may provide considerable material for adaptive evolution. In fact, among the few well-documented cases where we know the target of adaptive evolution at the genetic level in Drosophila, at least two are due to transposable element insertions. In one case, insecticide resistance is conferred by an accord insertion close to the Cyp6g1 gene, which results in increased expression levels of that gene [4]. In the other example, also involving insecticide resistance, a Doc insertion into the CHKov1 gene disrupts the gene, yielding an alternative set of transcripts [3].
We thus investigate the possibility that some of the insertions fixed in the Portuguese population were fixed via positive selection. We again ignored INE-1 insertions in this analysis, as INE-1 has not been active for >3 million years [58], [59], and thus are unlikely to be the targets of recent positive selection. We considered TE insertions in fixed in high recombining regions as promising candidates for recent positive selection, as fixed insertions in regions of normal recombination are unusual (132 cases, or <3% insertions when low recombination regions are excluded). To distinguish selection from genetic drift, we also required candidates for positive selection occur near regions of low Tajima's D values (TD) [71]. Negative values of TD indicate an excess of rare mutations, one possible signature of a sweep due to positive selection [72]. We identified insertions in the neighborhood of TD lower than the genome wide 5% quantile (i.e, within, or immediately adjacent to, windows of 500 bp with TD<−2.265 for the autosomes and TD<−2.397 for the X chromosome). We included the flanking windows of the actual TE insertion in this analysis, as the strongest signal of selection may not be directly at the site under positive selection [73].
As a result of this analysis, we identified 13 putatively positively selected TE insertions (Table 5), of which 5 are located on the X chromosome and 8 on the autosomes. We asked if two TE insertions involved in the adaptation to insecticides are among the candidates, an accord insertion close to the Cyp6g1 gene [4] (not in the reference genome), and a Doc insertion into the CHKov1 gene [3] (insertion FBti0019430 in the reference genome). Both were among the candidates identified here (Table 5). We also compared our results to a different set of putatively positively selection TE insertions identified by [5], and found that only the Doc insertion mentioned above overlaps between the two studies. This is likely due to the different criteria used— fixation in a single population (present study) vs. a frequency difference between African and non-African populations [5].
Of the 13 insertions identified as putatively positively selected here, 11 are present in the current annotation of D. melanogaster (5.31). These candidates belong to very different TE families and orders, with 2 TIR insertions, 8 LTR insertions and 3 non-LTR insertions. The location of these insertions with respect to the nearest gene varies: 4 are upstream of the next gene, one is in the CDS, 3 are in introns, one is in the 3′-UTR, and 4 are downstream of the next gene. The putative functions of these genes nearest to candidate TE insertions are diverse, ranging from wing disc pattern formation to spermatoid development (Table 5), and show no significant enrichment for a gene ontology category. The fact that only 2 of these insertions are in exons suggests that positively selected TE insertions mostly have an influence on expression of genes by cis-regulatory effects. The 3 intronic insertions may instead yield alternative transcripts.
These insertions represent only candidates for positive selection, and we cannot exclude other possibilities. For example, it may be that the TE insertion happens to be near the target of a selective sweep, resulting in a low TD, while the causative mutation is elsewhere. This may have been the case for the candidate TE insertion close to the Est-6 gene (FBti0063191), which was identified as ancestral, predating the split of D. melanogaster and D. simulans [51]. It has further been suggested that the cis-regulatory region of the Est-6, which co-segregates with the TE insertion has been the target of positive selection [51], [74].
Alternatively the regions neighboring the sweep may have exceptionally low TD for stochastic reasons, such as fluctuations in population size [75]. Furthermore, low TD values may partly be caused by non-synonymous sites, as we found that windows with low TD values contain exons slightly more often than other windows (low TD: 39.7%, high TD: 31.7%; Fisher's exact test; p<2.2e-16). We also examined the regions near our candidate insertions for a depression in nucleotide variability (see Figure S2) expected near strong selective sweeps [76]. We note, however, that such a signature strongly depends on the history of the selection event. Only for selective sweeps starting from a low population frequency is a pronounced trough in variability expected around the causative mutation. Given this limitation, we consider the fact that nine out of 13 candidates show a visually recognizable trough in variability a strong support for the non-neutral history of these TE insertions.
The true number of positively selected TE insertions in our natural population of D. melanogaster may, for several reasons, be higher than the 13 candidates presented in this work. That is, in addition to ignoring insertions in low recombining regions and nested insertions, our criteria also exclude incomplete sweeps, and may sometimes exclude sweeps fixed from standing variation [77], [78], which have been shown to contribute to TE insertion mediated adaptation [3], [5]. Hence, the number of TEs that contribute to adaptation of natural D. melanogaster populations may be substantially larger than this estimate indicates.
In this work, we have developed a method for the identification of population frequencies of TE insertions in pooled populations using paired-end sequencing (Figure 1). The primary advantage of this method is that it does not require previous knowledge of TE insertions allowing for relatively unbiased estimates of their frequencies. This is a substantial improvement over some prior methods, which measure TE polymorphism only at insertion sites known from the reference genome (e.g, [79]) and suffer the attendant ascertainment bias problems [35], [79]. Sequencing and assembling of every individual separately also allows ascertainment bias free frequency estimates, but is costly and error prone, as repetitive regions are notoriously difficult to assemble [80]. In contrast, our method requires sequencing of one pooled sample for the population of interest, a reference sequence and an appropriate TE database. Extension to other species with sequenced genomes should thus be straightforward. However, it does suffer a few limitations. It cannot identify insertions from TE families not in the supplied TE database, identify clustered and nested TE insertions, distinguish full-length from partial insertions and the locations of insertions are only roughly estimated. And finally, insertions segregating at low population frequencies can be missed, depending on the depth of coverage.
Because the method treats novel and known insertions equally, we are able to estimate the frequencies of large numbers of insertions whether or not they are present in the reference genome. In fact, most (66.8%) of the insertions identified in the Portuguese population were not present in the reference genome. The abundance of non-reference insertions is a natural consequence of the population frequency distribution of insertions, which tend to be either rare or fixed in this data set, consistent with previous reports based on smaller data sets [34], [35], [40], [42], [45], [46], [57], [81]. That is, the high proportion of TE insertions segregating at low frequencies— e.g, 48% occur at frequencies lower than 20%— implies that there will be many TE insertions not captured in the reference genome. Sampling more individual lines and/or increasing the coverage of insertion sites (which averages 31-fold in our study) will increase the number and proportion of novel insertions, as more rare insertions will be sampled. A similar effect can be seen in our data set by lowering the number of paired-end fragments required to identify and insertion from three to two, in which case the fraction of novel insertions increases to 81% and the number of novel insertions more than doubles (from 6,824 to 14,786).
Our improved frequency estimates confirm many inferences from previous work, but provide a more complete picture of the evolutionary forces acting on insertions within populations and put individual observations into context. For example, the population frequency of TE insertions strongly depends on the sampled TE families and orders (Figure 4; [35], [51])—e.g, DNA transposons were more frequently fixed than RNA transposons— and the sampled regions (Figure 2). Thus with different sampling strategies vastly different estimates of TE population frequencies will be obtained, which could be an explanation for the conflicting reports about TE population frequencies. In particular, estimates of frequencies of insertions from in situ methods are mainly limited to the euchromatin, where most insertions segregate at low population frequencies (Figure 2; [25], [41], [42], [57]) and may have missed the fixation of many elements in heterochromatic regions for technical reasons [46]. In contrast, when frequencies of insertions in the reference sequence are examined, most TE insertions are in low recombination regions, and are fixed or at appreciable frequencies (e.g, [35], [46], [51]).
Further, our data shed light on the nature of the forces affecting TE insertions in populations. These data provide evidence that ectopic recombination might counteracts the spread of TE insertions through populations, and that the abundance of TE insertions in low recombination regions is not, or not entirely, due to less functional selection in these regions. But they also suggest that an equilibrium model where transposition rate and purifying selection due to ectopic recombination are the primary factors affecting population frequencies may not provide a complete picture. New families invade the Drosophila genome [82], and recent successful invasions, coinciding with bursts of activity due to derepression, and the time since these bursts must have some influence on element frequencies. In fact, we see that the age of TE families, estimated from phylogenetic data, appears to be well correlated with population frequencies. And age may also contribute to the relationship between length and frequency—old, high frequency insertions will accumulate deletions and thus be short [46]. The strongest argument against age having an effect on element frequencies is that it requires a recent increase in activity in many LTR families to explain the abundance of low frequency/high copy number families in this order [35]. However, it is plausible that the enrichment of young TE families is due to rapid turnover in LTR families [36], [37]. That is, it may be that new LTR families invade Drosophila species more frequently (perhaps due to higher rates of horizontal transmission [38], [70] or differential targeting by small RNAs [38], [83]), and are also lost faster (due to more frequent ectopic recombination) than families from other orders.
Finally, our novel method of characterizing TE insertion population frequencies can be applied to any organism with a well-assembled reference genome. Application to other organisms will demonstrate the generality of the patterns observed in Drosophila.
We sequenced 113 isofemale lines cultured from D. melanogaster females collected in 2008 from northern Portugal (Povoa de Varzim), as described previously (PoPoolation DB [84]). The lines were kept in the laboratory for five generations, and five females from each line were combined into a pool of flies for sequencing. DNA was extracted from homogenized female flies with the Qiagen DNeasy Blood and Tissue Kit (Qiagen, Hilden, Germany) for generation of paired-end libraries using the Genomic DNA Sample Preparation Kit (Illumina, San Diego, CA). Briefly, 5 µg of DNA were sheared with a nebulizer and after end repair, A-tailing, and ligation of paired-end adapters, the library was size-selected on an agarose gel (300 bp) and amplified using 10 PCR cycles. Cluster amplification was performed using a Paired-End Cluster Generation Kit v2. Sequences were generated with the Illumina Sequencing Kit v3 and the Genome Analyzer IIx. Image analysis was performed with the Firecrest, Bustard and Gerald modules of the Illumina pipeline v. 1.4. In total, we sequenced 5 paired-end lanes, which produced 80 mio PE fragments (160 mio individual reads) with an average read length of 74 bp.
The goal of the mapping procedure used here was to identify cases in which one read of a paired end fragment maps to a TE, and the other maps to a location in the Drosophila genome. To achieve this, we used the D. melanogaster reference genome v 5.31 and transposable element sequences obtained from FlyBase (http://flybase.org/; [29], [85]). We retained only TE sequence having a length greater or equal to 40 bp). We also masked repeat sequences in the reference genome using RepeatMasker open-3.2.9 [86] with the rmblast 1.2 search engine (parameters: -no_is -nolow -norna -pa 4) using the length filtered TE sequences form FlyBase as custom repeat library. We then constructed a combined reference sequence consisting of the repeat masked reference genome of D. melanogaster (v5.31) and the length filtered TE sequences. We then mapped our ∼160 million paired end reads to this combined reference sequence using BWA-SW 0.5.7 [87] with default settings. BWA-SW uses a Smith-Waterman algorithm [87], which allows for a partial mapping of the reads, potentially useful for reads spanning a TE insertion site. As BWA-SW does not support mapping of paired end reads, paired end information was recovered using a custom Perl script (samro). The mapping results were further processed using samtools 0.1.8 [88]. Both paired-end reads were mapped for 69.6 million (86%) out of the 80.5 million PE fragments (Table S2). We identified 961,283 PE fragments indicating the presence of TE insertions, i.e.: PE fragments with one read mapping to the reference chromosome and the other one to a TE. Unexpectedly, the number of PE fragments confirming a TE insertion from the forward direction (forward reads) and the number from the reverse direction (reverse reads; Figure 1A) were unequal (414,123 reverse reads; 547,160 forward reads; Fisher's Exact Test; p<2.2e-16). We can only speculate as to what causes this bias, with one possibility being the heuristics applied in the BWA SW algorithm [87].
We clustered forward and reverse reads into distinct TE insertion sites, limiting this analysis to TE insertion fragments in the non-heterochromatic reference chromosomes (2L, 2R, 3L, 3R, X, 4), using a two step protocol. First, we clustered reads in the same direction if they: (i) were separated by less than 225 bp (insert size+2×standard deviation the average distance between reads of a PE fragment) and (ii) mapped to the same TE type (e.g, INE-1). We further required that an insertion be supported by a minimum of 3 PE fragments, each with a minimum mapping quality of 15. We identified 6,672 insertions by forward reads, and 6,566 by reverse reads; note that this ratio of clustered forward and reverse reads is more balanced than that of the unclustered ones. Next, we combined adjacent forward and reverse insertions of the same family separated by between 74 and 250 bp intro single insertions (where the minimum distance is the read length, and the upper limit is empirically determined to result in the lowest levels of misclustering, see below). In order to treat TE insertions that are in and not in the reference genome equivalently, we ignored repeat masked sequence in the reference genome in calculating the distance between forward and reverse insertion sites. Using this procedure, we identified 10,076 individual TE insertion sites. Our procedure for clustering forward and reverse reads is based on distance, and so may therefore sometimes result in incorrectly grouping multiple TE insertions, or, conversely, erroneously splitting single TE insertions into two. We estimated the accuracy of the clustering procedure using TE insertions known from the FlyBase annotation (v5.31). We assumed that a TE insertion identified in our data corresponds to an insertion in the reference sequence if both insertions belong to the same family, and if the paired reads supporting the insertion map to within 300 bp of the reference insertion. This analysis showed that a total of 150 insertions were erroneously clustered together, while 18 were falsely split. For further analysis, we corrected the clustering for these TE insertions, resulting in a total of 10,208 TE insertion sites, with 3,030 identified by both forwards and reverse reads (n2) and 7,178 TE insertions solely by forward or solely by reverse reads (n1; Figure 1B).
We estimated the total number of known TE insertions present in the sample using the following method. Let p be the probability of identifying a reference insertion present in the population, let n1 be the number of reference TE insertion only identified by reverse or forward reads, and let n2 be the number of reference TE insertions identified by both forward and reverse reads. Let nT further be the total number of reference insertions present in the sample. If the probability of identifying a reference insertion (p) is equal across insertion sites, then it is binomially distributed, with: n1 = 2p(1−p)nT, and n2 = pi2nT. Given the direct estimates of n2 and of n1 from the data (see above) p and nT. can be estimated. It follows that n0, the number of TE insertions not identified, can be calculated as: n0 = nT−n2−n1 This analysis was conducted for each TE order separately.
We estimate the frequency at which a TE is present at individual insertion sites as the ratio of the number of PE fragments that support the presence of the insertion (“presence fragments”) to the total number of reads covering the physical insertion site (including both “presence” and “absence” fragments; Figure 1C). While this is simple in principle, a practical difficulty arises from the fact that the precise TE insertion site is not known for all novel TEs, and, in these cases, we cannot determine with certainty whether a pair of reads map to either side of an insertion site, indicating the absence of the TE. Hence, we used the presence fragments to empirically define two 100 bp ranges in the reference genome on either side of the insertion site where we expect absence reads to map (Figure 1C: “range”). By truncating these ranges to 100 bp, we avoid overestimating the size of the ranges due to presence fragments with unrepresentatively large insert sizes, which could lead to an overestimate of the number of absence fragments. To estimate frequencies, we use only reads mapping within these ranges to tally either the presence or the absence of an insertion. Specifically, we considered absence fragments to be those where (i) both reads map in a proper pair, i.e, both reads map to the same reference, with the read mapped to the forward strand followed by the read mapped to the reverse strand, and with no overlap between the reads, and (ii) the end position of the 3′ read (or, for forward insertions, the start position of the 5′-read) maps within the 100 bp range (see Figure 1C for an example of a reverse insertion). We considered presence fragments to be those where (i) one read aligns to a TE sequence and the other read to the reference genome, and (ii) the position (end position for reverse reads and start position for forward reads) of the read mapping to the genome is within the same range as that used for the absence fragments. If a TE insertion is only identified by forward or reverse reads, the frequency estimate is solely based on the forward or reverse reads; otherwise, we averaged the estimates obtained from forward and reverse reads. We discarded insertion sites with lower than 10-fold coverage (defined as the sum presence and absence fragments), and TE insertion sites with overlapping ranges, yielding a total of 7,843 TE insertions with population frequency estimates (Table 1). See Text S1 for an assessment of the reproducibility of these frequency estimates.
Recombination rates for D. melanogaster were obtained for 100 kb windows from http://petrov.stanford.edu/cgi-bin/recombination-rates_updateR5.pl. The exact position of a TE insertion cannot be determined with our method, so we approximated positions using either the midpoint between forward and reverse reads identifying an insertion, or for TE insertions only identified by reads from one direction, using the last (first) position occupied by a forward (reverse) read plus (minus) 26 bp (1/3 inner distance between paired end reads).
We used the Flybase annotation to determine the functional category of the sequence surrounding the insertion, with categories expected to have stronger functional constraints taking precedence, as this is conservative for our purposes [in order of priority: exon (which can be further divided into CDS, 3′ UTR, 5′ UTR), ncRNA, regulatory, intron and intergenic]. We used chi-square tests to compare the number of TE insertions in a feature to the number in intergenic regions, and Fisher's exact test to compare the number of fixed and polymorphic TE insertions to those in intergenic regions. To analyse population frequencies, we used either the non-parametric Mann-Whitney U test, or linear models on log-transformed data. For linear modeling, we attempted to use arcsine transformed frequencies and generalized linear models with binomially distributed errors, but qq plots showed that these models fit poorly, while linear models fit to the log-transformed data fit well. As many of the tested models are non-nested, we used AIC to test model fit. Reduced models were obtained using the “step” function in R, which adds and drops terms based on AIC.
We calculated Tajima's D in non-overlapping 500 bp windows using PoPoolation v1.2.1 [89]. To do this, we trimmed (trim-fastq.pl, with base quality threshold of 18 and minimum length of 50) PE reads and subsequently mapped them to the D. melanogaster genome (5.31) using BWA 0.5.8 (parameters: -l 150 -n 0.01 -o 2 -e 12 -d 12). Paired-end information was restored using BWA SAMPE (0.5.8), and reads were filtered for unambiguous positions with samtools (0.1.8) [88] using a minimum mapping quality of 20. We converted the reads into a pileup file using samtools (0.1.8). The pileup file was sub-sampled to a uniform coverage of 30 bp using random sampling without replacement, a maximum coverage of 250 and a minimum base quality of 20. Tajima's D values were calculated using a minimum count of one and a window size of 500 bp; Tajima's π values were calculated using a minimum count of one and a window size of 2,500 bp.
For each of the candidate insertions, the nearest gene, the relative location with respect to the nearest gene and the ID of known TE insertions were obtained visually with IGV (1.5.06) [90], using the annotation of D. melanogaster (5.31). Putative functions of genes were obtained from FlyBase (http://flybase.org/) using either the first biological function, if available, or when not available the first molecular function. Analysis for an enrichment of GO terms was performed using FuncAssociate 2.0 [91].
The data are available from the European Nucleotide Archive (http://www.ebi.ac.uk/ena/) with the accession number SRA035392. The software used in this work is distributed as PoPoolation TE and available at Google Code (http://code.google.com/p/popoolationte/).
|
10.1371/journal.pgen.1005842 | A Parallel G Quadruplex-Binding Protein Regulates the Boundaries of DNA Elimination Events of Tetrahymena thermophila | Guanine (G)-rich DNA readily forms four-stranded quadruplexes in vitro, but evidence for their participation in genome regulation is limited. We have identified a quadruplex-binding protein, Lia3, that controls the boundaries of germline-limited, internal eliminated sequences (IESs) of Tetrahymena thermophila. Differentiation of this ciliate’s somatic genome requires excision of thousands of IESs, targeted for removal by small-RNA-directed heterochromatin formation. In cells lacking LIA3 (ΔLIA3), the excision of IESs bounded by specific G-rich polypurine tracts was impaired and imprecise, whereas the removal of IESs without such controlling sequences was unaffected. We found that oligonucleotides containing these polypurine tracts formed parallel G-quadruplex structures that are specifically bound by Lia3. The discovery that Lia3 binds G-quadruplex DNA and controls the accuracy of DNA elimination at loci with specific G-tracts uncovers an unrecognized potential of quadruplex structures to regulate chromosome organization.
| Non-canonical DNA structures, including four-stranded Guanine quadruplexes (G4 DNA), have been observed readily in vitro, but their regulatory importance within cells has been particularly challenging to demonstrate conclusively. We have discovered a G4 DNA binding protein, Lia3, that specifically regulates programmed DNA elimination events in Tetrahymena thermophila. This ciliate deletes nearly one-third of its germline genome from each developing somatic nucleus. These genomic deletion events must be accurate as the thousands of DNA regions excised are located near genes and/or their promoters, thus aberrant excision may alter gene expression. When we knocked out the gene encoding Lia3, we found that the boundaries of the excised regions were heterogeneous for a subset of loci that are flanked by G-rich (5’-AAAAAGGGGG-3’) boundary controlling sequences. When we tested whether Lia3 bound this sequence, we discovered that the sequence itself formed G4 DNA and that Lia3 bound only when the sequence adopted this conformation. Our findings that Lia3 binds G4 DNA and that deletion of the gene encoding Lia3 perturbs the boundaries of the excised loci which are flanked by this quadruplex-forming DNA provides compelling evidence that this non-canonical DNA structure has a critical role during development of these cells.
| Ciliates maintain distinct germline and somatic genomes that are partitioned into different nuclei, called micro- and macronuclei, respectively [1]. At each sexual round of the ciliate life cycle, the somatic genome is destroyed, and new germline and somatic genomes are created from identical copies of a zygotic genome formed after exchange of germline nuclei between conjugating partners. The subsequent differentiation of the somatic genome involves massive genome reorganization, which includes fragmentation of the chromosomes and elimination of a large fraction of the germline-derived sequence. In the ciliate Tetrahymena thermophila, more than 6,000 dispersed loci, comprising nearly one-third of the genome, are eliminated [2]. These internal eliminated sequences (IESs) consist of both unique and repetitive sequences that are most likely evolutionarily derived from the movement of transposable elements. DNA elimination serves as an effective genome surveillance mechanism that silences these potentially deleterious sequences by removing them from the transcribed nucleus [reviewed in 3].
The eliminated sequences are targeted for excision by small-RNA-directed heterochromatin formation. The targeting small RNAs (called scan RNAs) are produced during meiosis of the micronucleus and then assembled into effector complexes containing the argonaute/Piwi-related protein, Twi1 [4–6]. This mechanism is the evolutionary equivalent of the piRNA pathway, which employs small RNAs to silence transposons in the germline of multicellular organisms [see 7,8]. In Tetrahymena, the scan RNA-Twi1 complexes enter developing macronuclei during post-zygotic development and direct histone H3 lysine (K)9 and K27 tri-methylation (me3) to homologous regions [9,10]. The modified chromatin is recognized first by chromodomain proteins Pdd1 and Pdd3 [11,12] and then by additional proteins [13]. Finally, the domesticated piggyBac transposase, Tpb2, excises the IESs [14].
The widespread distribution of IESs throughout germline chromosomes, together with the high gene density of the somatic genome (the average intergenic region is 1 kbp) [15], necessitates accurate removal of the IESs to prevent loss of important coding or regulatory sequences. Previous work has revealed that cis-acting sequences located in the DNA flanking each IES specify excision boundaries [16–19]. Even so, the functionally equivalent controlling sequences of different characterized IESs share no obvious sequence similarity. The best studied of these cis-acting sequences is a polypurine tract (5’ AAAAAGGGGG 3’ or A5G5) located 45–50 bp outside each excision boundary of the extensively studied M IES [16]. These sequences on each side of the eliminated region reside in opposite orientation such that the G5 portion is proximal to the IES. This A5G5 tract is both necessary and sufficient to direct accurate excision [16,20]. However, the actual mechanism by which this critical sequence defines the M IES boundaries is unknown.
Here, we show that deletion of the novel gene LIA3 abolishes accurate excision of both the M IES and other IESs flanked by A5G5 tracts. Furthermore, we show that the Lia3 protein binds the M IES A5G5 boundary determinant when it adopts a non-canonical Guanine quadruplex (G4 DNA) structure. G4 DNA forms when Hoogsten base pairs stabilize interactions between four strands each composed of runs of three or more Gs [21]. G4 DNA may form during DNA replication, transcription, or other circumstances that free DNA strands from the double helix; however, in vivo evidence for formation of G4 DNA and its regulatory functions is limited. Studies have indicated that cells need to effectively manage sequences that have the potential to form G4 DNA to ensure genetic and epigenetic stability [22,23]. Furthermore, a G4-DNA-forming sequence was found to be critical for antigenic variation in Neisseria gonorrhoeae, illustrating that DNA elements that form non-canonical structures are indeed functional [24].
Early evidence that G4 DNA can form in eukaryotic cells came from studies of the telomeres of the multicopy nanochromosomes of Stylonychia lemnae in which telomeric G4 DNA and telomere binding proteins were shown to mediate attachment to the nuclear envelope [25,26]. The abundance of Stylonychia telomeres permitted ready detection of G4 DNA with the aid of structure-specific antisera. By using a similar approach, G4 DNA was more recently detected in vivo in multiple eukaryotic species including in mammalian cells [27–29]. Identification of proteins that bind and/or unwind G4 DNA has provided further evidence that these structures likely serve functional roles in vivo [21,30,31]. The in vitro binding and in vivo genetic data presented here identify a new role for G quadruplexes, in the control of genome-wide DNA elimination, and demonstrate clearly that such non-canonical DNA structures function in genetic regulation.
In our search for proteins that are important for the differentiation of the somatic genome, we identified candidates, including Lia3, that are expressed specifically during conjugation and localize to developing macronuclei [13]. Lia3 is a novel protein, which only has obvious similarity with three other Tetrahymena proteins of unknown function. To determine whether Lia3 has a critical role in macronuclear development, we created LIA3 knockout (ΔLIA3) strains lacking all germline and somatic copies of LIA3. We confirmed the replacement of the LIA3 coding region with the neo3 paromomycin-resistance cassette through genetic crosses and Southern blot analysis (S1 Fig), and loss of LIA3 expression by using rtPCR (Fig 1A). When we mated two LIA3 knockout lines together, we found that they completed all stages of development, reaching the wild-type (wt) end-point of conjugation, having resorbed one of the two micronuclei (Fig 1B); however, when mated ΔLIA3 cells were returned to growth media, only 15% of mated pairs produced viable progeny, whereas 70% of wt pairs did so (Fig 1C). These results indicated that LIA3 participates in, but is not essential for, development.
During macronuclear development, the germ-line derived genome is extensively reorganized and nearly one-third of the DNA is eliminated. To assess whether DNA elimination occurred efficiently in ΔLIA3 conjugants, we monitored the excision of a well-characterized locus containing two eliminated sequences, the M and R IESs. The M IES exhibits alternative excision, eliminating either 0.6kbp (Δ0.6) or 0.9kbp (Δ0.9) (Fig 2A). By using PCR primers outside the IES, we could detect both rearranged and unrearranged loci (Fig 2B). As all parent lines used in this study possessed only the Δ0.9 form in their macronuclei, detection of the Δ0.6 form during conjugation revealed if and when new excision had occurred in differentiating nuclei. Upon mating wt cells, M IES excision began by 12 hrs of conjugation, evident by a doublet of ~600 bp bands (Fig 2B); In contrast, M IES excision in ΔLIA3 mating cells was both delayed and aberrant, as newly excised forms were not observed until 16hrs after initiation of mating, and when observed, a ladder of PCR products was visible instead of the doublet (Fig 2B). We did not observe similar aberrancy in R IES elimination due to loss of Lia3. R IES excision may be delayed in ΔLIA3 matings, as the DNA fragment representing the unrearranged form was more abundant between 10 and 18 hrs than in wt, but this could not be unambiguously determined because de novo rearrangement of this IES cannot be distinguished from the DNA present in the parental macronuclei (Fig 2D and 2E). Nevertheless, no aberrant excision was evident, suggesting that the loss of LIA3 affects the accuracy of excision of only one of these two IESs.
We initially observed aberrant M IES excision in ΔLIA3 mating populations for which only a portion of cells survived. To determine whether the defective excision detected occurred primarily in the fraction of the population that died, we also examined M and R IES excision in individual surviving progeny cells. The nine individual progeny lines from ΔLIA3 crosses examined possessed an array of M IES excision products, which reflects the aberrancy observed within the full mating population (Fig 2C and 2B). Excision of the R IES again appeared to be largely unaffected (Fig 2F). Thus, aberrant excision was not limited to the ΔLIA3 progeny that died as cells with heterogeneous excision boundaries survived conjugation.
To determine how IES boundaries are positioned in the absence of Lia3, we cloned and sequenced a number of the M IES junctions of wild-type and ΔLIA3 progeny. The boundaries of eliminated DNA can be positioned hundreds of base pairs upstream or downstream of the major wild-type boundaries (Fig 2G). This aberrant elimination could occur due to improper cleavage of the genome by Tpb2 or, alternatively, cleavage could be normal, but the rejoining that must occur subsequent to cleavage could be perturbed. We took advantage of the observation that excised IESs will circularize to map presumed sites of cleavage in both wt and mutant cells [32,33]. The junctions of these circular, excised IESs were recovered by using PCR primers complementary to the excised regions to amplify outward across the joined ends, thus allowing us to map excision boundaries (see Fig 2H). In wt matings, circular products were observed starting either at 10 or 12 hrs into conjugation. In ΔLIA3 matings, R IES circular products of the predicted size were detected at the same hour into conjugation as they were found in wt matings, but M IES circular products appeared much later and were variable in size (Fig 2I and 2J and S1 Table). The IES boundaries of these circular products showed similar map positions as the boundaries of the rejoined DNA in progeny (Fig 2B and S1 Table). This observation is consistent with aberrancy in the cleavage of the M IES from the genome.
To determine why loss of LIA3 affects M IES but not R IES excision, we tested whether the perturbation in ΔLIA3 mutants was due to impaired recognition of the M IES or specification of accurate boundaries. Recognition of IESs occurs when complementary scan RNAs match regions of the developing somatic genome and mark them for elimination [4,34], whereas boundaries are determined by sequences flanking each IES, which are retained after excision [16–19]. Two lines of evidence indicate that the M and R IESs differ from one another in both their recognition requirements and boundary determinants: 1) Ema1, an RNA helicase that participates in scan RNA/Twi1 recognition, is required for M but not R elimination [35]; and 2) M and R IESs have functionally distinct and incompatible boundary-controlling sequences [16,17,20]. To determine whether Lia3 acts to identify the eliminated region of the M IES or its flanking boundary sequences, we generated chimeric IESs and tested their excision in both wt and ΔLIA3 conjugants. These chimeras contained the eliminated region of either the M, R, or a segment of the transposon-like TLR IES [36,37] inserted between either the M or R boundary-controlling flanking sequences. These chimeras were introduced into conjugating cells during nuclear differentiation on rDNA-based replicating vectors, and IES excision was monitored by Southern blot analysis of the transformant DNA. When any of the eliminated sequences, including that of the M IES, was positioned between the R IES’s flanking sequences, the chimera was accurately excised using the normal R IES boundaries, even in ΔLIA3 matings (Fig 3A–3D). In contrast, when any of these IESs was positioned between the M IES’s flanking DNA, each IES was correctly and efficiently excised in wt matings, but not in ΔLIA3 matings (Fig 3E–3H). We observed a significant decrease in excision efficiency for the M-flanked M and R IESs (Fig 3F and 3G), whereas the M-flanked TLR IES was efficiently deleted, but its excision lacked clearly defined boundaries, evident as a ladder of products (Fig 3H). The decrease in excision efficiency that we observed in ΔLIA3 progeny coincides with the decreased M IES excision observed upon deleting or otherwise mutating polypurine tracts flanking the M IES [16,20]. These experiments demonstrate that Lia3 acts in concert with the M IES flanking DNA to specify the boundaries of excision, but does not discriminate between the different IESs placed between these controlling sequences.
The boundary-controlling flanking sequences of the M IES consist of polypurine tracts, 5’-A5G5-3’, located approximately 45bp away from the major boundaries [16]. This A5G5 sequence is not present in the flanking region of the R IES, leading to our hypothesis that Lia3 interacts with this polypurine tract to determine the position of each excision boundary. If true, Lia3 represents the first protein known to position these boundaries. We first tested this possibility by identifying other IESs with similarly positioned polypurine tracts and assessed whether their excision was aberrant in ΔLIA3 progeny. All four additional IESs with polypurine tracts located near their boundaries exhibited aberrant excision in progeny of ΔLIA3 matings (Figs 4A, 4B and 4C and S2A) whereas the several other IESs tested that lacked obvious polypurine tracts were not affected by loss of LIA3 (Figs 4D, 4E and 4F and S2B–S2F). These findings are consistent with our analysis of chimeric IESs (Fig 3) that showed that diverse IESs are affected by loss of LIA3 only when flanked by G-rich polypurine tracts. Thus Lia3 appears to specifically control the excision boundaries of a class of IESs containing flanking polypurine tracts.
The obvious interpretation of our data is that the novel protein Lia3 directly binds to the M IES polypurine tracts and controls the extent of excision. To test the ability of Lia3 to bind DNA, we purified the protein after expression in E. coli (S3 Fig) and used it in electrophoretic mobility shift assays (EMSA). Initially, we incubated Lia3 with single stranded (ss) or annealed (ds) 30 nt oligonucleotides corresponding to either the leftmost M IES boundary (M1), centered on the A5G5 tract, or the equivalent region from the leftmost R IES boundary (R1) (Table 1). Lia3 bound strongly to ssM1 and weakly or not at all to the dsM1, ssR1, and dsR1. To confirm specificity, we used unlabeled oligonucleotides to attempt to compete away weaker interactions. The ssM1 oligonucleotide competed effectively for the initial binding observed when using the ssR1 and dsR1 substrates, whereas the ssR1 and dsR1 oligonucleotides could not compete for the binding to ssM1, indicating that the interaction that Lia3 had the highest affinity for the ssM1 probe (Fig 5A).
In these assays, the majority of the unbound ssM1 oligo exhibited an altered electrophoretic mobility, migrating significantly slower than expected, and it was this form of the probe to which Lia3 preferentially bound (Fig 5A, black arrow). This purine-rich oligonucleotide contains a run of Gs, leading us to test the possibility that the probe had adopted G4 DNA structure (Fig 5B). G4 DNA is known to form readily in the presence of KCl, but poorly in the presence of LiCl or without cations [38], so we denatured the oligonucleotide by boiling in either 10 mM Tris-HCl (ph 7.5) alone or supplemented with either 100 mM KCl or 100 mM LiCl, then slow cooled to room temperature before native gel electrophoresis. The M1 oligonucleotide migrated as expected for ssDNA in buffer without salt or with LiCl, but abnormally slow in the presence of KCl (S4A Fig). Mutation of either the first three Gs within the A5G5 segment to Cs, mutations known to abolish boundary function [20], or even simply changing the second G to C, was sufficient to prevent the M1 oligo from forming a higher-ordered structure (S4 Fig). All these observations are consistent with a G4 DNA structure. We confirmed that 30 nt, A5G5-containing oligonucleotides representing either the M1 and M2 flanking region (the sequences from the two alternative left side M IES boundaries centered around the A5G5, [16]) formed quadruplex structures by performing circular dichroism (CD) [39] (Figs 5C and S4). Parallel G4 DNA exhibits a diagnostic positive peak at 260nm and a negative peak at 240nm [39]. Both the M1 and M2 oligonucleotides displayed CD spectra diagnostic with formation of parallel G4 DNA when in the presence of KCl, but not, at least for M1, when in the presence LiCl (S4 Fig). This observation further supports our conclusion that these probes formed a quadruplex in the conditions used in our EMSA.
To rule out the possibility that a co-purifying contaminant in the extract was responsible for quadruplex binding, we performed parallel purification of Lia3 and the MS2-coat protein and used each in EMSA. Our initial binding assays were performed with a histidine-tagged Lia3 protein, which required denaturing lysis to recover from E. coli. We subsequently expressed Lia3 with a maltose binding protein (MBP) fused to its amino terminus, which allows purification in non-denaturing conditions, and isolated MBP-Lia3 along with MBP-MS2 (S5 Fig). The MBP-Lia3 specifically bound the G4 DNA M1 probe formed in the presence of KCl, but not the ssMI probe (in LiCl) (Fig 6). The specificity of Lia3 for G4 DNA is further revealed by the observation that none of the residual ssM1 DNA (lower band) remaining in KCl treated probe samples was shifted upon addition of protein (Figs 5 and 6). The MBP-MS2 protein bound neither the probe in LiCl or KCl. Excess unlabeled M1 oligonucleotide, but not the C3G2 mutant oligonucleotide, could compete away MBP-Lia3 binding, which further supports that Lia3 preferentially binds G4 DNA.
To further assess the specificity of Lia3 for parallel G4 DNA, we measured binding affinity of Lia3 to M1 G4 DNA, ssM1, dsM1, or Tetrahymena telomere sequence, which is known to form mixed quadruplex structures (Figs 6B and S6). We determined that the Kd of Lia3 for the M1 quadruplex was 144 nM. It also bound to the telomere quadruplex, but with lower affinity than to the M1 quadruplex (Kd = 11.5 μM). Lia3 had much higher affinity for either quadruplex probe than for the ss or ds linear forms of the M1 probe (extrapolated Kd over 0.2 mM). In competition experiments, oligonucleotides forming parallel G4 DNA (M1 or M2) were able to compete away the interaction of Lia3 with the M1 quadruplex, whereas linear oligonucleotides did not compete for binding, which further shows that Lia3 binds specifically to parallel G4 DNA in vitro (S7 Fig). It is important to note that addition of LiCl to the binding reaction did not inhibit Lia3 binding to the M1 probe when it was pre-assembled into the quadruplex form (S7 Fig–ss competitor).
Together, our genetic and biochemical analyses indicate that Lia3 binds to a parallel G quadruplex that forms near the boundaries of the IESs flanked by A5G5 sequences to direct accurate excision. We attempted to directly detect these structures in developing macronuclei using available anti-G4 DNA antibodies without success. We could detect putative G4 DNA in the macronuclei of unmated cells and the parental macronuclei of late stage conjugants, possibly due to the very abundant telomeres (S8 Fig). The failure of this approach may indicate that Lia3 binding masks the G4 DNA epitope in developing macronuclei or that the amount of these structures present is below the level needed for detection with these reagents.
Although it is easy to envision how intermolecular association of four oligonucleotides can allow formation of G4 DNA in our gel shift assays, how a four-stranded structure might form at chromosomal loci given that each side of the IES contains a single run of Gs is less obvious. We investigated one possibility, that two of the four strands could be RNA. Hybrid DNA/RNA quadruplexes can form during transcription [40], and transcription of IESs occurs before their excision [35,41]. Transcription would unwind the flanking G tracts, freeing them to interact with other G-rich strands. In this model, the non-coding transcripts created provide the two additional strands needed to complete this structure. To test whether RNA is available to participate in defining M IES boundaries, we used rtPCR to look for transcripts at the time that the Lia3 protein accumulates (S9 Fig) and detected RNAs that span the A5G5 tract (Fig 7A). We also found that RNA oligonucleotides with the M1 flanking region sequence can form quadruplexes, and that these RNA quadruplexes can compete for Lia3 binding to the M1 G4 DNA probe (S10 Fig). Although these findings support the possibility that non-coding transcripts participate in controlling the boundaries of eliminated sequences, they certainly do not exclude other mechanisms discussed below.
The polypurine tracts flanking the M IES were first shown to control its excision boundaries 25 years ago [16]. Despite the identification of similarly positioned controlling sequences flanking other IESs, how these diverse cis-acting sequences are recognized has remained a mystery. We show here that Lia3 is required to accurately excise the M IES and other IESs possessing flanking polypurine tracks. Lia3 is a novel protein, expressed exclusively during post-zygotic development. In our efforts to characterize its binding to DNA, we discovered that it binds specifically to parallel G4 DNA formed by the M IES A5G5 sequence. As both this sequence and Lia3 determine IES boundaries, our data strongly support our hypothesis that G4 DNA can form at internal chromosomal loci (not just telomeres) and define specific regulatory domains.
Although we were not surprised that the A5G5-containing oligonucleotides we used as EMSA probes could form G4 DNA, we did not expect that Lia3 would preferentially bind this structure. Each side of the IES has a single G5 tract, and formation of a quadruplex would require four independent copies to come together. The simplest way we can envision this forming in vivo involves the transcription from the C5 strand, providing G5-containing RNA copies that participate in quadruplex formation such that the quadruplex includes the G5 DNA tracts on each side of the IES and the two RNA strands (Fig 7B). Four strands of the flanking regulatory A5G5 DNA are also produced during a round of DNA replication that precedes DNA elimination (Doerder and Debault 1975). If RNAs are not part of the quadruplex, it is likely that the G tracts on each side of the IES from both sister chromatids form the G4 DNA structure. It is also possible that the G tracts of different A5G5-controlled IESs interact to form a single quadruplex.
By showing that Lia3 is both a parallel G4 DNA binding protein and a specific regulator of the excision of IESs containing A5G5 tracts, we report a compelling case for a role for non-canonical DNA structures in regulating genome organization. The G4 DNA bound by Lia3 appears to bring together G tracts located on each side of the IESs. The formation of G4 DNA through the association of distal G tracts located on different DNA strands is not the obvious outcome, and therefore our findings elucidate an unforeseen potential of dispersed, G-rich DNA sequences to interact. The proposed involvement of transcription in this structure serves two purposes: to unwind the DNA to allow the G tracts to interact with distal partners, and to provide additional G-rich strands to promote a four-stranded structure to form. If such a predicted structure forms in vivo, Lia3’s ability to bind these structures may permit this protein to serve as a probe for such structures in genomes beyond Tetrahymena. Long non-coding RNAs and non-genic transcription appear to be prevalent in genomes. The model we present in Fig 7 suggests a novel mechanism for these RNAs to interact with DNA and affect chromosomal DNA organization. We believe the ability of Lia3 to bind novel quadruplex structures represents another case in which studies of ciliate genome rearrangements have uncovered new regulatory potential in eukaryotes.
To assist the quadruplex formation between distal G tracts, we propose that the formation of heterochromatin (i.e., establishment of H3K9me3 and H3K27me3) across the IES and subsequent binding of chromodomain-containing proteins Pdd1 and Pdd3, and other DNA excision proteins, position the IES flanking regions in proximity to one other. This organization of IES chromatin aids the formation of the quadruplex, which is stabilized by Lia3 binding. This proposal is consistent with data showing that G4 DNA is enriched in the heterochromatic regions in Drosophila polytene chromosomes [28]. The interaction of distal G tracts on different strands represents a novel mechanism to partition chromosomal loci into distinct domains. Once bound, Lia3 guides the domesticated transposase Tpb2 to preferential boundary sites either by directly interacting with the transposase or simply preventing it from cutting elsewhere.
Multiple results from mutational analyses provide evidence that cis-acting sequences on each side of an IES interact with one another. For instance, deletion or other disruption of the boundary-controlling sequence on one side of an IES did not lead simply to inaccurate specification of the boundary on the mutated side of the IES, but instead severely decreased overall rearrangement efficiency [16,17]. Furthermore, chimeric IESs containing one M and one R IES flanking sequence did not exhibit excision at the native M and R boundaries present in the construct, but instead used the native M boundary on one side and a novel boundary that is 45–50 bp away from a cryptic A5G5 tract present by chance within the R IES sequence [16]. These data are consistent with our model in which G5 tracts on each side of the IES come together to form parts of a common structure. Coupled cleavage on both sides of an IES may have been selected for during the domestication of the Tpb2 piggyBac transposase to ensure accurate excision and prevent aberrant double-strand breaks during genome-wide DNA elimination events. Coordinated cleavage on both sides of an IES occurs in the ciliate Paramecium [42,43], which also uses a domesticated piggyBac to perform its genome rearrangements [44], indicating that communication between IES ends is a conserved mechanism.
Although we favor a model in which IES heterochromatin is established, and subsequent organization of this chromatin structure helps to bring distal A5G5 sequences together to form a G quadruplex, we cannot rule out the possibility that Lia3 stabilizes this structure prior to the completion of these chromatin modifications and acts to limit the spread of small-RNA-directed heterochromatin. In the future, we will determine the enrichment of H3K9me3 and H3K27me3 across the developing genome in the presence and absence of Lia3 to assess whether the cis-acting sequences that control IES boundaries actually serve as barrier elements blocking the spreading of chromatin modifications.
Only a subset of IESs have flanking A5G5 tracts and are controlled by Lia3; yet there are thousands of IESs, which vary greatly in size and sequence, that are faithfully excised during differentiation of the somatic genome. The adjacent M and R IESs are known to use functionally distinct boundary-controlling sequences [16,17]. The use of distinct cis-acting sequences by neighboring IESs would prevent aberrant elimination events between the distal ends of adjacent IESs. Some likely candidates to define the ends of the non-A5G5 IESs are three Tetrahymena proteins with homology to Lia3. Like LIA3, each is expressed exclusively during post-zygotic development [45], and the two of which that we have examined localize to developing macronuclei (S11 Fig). Although these Lia3-like (LTL) proteins do not have obvious homologs in other organisms, database annotation indicates that their amino termini possess similarity to DNA binding proteins [46]. In our preliminary investigations, disruption of LTL1 (Ttherm_00499370) results in aberrant excision of several non-Lia3 regulated IESs. It will be interesting to determine whether these related proteins control the boundaries of other IESs by binding to other non-canonical DNA structures. Their study could provide evidence for novel mechanisms used to bring together cis-acting sequences to define specific regulatory domains within genomes.
Tetrahymena cells were grown at 30°C in either SPP or Neff’s medium under standard conditions [47,48]. Strains CU428 (Mpr1-1/Mpr1-1 [VII, mp-s]), B2086 (II), and CU427 (Chx1-1/Chx1-1 [VI, cy-s]), B*VI (VI), and B*VII (VII), were used to construct knockout strains or were transformed with rDNA constructs. Strains B-VII-427(Chx1-1/Chx1-1 [VII, cy-s]) and B2086 were used for excision assays because both contain only the Δ0.9 form of the M IES in their macronuclei. To promote synchronous mating, cells were starved at 30°C overnight in 10 mM Tris, pH 7.5, prior to mixing at equal cell densities (~2.5x105 cells/ml).
Individual pairs, >6 hrs after initiating mating, were transferred to individual drops of SPP and allowed to complete conjugation. Drops containing living cells after 2 days were transferred to 96 well plates containing starved CU428 or B2086. Throughout the day wells were screened for mating pairs. Wells containing paired cells indicated that the initial drop plates had contained back-outs instead of progeny as progeny would not be sexually mature yet. Survival was scored as the percent of drops containing progeny versus the number of drops plated. To score conjugation endpoint, cells were fixed with 2% paraformaldehyde 24 hrs into mating and stained with DAPI.
Total RNA (4μg), isolated by RNAsol extraction [51], was converted to cDNA using SuperScript II reverse transcriptase as described [50]. PCR was performed using Lia3rt_FW and Lia3rt_RV primers (S2 Table) to monitor LIA3 expression and HhpI_FW and HhpI_RV primers (S2 Table) to monitor HhpI expression as a loading control.
gDNA was isolated from mating cells at indicated times. Detection of IES junctions was performed by using PCR primers that amplify across the IES junction as described [2]. Detection of excised IES circles was performed by nested PCR using primers (S2 Table) pointing outward from the IES [52] [53]. PCR products were gel isolated and then TA cloned prior to sequencing.
Electroporation of wild-type or ΔLIA3 mating cells with IES-containing rDNA vectors was performed as described [17,54]. Plasmids containing M or R IES sequences or chimeric IESs (pMgtwM_m, pMgtwM_r, pRgtwR_m, pRgtwR_r) were created by first replacing the IES sequence with a gateway recombination cassette then recombining the desired IES sequence into the desired vector. For Southern blot analysis, three individual paromomycin-resistant progeny lines, obtained after electroporation of wild-type or mutant mating cells with IES-containing vectors, were co-cultured for genomic DNA isolation. Ten μg of each DNA preparation was digested with either NotI or BamHI, fractionated on 1% agarose gels, transferred to nylon membranes and hybridized to M or R IES (from pDLCM3 or pDLCR5, respectively) [55].
DNA encoding an N-terminal His tagged Lia3 (His-Lia3) was codon optimized for expression in E. coli by Life Technologies. His-Lia3 was cloned into NcoI and XbaI sites of pBAD (a gift from Dr. R. Kranz, Washington University) to make pBAD-HisLia3. His-Lia3 was expressed in E. coli strain BL21(DE3). After reaching an OD600 ~0.8, 0.2% wt/vol Arabinose was added and cells continued to grow for 4hrs before harvesting. Cell pellet was resuspended in native lysis buffer (50 mM NaH2PO4, 300 mM NaCl, 10 mM Imidazole, and protease inhibitors) and lysed using a French press (1200 psi). Cells were centrifuged at 40,000 rpm for 45min at 4°C and the pellet was resuspended in denaturing buffer (100 mM NaH2PO4, 10 mM Tris-HCl pH 7.5, 8M Urea, and protease inhibitors) and stirred on ice for 1 hr. Cell lysate was spun at 10,000 x g for 30 min and the supernatant was incubated with Ni-Nta resin for 1 hr before loading onto the column. After washing with 50 mM NaH2PO4, 300 mM NaCl, and 20 mM Imidazole, proteins were eluted in 50 mM NaH2PO4, 300 mM NaCl, and 125 mM Imidazole and subsequently dialyzed against 25 mM Tris pH 7.5, 100 mM KCl, 1 mM DTT, 1 mM EDTA, and 10% glycerol before storage at -80°C.
The His-Lia3 coding sequence was PCR amplified using oligonucleotides listed in S2 Table to add an N-terminal TEV protease cleavage sites and BamHI and HindIII sites. The amplified DNA was cloned into the pMAL-C2X expression vector (New England Biolabs, Ipswich, MA) to create pMAL-TEV-hisLIA3. The plasmid was transformed into BL21(De3) cells and the recombinant protein was purified by using an amylose resin as described [56] and then dialyzed against 25 mM Tris pH 7.5, 100 mM KCl, 1 mM DTT, 1 mM EDTA, and 10% glycerol before storage at -80°C.
Oligonucleotides (Table 1) were labeled by incubation with T4 PNK and [γ -32P] ATP for 1 hr at 37°C and then purified using Roche oligo spin columns. Oligos were made double-stranded by mixing equal amounts of complementary oligonucleotides in 10 mM Tris pH 7.5, 5% glycerol, and 100mM LiCl and boiling for 5 min, followed by slowly allowing the oligonucleotides to cool to RT. Prior to gel shifts, oligonucleotides were boiled for 5 min in 10 mM Tris pH7.5, 5% glycerol, and either 100 mM KCl or 100 mM LiCl and slow cooled to RT to allow structures to form.
For binding and competition experiments, 50–400 nM His-Lia3, MBP-Lia3, or MBP-MS2 was incubated with unlabeled competitor oligonucleotides (Table 1) in binding buffer (10 mM Tris pH 7.5, 1 mM EDTA, 0.1 mM DTT, 5% vol/vol glycerol, 0.010 mg/ml BSA, and either 100 mM KCl or 100 mM LiCl) for 15 min at RT. KCl was used in all binding reactions except when LiCl was use to limit quadruplex formation. Followed by addition of 4 nM 32P-labeled oligonucleotide and incubation for another 15 min at RT before 4 μl was loaded onto a 4.5% polyacrylamide (75:1 acrylamide:bisacrylamide) gel. After pre-running gels at 140V for 30 min, samples were fractionated by electrophoresis at 140V for 1 hr 45 min. Gels were subsequently vacuum dried for 1 hr prior to exposure to X-ray film or to a Phosphorimager screen. For binding curve experiments, 4 nM 32P-labeled oligo was incubated with 0, 50, 75, 100, 150, 200, 250, 300, 400, 500, 700, 900, 1100, 1300, 1500, 1900, 2100, 2500, or 3000 nM His-Lia3 for 20 min at RT prior to loading on gel.
Oligonucleotides were boiled for 5 min in 10 mM Tris pH 7.5, and either 100 mM KCl or 100 mM LiCl and slow cooled to 4°C. The CD spectra were recorded on a J-810 spectropolarimeter (Jasco). The measurements were carried out with 500 μL 3 μM ODN samples at 4°C under nitrogen. Spectra shown are the average of 3 scans in a range from 220 to 300 nm with a band width of 1 nm, response time of 0.5 s, data pitch of 0.2 nm, and scan speed of 50 nm/min. A blank sample of 10 mM Tris pH 7.5 with 100 mM KCl or 100 mM LiCl was used for baseline correction.
Strains CU428 and B2086 were starved in 10 mM Tris-HCl (PH 7.5) and mixed to induce mating. Between 9 and 10 hours post-mixing, mating Tetrahymena cells were fixed in 3% PHEMS-paraformaldehyde essentially as described [57], incubated overnight with a 1:200 dilution of anti-G4 antisera (1H6- EMD Millipore, Billerica, MA) [27], and detected with Alexa 488-conjugated, goat, anti-mouse antisera. Cells were counterstained with DAPI (4',6-diamidino-2-phenylindole) and imaged on a Nikon E600 epiflourescent microscope equipped a Retiga EX CCD camera (Q imaging, Burnaby. B.C. Canada) with Openlab acquisition software v404 (Improvision).
|
10.1371/journal.pntd.0005557 | Yaws resurgence in Bankim, Cameroon: The relative effectiveness of different means of detection in rural communities | Yaws is an infectious, debilitating and disfiguring disease of poverty that mainly affects children in rural communities in tropical areas. In Cameroon, mass-treatment campaigns carried out in the 1950s reduced yaws to such low levels that it was presumed the disease was eradicated. In 2010, an epidemiological study in Bankim Health District detected 29 cases of yaws. Five different means of detecting yaws in clinical and community settings were initiated in Bankim over the following five years.
This observational study reviews data on the number of cases of yaws identified by each of the five yaws detection approaches: 1) passive yaws detection at local clinics after staff attended Neglected Tropical Disease awareness workshops, 2) community-based case detection carried out in remote communities by hospital staff who relied on community health workers to identify cases, 3) yaws screening following mass Buruli Ulcer outreach programs being piloted in the district, 4) school-based screening programs conducted as stand-alone and follow-up activities to mass outreach events, and 5) house to house active surveillance activities conducted in thirty-eight communities. Implementation of each of the four community-based approaches was observed by a team of health social scientists tasked with assessing the strengths and limitations of each detection method.
Eight hundred and fifteen cases of yaws were detected between 2012 and 2015. Only 7% were detected at local clinics. Small outreach programs and household surveys detected yaws in a broad spectrum of communities. The most successful means of yaws detection, accounting for over 70% of cases identified, were mass outreach programs and school based screenings in communities where yaws was detected.
The five interventions for detecting yaws had a synergistic effect and proved to be valuable components of a yaws eradication program. Well planned, culturally sensitive mass outreach educational programs accompanied by school-based programs proved to be particularly effective in Bankim. Including yaws detection in a Buruli Ulcer outreach program constituted a win-win situation, as the demonstration effect of yaws treatment (rapid cure) increased confidence in early Buruli ulcer treatment. Mass outreach programs functioned as magnets for both diseases as well as other kinds of chronic wounds that future outreach programs need to address.
| Yaws is an infectious and disfiguring disease of poverty primarily affecting children in rural communities in tropical areas. Yaws is easily treated by a single dose of antibiotics and is on the World Health Organization’s eradication list. Yaws was thought eradicated in the Cameroon in the 1950s following aggressive mass-treatment campaigns. In 2010, epidemiological research revealed a resurgence of the disease. This paper discusses the relative success of five different means of detecting yaws in rural areas of Bankim District between 2012 and 2015. While few cases of yaws were detected at local clinics during this time, many cases were detected in the community. The most successful means of detecting yaws were mass outreach programs designed to educate the public about neglected tropical diseases found in the region, and follow up school-based screening programs. These programs were supported by local chiefs and traditional healers and found to be the best way of increasing community awareness about yaws, motivating community health workers to participate in outreach, and fostering trust in the free medical treatment being provided.
| Yaws is an infectious, debilitating, and disfiguring disease of poverty that mainly affects children and adolescents living in rural communities in tropical areas of Africa, the Pacific Islands, and Southeast Asia with high levels of rainfall. Caused by the spirochete bacteria Treponema pallidum, subspecies pertenue is closely related to syphilis and one of three endemic non-venereal treponemal diseases. The bacterium causes a chronic relapsing treponematosis characterized by highly contagious primary and secondary cutaneous lesions and non-contagious tertiary destructive lesions of the bones. Humans are the primary reservoir for yaws and transmission occurs through skin to skin contact with the exudate of lesions by those who have skin abrasions or cuts. Yaws is usually contracted in childhood (75% of cases occur before age 15) and infectious lesions are infrequent after the age of 30 [1, 2].
In the early stage of the disease, which may last from weeks to months, skin lesions are highly contagious and present differently by season with more open infectious lesions and papillomatous frambesides in the wet season and drier, scalier, maculopapular lesions in the dry season. Painful and itching lesions commonly appear on the upper and lower limbs, fingers, toes, soles of the feet, face, genital areas, and buttocks. The early stage is typically characterized by a single elevated primary lesion that develops a crust that is shed, followed by secondary lesions on other parts of the body. After 3–4 months lesions subside due to host immune response. The disease then becomes latent. In about 10% of untreated patients, the infection progresses to the tertiary stage characterized by destruction of tissue, bone, and cartilage resulting in disfigurement and disability.
Once widespread in the tropics, the incidence of yaws has been controlled though a combination of mass treatment with single dose of antibiotics (injectable benzathine benzylpenicillin) along with better hygiene and access to clean water. It has been estimated that yaws control efforts mounted by the World Health Organization (WHO) and United Nations International Children's Emergency Fund (UNICEF) resulted in up to a 95% reduction of the disease burden worldwide. Efforts are currently underway to eradicate the disease by 2020 following the Morges strategy, which calls for an initial mass treatment of endemic communities with Azithromycin in tablet form [2] followed by ongoing active community-based surveillance system and if required surveys every 3–6 months to detect and treat remaining cases of yaws and their contacts [3]
Yaws continues to be endemic in at least 13 countries globally, of which Cameroon is one [4]. Eradication will require better surveillance, health worker training, community outreach, and targeted mass drug treatment when and where necessary. In Cameroon, mass-treatment campaigns carried out in the 1950s reduced yaws to such low levels that it was presumed the disease was eradicated except among groups of pygmies living in the dense forest. In 2007 and 2008 outbreaks of yaws occurred among pygmy groups in Lomié health district. Cameroon’s National Neglected Tropical Disease (NTD) Control Program (covering Buruli ulcer (BU), leishmaniasis, yaws, and leprosy) working in conjunction with the NGO FAIRMED carried out an epidemiological survey in the district of Lomié in 2009. One hundred sixty-seven cases of yaws were detected in 35 small communities surveilled. Seventy five percent of cases were children under the age of 15 years with a majority between 9–11 years of age.
Yaws surveillance was not included in the routine disease surveillance system elsewhere in Cameroon and assumed to be a problem confined to the pigmy population. This changed when an epidemiological study of leprosy, yaws, and BU was carried out in Bankim district in 2010. The study entailed an intensive house-to-house survey conducted in late March to mid-April during which time 9,344 households were visited and 48,962 people examined. Twenty-nine confirmed cases of yaws were detected [5]. It became evident that either those afflicted with yaws were not coming to clinics for treatment or health staff were failing to recognize and treat the disease, confusing it perhaps for scabies.
As a follow up to the survey, three day NTD workshops were conducted by the National Disease Control Program in 2012 in Bankim and surrounding districts attended by hospital and clinic staff. The objective of the workshops was to better familiarize health workers with the signs of BU, leprosy, and yaws; encourage them to identify presumptive cases; and send swabs for laboratory confirmation. Disease control officers began visiting communities in 2012–2013 in an attempt to identify cases and inform community health workers (CHWs) about the disease. During these visits, CHWs were shown posters displaying the signs of yaws and BU and asked to identify suspected cases in their communities.
In 2013, an innovative community-based outreach program was launched in Bankim by the NGO FAIRMED working in conjunction with the government health service and the Stop Buruli Consortium. The three objectives of the outreach program were to raise awareness about BU, identify cases, especially early category one cases, and create collaborative relationships between clinic staff, CHWs, traditional healers, and local chiefs. Community health workers were mobilized and tasked with organizing mass community BU outreach events.
The culturally sensitive program that was introduced differed from previous outreach programs conducted in the Cameroon. In the past, information about BU was conveyed from health staff to the local population in a top down manner without community feedback elicited. The innovative program being piloted drew upon a year of formative research carried out by teams of social scientists in Bankim on patterns of health care seeking for BU and other chronic ulcers. The education program introduced went well beyond educating the public about the signs of BU. It employed a question and answer format that encouraged two-way dialogue between community members and health staff. Participants were shown before and after photographs of BU-related wounds depicting the healing process when appropriate treatment was followed. Time was allotted for testimonials by those cured of BU. Former patients attested to the quality of care they had received by clinic staff in what was described as a newly upgraded BU treatment program. Community members were also given explanations for all health staff actions including the collection of blood for disease confirmation. Following the educational program, screening by government health workers took place for those having lesions that were possible signs of BU. Although the focus of the outreach program was BU, many cases of yaws began to be detected. In communities where yaws was identified, teams returned and conducted school-based yaws screening and education programs.
This paper examines the relative utility of five approaches to yaws detection in rural settings of Cameroon:
We then present a brief overview of data collected on the distribution of yaws cases in the community and lessons learned about the best times to conduct yaws detection activities.
The study took place in Bankim district located in the northwest Adamawa region of Cameroon (Fig 1).
Bankim is situated in the Mape River Valley, where a hydro-electric dam was built more than twenty-five years ago. The Mape Dam splits the area into isolated islands and scattered communities. In the last two decades, increased irrigation has enabled rice cultivation. Inhabitants of the region also engage in growing maize, cassava, and peanuts as well as various forms of hunting and fishing. Much agriculture is done on plots of land some distance from the homes of community members during the months of January through May. Population movement and residence in the region is fluid and seasonal.
Bankim Health District is a challenging place to initiate a community outreach program due to both its rugged terrain and the wide variety of ethnic groups inhabiting the region. These groups speak a variety of languages and dialects in addition to French and Pidgin English. The district is served by a district hospital and 10 satellite clinics of which eight are government and two private and mission supported. All clinics report cases of yaws to Bankim hospital. Beginning in 2013 this hospital conducted rudimentary laboratory analysis for cases of yaws using Trepanoma Pallidum Hemagglutination Assay (TPHA), an inexpensive test indicative of, but not specific for, yaws.
Each of the five approaches for detecting yaws in Bankim health district described in this paper were implemented between January 2012 and December 2015. During this time a range of NTD related community outreach activities were ongoing involving local health staff and community health workers (CHWs) in different capacities. Health facility records of cases of yaws identified at Bankim hospital and local clinics were reviewed along with the records of yaws cases identified during small-scale NTD outreach activities documented by the disease control officer attached to Bankim Hospital.
Innovative community outreach programs for BU attracting large gatherings of community members took place from January 2013 through March 2015. During these programs trained health staff delivered culturally sensitive education programs developed and pretested by Cameroonian social scientists working with the Stop Buruli consortium. Outreach events generally attracted 400–600 participants and made use of image rich PowerPoint presentations conveying key messages about BU and its treatment. Meetings were organized by community health workers who also assisted in translating messages into local dialects. These events were attended by local chiefs, healers, and former patients who gave testimonials about the quality of care they had received.
Mass outreach events were generally held in the evening and lasted two to three hours. At the end of the event there was an opportunity for people to have their wounds screened for BU or to arrange for screening at a clinic. Local clinic staff and CHWs learned how to identify the signs of BU and yaws by observing trained hospital staff who used these events as teachable moments. CHWs were then encouraged to refer potential cases of yaws to clinics and screenings following future outreach events.
Patients suspected of having yaws were treated on the spot with free injections of benzathine penicillin. Blood samples from lesions were taken from a nonrandom sample of 120 patients by digital puncture or puncture at the heel and then transported to the laboratory of the Bankim hospital in 0.2 ml Eppendorf bottles with Ethylenediaminetetraacetic acid. Two tests were carried out by an experienced laboratory technician: rapid plasma regain (RPR) and TPHA. Both are routinely used and reactive in the screening of yaws and syphilis.
In communities where cases of yaws were detected during BU outreach events, health staff followed up with school-based yaws education and detection programs. Data on the age and gender of yaws patients were recorded along with information on school attendance and the clustering of cases in particular households and communities. In 2015, there was a gap in funding for mass BU outreach events. During this time another source of funds became available for two intensive house to house NTD surveys conducted in the months of August and November. One thousand eight hundred and eighty-nine (1,889) households residing in thirty eight communities were visited by health staff.
The National Ethics Committee for Health Research overseen by the Cameroon Ministry of Public Health Cameroon approved this study. All study participants voluntarily opted into the study through documented informed consent. In cases where children were interviewed or their blood was drawn for testing, consent forms were secured from parents after being informed why a test was being administered.
Eight hundred and fifteen (815) cases of suspected yaws were detected between 2012 and 2015 in Bankim district. Blood samples from 120-suspected cases were sent to Bankim Health District Laboratory for testing, of which 100 cases were < 16 years of age, 16 were 16–30 years of age, and four over > 30 years of age. The RPR confirmation rate was 85% and the TPHA confirmation rate 77%. It is possible that some of the remaining 23% of cases included people treated for yaws or syphilis in the past. All 815 cases were followed up four weeks after antibiotic treatment was administered. Complete recovery was observed in 89% of cases with the remaining 11% of cases found symptom free three weeks later.
Tables 1 and 2 summarize how the 815 yaws cases were detected. As a means of assessing the cumulative effect of the four community outreach activities, it may be noted that no yaws cases were recorded in Bankim district in the five years prior to 2012, when outreach activities were initiated. Furthermore, between 2012 and 2015 only four cases of yaws were reported at clinics in the neighboring district of Malentouen, although health staff in this district had attended a three-day NTD workshop in 2012 alerting them to the presence of yaws in the region. In Malentouten district, community based outreach activities had yet been introduced.
Five observations may be highlighted. First, even after yaws awareness training for health staff working in clinics and three years of outreach activities where health staff encouraged community members with yaws-like symptoms to visit clinics, only 7% of all yaws cases were detected at clinics. This suggests that community members with yaws-like symptoms are not commonly visiting clinics for treatment (see Agana-Nsiire [6] for a similar finding in Ghana). Ethnographic research confirmed this observation. The symptoms of yaws (itching and moderate levels of pain) are not seen to be serious enough to warrant seeking care at a clinic, especially if a clinic is distant. The fact that the symptoms of yaws eventually subside (as the disease becomes latent) led some community members to conclude that the disease was self-limiting, recurrent, or seasonal. Despite a rising level of awareness within the local population about yaws and the effectiveness of drug therapy resulting from outreach programs, most community members afflicted with yaws-like symptoms preferred to wait for outreach screening events rather than travel to clinics. This is evidenced by clinic data that documents only a small increase in yaws cases seen at clinics in Bankim during the four-year period.
Second, NTD outreach activities in remote communities identified yaws cases largely based on the mobilization efforts of CHWs. In 2012 and 2013 health staff visited six small to moderate sized (< 100 households) remote communities in December and January. Thirty-one cases of yaws were identified and treated. In December–January 2014, the Stop Buruli team visited another 10 remote communities (of similar size) searching for both BU and yaws cases. CHWs exposed to basic information about the two diseases were asked to identify possible cases in their community. Together with health staff, CHWs detected sixty cases of yaws. In total 11% of all cases of yaws were identified through this outreach approach, yielding a mean of 5.5 cases per outreach activity.
Third, a big spike in yaws detection occurred with the initiation of mass BU outreach program events followed by school screenings in communities where yaws was detected. These programs were held in mid-November through January, months when a majority of the local population reside in their homes and are not engaged in agricultural activities some distance away. Programs were conducted in moderate to large sized communities with schools. They were held in the early evening, attended by community leaders, and seen by community members as a major event. Light entertainment preceded the education program.
At first, the outreach team only focused on BU and did not pay much attention to other kinds of skin lesions. However, after a number of yaws cases were detected among children in 2013, the decision was made to be more proactive and screen for yaws. Between 2013 and 2015, 44 screenings were held after mass BU outreach events. When cases of yaws were identified in a community, school screenings were arranged and carried out. Three hundred and twenty-eight cases of yaws were detected with a mean yield of 9.4 case per mass event/school screening activity set. Notably, the case yield increased over time as people came to see health staff as accessible and free medication available for both yaws and BU. In 2013, 4.4 cases of yaws per activity set were detected. In 2014 the case yield was 8.9, and in 2015 17.5 cases.
A fourth observation focuses attention on the importance of school based yaws detection programs. Twenty-seven percent of all yaws cases detected between 2012 and 2015 were identified during stand-alone school screenings in communities where no mass BU outreach event had been held. In 2012, school programs held in five moderate sized communities yielded 8 cases of yaws per activity. By 2015, stand-alone school screening events yielded 9.8 cases per screening. Social scientists found that large BU outreach events supported by community leaders in nearby communities helped legitimize school- based screening programs. The parents of children had a much better idea of why screenings in schools were being held and had confidence in the medication offered given the circulation of stories of successful yaws treatment.
In addition to detecting cases at schools, school-based programs taught students how to identify the signs of yaws. The social science team investigated whether identifying children with yaws in the school would be stigmatizing. Observations and interviews with children did not find this to be the case. Those conducting the education program made it clear that yaws was easily treated with just one injection and the demonstration effect of classmates recovering rapidly from symptoms made the program popular. By 2014, students were asked to examine each other for “tell-tale signs” of yaws and encouraged to identify possible cases of yaws in children either too young to attend school or who stopped going to school as a result of painful or unsightly lesions. In short, schoolchildren were enlisted to assist in community based identification of yaws among their peers and those children they helped care for at home.
A fifth observation entails the effectiveness of house-to-house surveys as a strategy for achieving yaws eradication in rural areas of Cameroon. As the result of an unfortunate break in funding in 2015, Stop Buruli mass outreach activities were suspended. To keep up the momentum of NTD activities, FAIRMED in conjunction with Cameroon’s government NTD program conducted school based screenings as well as an intensive house-to-house NTD survey. The two phase survey was carried out during late August and mid-November. In all, 1,889 houses in 38 communities were canvassed and 110 cases of yaws detected, 13% of total yaws cases identified between 2012 and 2015. Two points may be made. First, the number of yaws cases in the 2015 survey far exceeded the number of cases detected in 2010, when a much larger survey (N = 9,344) was carried out in the month of March. March is a busy month for agriculturalists and many people are working in fields far from their community. In 2015, yaws was detected in 6% of all households while in the 2010 survey cases were found in only .3% of households visited. A second point was that the program required a significant investment of health staff. In 2015, the participation of seven hospital staff members was required for 10 days of arduous surveillance activities. This constituted a significant opportunity cost for a busy district hospital like Bankim, which serves a population of about 100,000 inhabitants. Table 3 summarizes the relative advantages, limitations and logistical challenges of each type of yaws detection activity observed.
Data was collected on the distribution of yaws cases by age, gender, ethnic group, and household occupation as well as season. An analysis of cases of yaws by age conforms to a well-described epidemiological pattern (Fig 2). Eighty-four percent (84%) of yaws cases were under 15 years of age with 26% of children being under the age of 5 years. The large number of young children suffering from yaws suggests that school-based programs alone are insufficient to reach a significant percentage of high-risk children.
The gender distribution of children detected with yaws is presented in Table 4. We found a greater number of male cases in all age categories over the course of the four years of the study.
In Bankim, as in much of West Africa, sibling care is common. Sixty-six percent (66%) of all children symptomatic for yaws are of school-going age. They not only have frequent skin to skin contact with classmates in school, but younger siblings. Observations of sibling childcare by social scientists documented that while school-going females more commonly care for younger siblings, school aged males do so as well. Breaking the chain of transmission required teaching school children how to recognize the signs of yaws in their siblings.
The most common types of yaws lesions detected between 2012 and 2015 are presented in Table 5. Notably, the most common lesions are also the most contagious (ulcers—69% and papilloma—19%).
We examined the distribution of yaws cases by locale as well as ethnic and occupational group. Data on the six health areas that comprise Bankim district revealed that yaws cases were widely distributed with hot spots in both towns and villages. The maximum number of cases was detected in locales proximate to the Mbam River and Mape dam.
Analysis of data on yaws cases by ethnic groups likewise revealed broad distribution of cases across the six largest ethnic groups in the district as distinct from clustering in any particular group. One finds both single and mixed ethnic group settlements in Bankim. The largest ethnic groups were the groups with the most exposure to outreach programs and the most cases of yaws detected. Analysis by occupation found broad distribution across communities that rely on both agriculture and fishing.
We next looked at yaws cases detected by month. The majority of cases were identified during outreach screening activities when community members were more likely to be at home. Although carried out throughout the year, outreach activities were easier to conduct in some seasons due to climate, transportation, agricultural cycles, school registration, and ritual activities in the region. Peak months of yaws detection in the community were August–September and November–December. Fewer cases were detected from January to June.
Over the last decade, there have been repeated calls for integrated NTD programs [7, 8, 9] including NTDs that cause skin lesions [10, 11]. To date, most examples of integration have entailed the integration of preventive chemotherapy programs [12, 13] including those that have used schools as sites for community-based NTD control activities [14, 15, 16]. In Bankim district NTD integration evolved in response to a community based outreach program for BU. Few cases of yaws were diagnosed in the district until a community based outreach program for BU was initiated attracting large gatherings of community members. These gatherings proved to be magnets for community members with chronic lesions and wounds. While yaws cases was not an initial focus of wound screening, this changed when many cases of yaws were detected, much to the surprise of health staff. Offering free treatment for yaws proved to be a win-win scenario for both BU and yaws outreach. The effectiveness of yaws treatment increased confidence in medications offered to treat cases of BU.
The five interventions of yaws detection profiled in this study were found to have a synergistic effect and constitute valuable approaches to eradicating yaws. The most time and cost effective means of outreach in Bankim were well-planned mass education programs followed up by school-based programs. Piggy backing yaws eradication onto BU outreach programs made sense in Bankim because mass events are given far more importance by community members than short outreach activities conducted solely by health staff. These events are attended by chiefs and influential traditional healers who offer support, increase program credibility, and motivate CHWs to be more proactive. They also boost community trust in school based NTD activities that other studies have found do better when accompanied by broader based community outreach [17].
In Bankim, outreach activities carried out in schools not only served as opportunities for identifying cases of yaws among the segment of the population most likely to suffer from yaws, but educated children to detect yaws in the future. Children not only brought health messages home to parents, they were trained to detect yaws among children too young to attend school. Given that school-aged children of both genders commonly engage in sibling care, they are both a good detection resource and an important link in the chain of yaws transmission. Teachers were also found to be a good source of knowledge about children who appear to have dropped out of school due to skin diseases.
In this study, we ascertained the relative effectiveness of five methods of detecting yaws from a review of records of patients treated for the disease in different clinic and community contexts. One of the lessons learned is that single NTD disease focused outreach programs, like the mass BU events described in this paper, attracts community members with a wide range of chronic skin diseases like yaws. We initially identified yaws cases serendipitously. Over time we felt the need to be more proactive in identifying yaws cases, as we deemed it unethical not to do so. A limitation of the project was that we did not add yaws messages to the mass BU outreach events given that the novel BU education approach piloted was being evaluated. The addition of yaws messages might have constituted a confounding variable negatively impacting BU message evaluation. Furthermore, we had not conducted formative research on yaws necessary for the design of culturally appropriate messages. Basic yaws recognition was included in school-based programs. In the near future, we will integrate yaws messages into all community and school based NTD education programs once such messages are pretested.
WHO recommendations for eradicating yaws include mass treatment with oral azithromycin, the use of recently developed rapid diagnostic tests, and three to six month follow up in endemic communities. Cameroon is in the process of adopting these measures as soon as resources and manpower become available. In remote locations like Bankim, yaws eradication will prove challenging due to poor transportation, population movements, and ethnic diversity requiring that education to be delivered in multiple languages. The five kinds of interventions described in this paper will need to be coupled with mass treatment strategies in order to achieve the level of community outreach needed for eradication.
Yaws eradication may also require a more comprehensive approach to neglected tropical skin diseases and wound care. During Bankim outreach activities, health staff encountered many cases of chronic wounds that were either neglected or being treated inappropriately, leading to complications. Only focusing on BU and yaws and neglecting other wounds and lesions sends a message to the community that some wounds and skin diseases matter more than others, and that treating specific diseases matters more than providing care to people suffering from debilitating skin conditions. Better diagnostic testing will only add to this perception as more lesions, that to community members look similar to yaws or BU, are ruled out for free treatment [18, 19, 20].
Outreach programs in community and school settings combining NTD disease surveillance with wound care education, and when appropriate free treatment, would help address this problem. Such programs would serve a broader primary health care agenda [21] and minimize the kinds of rumors that have undermined other NTD programs in Africa [22, 23, 24] including BU [25]. In West Africa, local perceptions of illness and the agenda of those conducting public health programs matter. In order to be sustainable, NTD programs will need to build community trust. Much of the success of the Bankim program can be attributed to concerted efforts to respect and involve all community stakeholders in NTD activities.
|
10.1371/journal.pbio.3000098 | Triplet-pore structure of a highly divergent TOM complex of hydrogenosomes in Trichomonas vaginalis | Mitochondria originated from proteobacterial endosymbionts, and their transition to organelles was tightly linked to establishment of the protein import pathways. The initial import of most proteins is mediated by the translocase of the outer membrane (TOM). Although TOM is common to all forms of mitochondria, an unexpected diversity of subunits between eukaryotic lineages has been predicted. However, experimental knowledge is limited to a few organisms, and so far, it remains unsettled whether the triplet-pore or the twin-pore structure is the generic form of TOM complex. Here, we analysed the TOM complex in hydrogenosomes, a metabolically specialised anaerobic form of mitochondria found in the excavate Trichomonas vaginalis. We demonstrate that the highly divergent β-barrel T. vaginalis TOM (TvTom)40-2 forms a translocation channel to conduct hydrogenosomal protein import. TvTom40-2 is present in high molecular weight complexes, and their analysis revealed the presence of four tail-anchored (TA) proteins. Two of them, Tom36 and Tom46, with heat shock protein (Hsp)20 and tetratricopeptide repeat (TPR) domains, can bind hydrogenosomal preproteins and most likely function as receptors. A third subunit, Tom22-like protein, has a short cis domain and a conserved Tom22 transmembrane segment but lacks a trans domain. The fourth protein, hydrogenosomal outer membrane protein 19 (Homp19) has no known homology. Furthermore, our data indicate that TvTOM is associated with sorting and assembly machinery (Sam)50 that is involved in β-barrel assembly. Visualisation of TvTOM by electron microscopy revealed that it forms three pores and has an unconventional skull-like shape. Although TvTOM seems to lack Tom7, our phylogenetic profiling predicted Tom7 in free-living excavates. Collectively, our results suggest that the triplet-pore TOM complex, composed of three conserved subunits, was present in the last common eukaryotic ancestor (LECA), while receptors responsible for substrate binding evolved independently in different eukaryotic lineages.
| Mitochondria carry out many vital functions in the eukaryotic cells, from energy metabolism to programmed cell death. These organelles descended from bacterial endosymbionts, and during their evolution, the cell established a mechanism to transport nuclear-encoded proteins into mitochondria. Embedded in the mitochondrial outer membrane is a molecular machine, known as the translocase of the outer membrane (TOM) complex, that plays a key role in protein import and biogenesis of the organelle. Here, we provide evidence that the TOM complex of hydrogenosomes, a metabolically specialised anaerobic form of mitochondria in Trichomonas vaginalis, is composed of highly divergent core subunits and lineage-specific peripheral subunits. Despite the evolutionary distance, the T. vaginalis TOM (TvTOM) complex has a conserved triplet-pore structure but with a unique skull-like shape suggesting that the TOM in the early mitochondrion could have formed three pores. Our results contribute to a better understanding of the evolution and adaptation of protein import machinery in anaerobic forms of mitochondria.
| Mitochondria originated from proteobacterial endosymbionts [1], and over time, massive endosymbiotic gene transfer to the host nucleus or gene deletion forged the development of a mechanism for retargeting of nuclear-encoded proteins to the evolving organelle [2]. To cross the double membrane of the mitochondrion, the proteins had to pass through the translocase of the outer (TOM) and inner (TIM) membranes. It has been inferred that most modules of the import machinery were created de novo and the ancient TOM complex comprised at least three components, the β-barrel translocation channel-forming Tom40 and two tail-anchored (TA) proteins, Tom22 and Tom7 [3,4].
The TOM complex in yeast consists of Tom40 and six α-helical proteins: two that are anchored to the outer mitochondrial membrane (OMM) by an N-terminal transmembrane domain (TMD; Tom20 and Tom70) and four that are anchored by a C-terminal TMD (Tom22, Tom5, Tom6, and Tom7). Tom20 and Tom70, both carrying tetratricopeptide repeat (TPR) domains, serve as primary receptors recognising proteins with N-terminal targeting sequence (NTS) and internal-targeting sequences (ITSs), respectively [5,6]. A prominent feature of the TOM complex is the variation in receptors across different eukaryotic lineages. A signal-anchored Tom20 is present in animals and fungi, whereas plant Tom20 evolved independently with a C-terminal anchor [7]. Lineage-specific Tom20 and Tom60 without any TMD are present in amoebozoans [8,9]. Tom20 and Tom70 are essentially absent in the eukaryotic supergroup Excavata [10–12]. In the excavate Trypanosoma brucei, the TOM complex (named the archaic translocase of the outer membrane [ATOM]) has only two orthologues, a highly divergent Tom40 (ATOM40) and a Tom22-like protein (ATOM14) [11,13]. Instead of Tom70 and Tom20, two unique receptors were identified, a TA protein ATOM69 and a signal-anchored ATOM46 [11].
Structural studies of the contemporary TOM complex are exclusively based on fungi, Saccharomyces cerevisiae and Neurospora crassa [14,15]. The yeast TOM complex is highly dynamic, with the mature trimeric complex formed by three pores, alternately switching with a dimeric form containing two pores, which serves as a platform for the integration of a new Tom40 into the complex [16]. The assembly of the Tom40 precursor in the OMM is mediated by the sorting and assembly machinery (SAM) that consists of a central β-barrel subunit Sam50 and two peripheral subunits Sam35 and Sam37 in yeast. To promote β-barrel biogenesis, TOM and SAM form a transient supercomplex [17,18]. The dimeric and trimeric TOM structures are stabilised by the highly conserved TMD of Tom22 [19]. This specific function of Tom22 and its conservation in most eukaryotes led to speculation that the ancient TOM complex may have been a trimeric form [12]. However, this concept remains unsettled as it has not been clarified whether N. crassa TOM complex forms a three-pore or a two-pore structure [15,20], and so far, the information on TOM structure from other organisms is unavailable. Thus, to understand what subunits contributed to the formation of the earliest translocases and to reconstruct the evolutionary steps, it is important to study the composition and the structure of the translocases in organisms harbouring different variants of mitochondria as well as in organisms from different eukaryotic supergroups. Highly reduced mitochondria known as hydrogenosomes and mitosomes are found in certain organisms adapted to an anaerobic lifestyle [21] with simplified import machinery. The most studied hydrogenosomes are those found in the Parabasalia group of excavates, which includes the human parasite Trichomonas vaginalis. T. vaginalis hydrogenosomes have lost the tricarboxylic acid (TCA) cycle, and the oxidative phosphorylation has been replaced by substrate-level ATP synthesis, with the concomitant production of hydrogen [22]. Hydrogenosomes have lost the organellar genome entirely [23], and consequently, all hydrogenosomal proteins are imported from the cytosol. Like mitochondria, the import of proteins into hydrogenosomes is dependent on the hydrogenosomal NTS [24]. However, some matrix proteins are imported into hydrogenosomes independent of an NTS, and therefore the NTS-independent route was proposed to represent an ancestral mode of protein import [25,26]. Previous proteomic analysis of T. vaginalis hydrogenosomes revealed the presence of several β-barrel proteins of the mitochondrial porin 3 superfamily that were designated as putative Tom40. However, the protein sequences were highly divergent from known homologues, making it difficult to unequivocally distinguish between Tom40 and voltage-dependent anion channel (VDAC) [10]. Other hydrogenosomal β-barrel proteins include Sam50 and paralogues of two proteins of unknown function, hydrogenosomal membrane protein 35 (Hmp35) and Hmp36 [10,27]. Neither genomic nor proteomic analyses indicated the presence of other TOM components [10,28]. Hydrogenosomes also lack Tim50 and its regulatory subunit Tim21 that links the TOM complex with TIM in the intermembrane space (IMS) [10,28,29]. Furthermore, five paralogues of the Tim17/22/23 family that constitute the TIM channel have been detected. However, limited similarity of these hydrogenosomal proteins to Tim17, Tim22, and Tim23 subfamilies prevented determining whether they form a single multifunctional channel or distinct TIM23 and TIM22 channels for the import of matrix and inner membrane proteins, respectively [10]. Thus, structure and function of the hydrogenosomal protein import machineries remains elusive.
In the present study, we focus on the T. vaginalis TOM complex (TvTOM) and demonstrate that this highly divergent translocase mediates protein import into hydrogenosomes. Despite remarkable divergence in both primary structure and evolutionary distance, electron microscopy revealed some structural similarity between TvTOM and yeast three-pore TOM complex. However, the presence of an extra density provides a unique skull-like shape to TvTOM. Mass spectrometry (MS) of TvTOM and bioinformatic analysis identified two conserved and three lineage-specific TOM subunits, including two receptors, and revealed an association of TvTOM with Sam50. Although we did not identify Tom7 in TvTOM, our phylogenetic profiling predicted Tom7 in free-living representatives of Excavata. We propose that Tom40, Tom22, and probably Tom7 were present in the last common eukaryotic ancestor (LECA) and constituted a triplet-pore TOM complex, whereas the receptor subunits evolved independently in different eukaryotic lineages.
Seven Tom40-like proteins, named TvTom40-1 to TvTom40-7, identified in the hydrogenosomal proteome [10] displayed remarkably low sequence identity with fungal Tom40 sequences (e.g., 10%–14% identity compared with N. crassa). All TvTom40 proteins carry a conserved β-motif, PxGxxHxH (P = polar; G = glycine; H = hydrophobic; x = any amino acid) in the last β-strand similar to Tom40s and VDACs of other eukaryotes except TvTom40-3, where the last hydrophobic amino acid has been replaced by a polar hydroxylic residue, serine (S1 Fig). Bioinformatic analyses for all the seven proteins using HHpred tool identified TvTom40-2 (TVAG_332970) as the closest homologue to Tom40 (S1 and S2 Tables). Next, we built a local Tom40 hidden Markov model (HMM), based on 24 well-annotated Tom40 sequences (S1 Data) that was employed to scan the T. vaginalis proteome with HMMER jackhmmer tool, and again, TvTom40-2 was identified as the best Tom40 candidate.
A homology model of TvTom40-2 was constructed based on the N. crassa Tom40 template. TvTom40-2 forms a typical 19-strand β-barrel structure, but with only one N-terminal helix instead of two helices observed in Tom40 of other eukaryotes. Furthermore, TvTom40-2 contains a unique loop between β-strands five and six that is positively charged (Fig 1A). Most of the positions responsible for the interactions with other TOM proteins in yeast [16] are not conserved in TvTom40-2 (S2 Fig). A comparison of the electrostatic potential revealed that TvTom40-2 and N. crassa Tom40 share both positively and negatively charged patches inside the barrel, whereas mouse VDAC is almost uniformly positively charged (Fig 1B). Hence, based on homology searches and modeling, TvTom40-2 was chosen for further experimental studies.
To verify the cellular localisation of TvTom40-2, a strain expressing C-terminally human influenza hemagglutinin (HA)-tagged TvTom40-2 was prepared. Immunofluorescence microscopy visualised TvTom40-2 as a ring, staining the membrane of hydrogenosomes. Malic enzyme, a hydrogenosomal marker enzyme, stained the organellar matrix (Fig 2A). Cell fractionation and immunoblotting revealed the presence of TvTom40-2 exclusively in the hydrogenosomal fraction (Fig 2B). Treatment of hydrogenosomes carrying HA-tagged TvTom40-2 with proteinase K resulted in a shift of the molecular weight from 37 kDa to 28 kDa, indicating that the protein was likely cleaved within the loop between the fourth and fifth β-strands that is oriented towards the cytosol (Figs 2B and 1A). Then, the isolated hydrogenosomes were solubilised with varying concentrations of digitonin (1%–3%), and the samples were subjected to blue native-PAGE (BN-PAGE). TvTom40-2 was observed to be present in two high molecular weight complexes of 570 kDa and 330 kDa (Fig 2C). These experiments demonstrate that TvTom40-2 is present in a high molecular weight complex embedded in the hydrogenosomal outer membrane.
The striking divergence of hydrogenosomal TvTom40-2 from Tom40 orthologues prompted us to test whether biogenesis of TvTom40-2 is specific to the hydrogenosomal machinery or whether, despite the variance in the sequence, it could be integrated into the OMM of distant eukaryotes from Opisthokonta lineage. We expressed TvTom40-2 with a C-terminal HA tag in S. cerevisiae. TvTom40-2 appeared in the mitochondrial fraction together with the mitochondrial marker, aconitase (Fig 3A). Alkaline extraction showed that most of the TvTom40-2 was present, similar to the OMM protein, mitochondrial fission 1 (Fis1), in the membrane fraction (Fig 3B). Finally, treatment of isolated mitochondria with proteinase K resulted in the formation of a proteolytic fragment of TvTom40-2 that resembled the one observed with isolated hydrogenosomes (Fig 3C). As expected, this fragment was completely degraded upon solubilisation of the organelles with the detergent. Collectively, these observations indicate that TvTom40-2 is localised in the OMM in yeast. In addition, to check whether TvTom40-2 could form an oligomeric complex in yeast mitochondria, we performed BN-PAGE. TvTom40-2 migrated in a 230 kDa complex, while ScTom40 migrated in a 480 kDa complex (Fig 3D). This suggests that TvTom40-2 can form in yeast mitochondria a high molecular weight complex, although of smaller size than that in hydrogenosomes.
Because TvTom40-2 was integrated into the OMM of yeast, we wanted to test whether it could functionally replace ScTom40. First, we prepared a yeast mutant, tet-TOM40, such that the TOM40 promoter was replaced by a tetracycline promoter via homologous recombination, which would deplete ScTom40 in the presence of doxycycline (Dox). As expected, the addition of Dox to the growth medium resulted in a growth retardation of the tet-TOM40 mutant. When TvTom40-2 was overexpressed, it could not rescue the growth defect of the tet-TOM40 strain on fermentable medium (synthetic drop-out medium without leucine, SD-Leu) but could do so on nonfermentable medium (yeast extract-peptone-glycerol [YPG]) (Fig 3E). To substantiate this observation, we performed functional complementation studies using two yeast strains harbouring temperature-sensitive alleles of TOM40—tom40-25 and tom40-34. When grown at 30 °C, the overexpression of TvTom40-2 partially restored the growth phenotype of the tom40-25 strain both on fermentable and nonfermentable media (Fig 3F). Such an effect was not observed in the same strain grown at elevated temperature (37 °C, Fig 3F). The growth of tom40-34 was not restored even at lower temperatures (Fig 3F). Thus, it seems that TvTom40-2 can only partially replace yeast Tom40 function.
To identify interaction partners for TvTom40-2, we performed co-immunoprecipitations (coIPs) of HA-tagged TvTom40-2 under crosslinking and native conditions, and the eluted proteins were analysed using label-free quantitative MS (LFQ-MS). CoIPs using anti-HA antibody were performed with hydrogenosomes isolated from both the strain expressing HA-tagged TvTom40-2 and the wild-type (WT) strain, used as a negative control. The analysis revealed that 50 and 36 proteins were enriched with HA-tagged TvTom40-2 under crosslinking and native conditions, respectively (S2 Data). As TOM proteins are embedded in the hydrogenosomal outer membrane, we searched for proteins with TMDs in the data sets using TMHMM and found 19 and 13 proteins for crosslinking and native coIPs, respectively. The intersection between the two data sets and the hydrogenosomal membrane proteome [10] contained five TvTom40 isoforms, two TA proteins named Tom36 and hydrogenosomal outer membrane protein 19 (Homp19), two Sam50 paralogues, and Hmp35 (S2 Data and Fig 4A). In addition, the intersection between the coIP data set under crosslinking conditions and the membrane proteome contained two more TA proteins named Tom46 and Homp38. Based on our previous results [10], we selected Tom36 for the reciprocal coIPs.
Proteins enriched in the HA-tagged Tom36 coIPs under crosslinking conditions included three isoforms of TvTom40, Sam50, Hmp35, Homp38, Tom46, and Homp19, whereas under native conditions, three isoforms of TvTom40, Sam50, and Hmp35 were enriched (S2 Data and Fig 4B). Altogether, the coIP and MS data indicated four TA candidate proteins, Homp19, Tom36, Homp38, and Tom46. InterProScan [30] predicted that Tom36, Homp38, and Tom46 would carry an N-terminal heat shock protein (Hsp)20-like chaperone domain, three TPR-like domains, and a C-terminal TMD. This domain architecture resembles the recently reported ATOM69 in T. brucei [11] (Fig 4C). Indeed, HHpred searches using Tom36 and Homp38 as queries against the T. brucei proteome revealed ATOM69 as the first hit, with e-values of 4.9 × 10−17 and 2.3 × 10−11, respectively. HHpred searches with Tom46 recognised various proteins with TPR domains, whereas no significant homology was observed for Homp19.
The coIP-MS data did not identify homologues of either Tom22 or Tom7. Thus, we used HMM to search for Tom22 and Tom7 sequences in the T. vaginalis protein database. The searches for Tom22 identified a small protein with a predicted molecular weight of 6.4 kDa, containing a C-terminal TMD. It has a conserved Tom22 motif, including a tryptophan residue at the second position, followed by a few hydroxylated residues, with a serine at the +4 position and an invariant proline residue in the TMD; hence, we named it Tom22-like protein (TVAG_076160) (S3 Fig). In comparison to the fungal Tom22, Tom22-like protein is substantially shorter, similar to Tom22-like proteins in plants, apicomplexans, and kinetoplastids [4,31,32]. However, unlike Tom22, Tom22-like protein lacks a C-terminal IMS domain (S3 Fig). Searches for Tom7 in the T. vaginalis protein database did not identify a convincing orthologue.
Interestingly, Sam50 that only transiently associates with TOM in yeast [17] was copurified when both TvTom40-2 and Tom36 were pulled down both under crosslinking and native conditions, which may suggest a more stable association between TvTOM and Sam50. Therefore, we performed reciprocal coIPs using a strain expressing HA-tagged Sam50. LFQ-MS analysis revealed a similar spectrum of proteins as observed in the previous experiments that supports TvTOM-Sam50 association (S2 Data and Fig 4D). The presence of HA-tagged proteins in the eluates from TvTom40-2-HA, Tom36-HA, and Sam50-HA crosslinking and native coIPs were verified via immunoblotting (Fig 4E).
To verify the localisation and topology of identified TA proteins, we prepared double transfectants that expressed TvTom40-2-HA together with one of the candidate proteins, all of which were C-terminally tagged with V5. In all cases, the TA protein was present in the hydrogenosomal fraction (Fig 5A). Treatment of isolated hydrogenosomes with proteinase K showed the presence of a truncated fragment that was protected from externally added proteinase K (Fig 5A). Next, we visualised V5-tagged candidate proteins, together with HA-tagged TvTom40-2, in the double transfectants using Stimulated Emission Depletion (STED) microscopy. All five candidates exhibited a ring-like pattern in the hydrogenosomal outer membrane similar to that observed with TvTom40-2 (Fig 5B). A Pearson correlation coefficient displayed the highest degrees of colocalisation with TvTom40-2 for Tom46 (77%) and Tom22-like protein (63%). Decreasing degrees of colocalisation with TvTom40-2 were observed for Tom36 (46%), Homp19 (26%), and Homp38 (17%). These experiments showed that all the selected TA proteins reside in the hydrogenosomal outer membrane.
To obtain further support for the association of identified TA proteins and Sam50 with the TvTOM complex, TvTom40-2-HA was pulled down from hydrogenosomes isolated from the double transfectants, and the samples were probed for V5-tagged proteins and Sam50 via immunoblotting using α-V5 and polyclonal α-Sam50 antibodies, respectively. Under crosslinking conditions, TvTom40-2 pulled down Tom36, Tom46, Homp19, and Tom22-like protein, while under native conditions, we observed a strong signal for Tom36, Homp19, and Tom22-like protein and a weaker signal for Tom46 (Fig 6A). Homp38 was not co-immunoprecipitated from the double transfectant under these conditions. On the other hand, Sam50 was detected in all samples analysed (Fig 6A). Furthermore, to validate whether the TvTom40-2-associated proteins are present in the high–molecular-weight complexes, hydrogenosomes isolated from the recombinant strains were subjected to BN-PAGE and immunoblotted with corresponding antibodies. Both Tom36 and Tom22-like protein migrated in 570 kDa and 330 kDa complexes. Tom46 and Homp19 migrated only in a 330 kDa complex, while Homp38 did not appear to be present in any high molecular weight complex. TvTom40-2, used as a reference, migrated at 570 kDa and 330 kDa under the same conditions when immunodecorated with α-HA antibody (Fig 6B). HA-tagged Sam50 migrated at 570 kDa and 55 kDa, which corresponded to the high molecular weight of TvTOM complex and to Sam50 monomer, respectively (Fig 6B). These results confirmed the association of Tom36, Tom46, Homp19, Tom22-like, and Sam50 with TvTom40-2, and their ability to incorporate into high molecular complexes.
To demonstrate that the predicted TvTom40-2 participates in hydrogenosomal protein import, we performed an in vitro protein import and coIP assay. As an import substrate, we used the hydrogenosomal matrix protein ferredoxin (TvFdx1), which has an NTS fused to Escherichia coli dihydrofolate reductase (DHFR) at the C-terminus. TvFdx1-DHFR was synthesised in vitro in the presence of [35S]-methionine. Under standard in vitro import conditions, using hydrogenosomes isolated from the double-transfected TvTom40-2-HA/Tom36-V5 strain, TvFdx1-DHFR was imported into hydrogenosomes, which was confirmed by a protease protection assay. The autoradiograph showed a time-dependent import of TvFdx1-DHFR (Fig 7A). Next, in vitro import assay was performed in the presence of methotrexate, which is known to cause the folding of DHFR and therefore arrests the translocating protein at the mitochondrial protein import site [33]. As expected, TvFdx1-DHFR was arrested at the hydrogenosomal outer membrane, and the exposed region was degraded when the hydrogenosomes were treated with proteinase K (Fig 7B). Finally, to prove that TvTom40-2, Tom36, and the substrate are present in the same complex, we performed in vitro import assay for TvFdx1-DHFR either in the presence or absence of methotrexate, crosslinked the interacting proteins, and immunoprecipitated the complex via TvTom40-2-HA. Autoradiography of the eluted sample revealed the presence of arrested TvFdx1-DHFR associated with the complex when methotrexate was added (Fig 7C). The two bands present on the autoradiograph (lane 1) correspond to TvFdx1-DHFR (30 kDa) and its proteolytically cleaved product (29 kDa) most likely. Immunoblot analysis of the complex confirmed the presence of TvTom40-2 and Tom36 in the same sample (Fig 7D). No substrate signal was observed when methotrexate was omitted from the reaction mixture (Fig 7C). These results demonstrate that TvFdx1-DHFR was imported into hydrogenosomes in an unfolded state and the arrested TvFdx1-DHFR was associated with TvTom40-2 and Tom36.
Because both Tom36 and Tom46 interact with TvTom40-2, are present in high–molecular-weight complexes, carry TPR-like domains and Hsp20-like chaperone domain that are involved in protein–protein interactions, and are paralogues, we selected these proteins as receptor candidates. To test whether they can bind to hydrogenosomal proteins, we performed in vitro binding assay. The cytosolic domain of Tom36 (Tom36cd, residues 1–308) and Tom46 (Tom46cd, residues 1–402) were expressed with a C-terminal polyhistidine (His) tag in E. coli BL21 (DE3) strain, respectively, and coupled with Ni-nitrilotriacetic acid (Ni-NTA) agarose beads (S4 Fig and Fig 8A and 8B). Beads preincubated with untransformed E. coli lysate were used as a negative control. A cytosolic protein cytochrome b5 was used as a negative control. Radiolabelled precursors of two hydrogenosomal matrix proteins, frataxin and the α-subunit of succinyl coenzyme A (CoA) synthetase (αSCS), with the latter fused to DHFR at the C-terminus (αSCS-DHFR), were incubated with Tom36cd-His or Tom46cd-His coupled with or mock-treated beads for 1 hour. Then, the His-tagged proteins with the bound substrates were eluted with imidazole. The eluate from the Tom36cd-His and Tom46cd-His binding assay showed the presence of two radiolabeled proteins, frataxin and αSCS-DHFR (Fig 8C, top panel). The cytosolic cytochrome b5 was not observed to be bound to either Tom36cd-His or Tom46cd-His (Fig 8C, top panel). Furthermore, the eluates were immunoblotted with anti-His antibody to verify the presence of His-tagged proteins (Fig 8C, bottom panel). These experiments indicate that the cytosolic domain of Tom36 and Tom46 can bind hydrogenosomal preprotein substrates.
The diversity of TvTom40 paralogues and the presence of unusual components in the TvTOM complex prompted us to investigate the structure of the TvTOM complex via electron microscopy analysis. The hydrogenosomal TOM complex was purified from T. vaginalis expressing TvTom40-2-HA under native conditions. The isolated hydrogenosomes were solubilised with digitonin to release the complex, and then the TvTOM complex was purified by IP using α-HA antibody coupled to Dynabeads and negatively stained for electron microscopy. The identity of the HA-tagged TvTom40-2 in the IP eluate was verified by immunoblotting and silver staining (S5A and S5B Fig). The unprocessed electron micrographs mainly showed particles composed of ring-shaped structures with one, two, or three centers of stain accumulation (representative micrograph in S5C Fig). These stain-filled openings are interpreted as pores, each of which represents one channel of the protein translocase. A total of 10,038 particles were selected from 650 micrographs for further processing. Two-dimensional (2D) classification with 3,412 particles (34% of 10,038 particles) resulted in class averages representing TvTOM with one, two, or three pores of resolution between 21 and 34 Å (Fig 9A–9C). TvTOM with one or two pores were the most prominent, accounting for 35% (n = 1,175) and 40% (n = 1,377), respectively, while TvTOM with three pores accounted for 25% (n = 860). The single-pore particles were oval, 70 × 125 Å in size with an eccentric pore placement. Two-pore particles were oval or triangular and 140 × 100 Å in size. The particles with three pores were skull-shaped and measured 150 × 175 Å in size, although a fourth spot of stain accumulation with a low contrast was observed in one of the class averages (Fig 9C). A single translocation channel measured 70 Å in diameter, and the inner pore size of the channel measured 25–30 Å. The distance between two pore centers measured 50–60 Å. The most striking difference from the yeast TOM is the presence of an extra density, measuring 50 Å in diameter observed in most classes of single-, double-, and triple-pore TvTOM particles, suggestive of a subunit(s) interacting with the peripheral part of the channel formed by TvTom40.
Conservation of Tom40 and Tom22, and the identification of two novel peripheral components with Hsp20 and TPR domains (Tom36 and Tom46) suggest a peculiar evolutionary history for TvTOM complex. Therefore, we searched for orthologues of TOM components using a local HMM in selected genomes across different eukaryotic supergroups, with a focus on Excavata to estimate the conservation, gain, and loss of components (S3 Table and S3 Data). For our evolutionary scheme (Fig 10), we adapted a view that Excavata has two major sister groups: Metamonada, comprising anaerobic protists such as T. vaginalis, and Discoba, comprising T. brucei [34,35], although an alternative placement of Metamonada has been suggested [36]. Our phylogenomic profiling supported the current view that at least Tom40 and Tom22 are conserved in all eukaryotes and might have been present in the TOM complex of LECA (Fig 10). The only exception is Monocercomonoides sp., which has completely lost mitochondria including all genes coding for TOM and TIM components [37] (S3 Table and Fig 10). Support for Tom7 was less clear because neither T. vaginalis nor T. brucei seems to possess Tom7 (S3 Table and Fig 10). However, we took advantage of the available genome sequences of some free-living excavates [38–40] and identified putative Tom7 orthologues in Carpediemonas membranifera of Metamonada, and Euglena gracilis and Stygiella incarcerata of Discoba lineages (S3 Table and Fig 10). As expected, our searches showed that Tom20 and plant Tom20 were most likely gained independently in Opisthokonta and Viridiplantae, respectively, and their orthologues are not present in other lineages, including Excavata (S3 Table and Fig 10). The evolutionary history of Tom70, Tom5, and Tom6 is more complex. All three components have been found in opisthokonts, while only Tom5 and Tom6 are present in Viridiplantae. Conversely, in the supergroup Stramenopiles, Alveolata and Rhizaria (SAR), which is related to Viridiplantae [34], Tom5 and Tom6 are absent, whereas Tom70 was reported in Blastocystis, other SAR species, and the haptophyte Emiliania huxleyi [41] (S3 Table and Fig 10). In our searches, none of these three components have been identified in both Excavata and Amoebozoa (S3 Table and Fig 10). The most puzzling aspect is the appearance of unique peripheral TOM components in the Excavata group. The searches for proteins with the same domain structure as Tom36 (Hsp20-TPR-TMD) in the available genome of 11 excavates and in the genome of selected organisms from other eukaryotic supergroups revealed the presence of homologous proteins only in Tritrichomonas foetus, a close relative of T. vaginalis (Parabasalia lineage), in kinetoplastids, and interestingly, in a fungus Neocallimastix californiae (S3 Table and Fig 10).
Next, we performed homology searches using Tom36 or ATOM69 as queries against the National Center for Biotechnology Information (NCBI) nonredundant protein database regardless of the domain composition that resulted in a data set of 299 eukaryotic, 810 bacterial, and 5 archaeal sequences that were analysed using CLuster ANalysis of Sequences (CLANS) algorithm [43] (Fig 11A and S4 Data). Tom36 and Tom46 formed a cluster together with 10 other T. vaginalis and four T. foetus homologues (Fig 11A). All these homologues share Hsp20-TPR domains, two of them without any predicted TMD. A distinct cluster included seven ATOM69 homologues found in kinetoplastids that included dixenic, monoxenic, and free-living species (Fig 11A). The other clusters were formed by various TPR proteins, including elongation factor 2 kinase and endoplasmic reticulum-associated protein degradation (ERAD)-associated E3 ubiquitin-protein ligase (Fig 11A). The largest cluster predominantly contained bacterial proteins (Fig 11A). The formation of distinct clusters for Hsp20-TPR-TMD proteins of trichomonads and kinetoplastids suggests that Tom36/Tom46 and ATOM69 may have evolved independently in their respective lineages (Fig 11A). This view is supported by our phylogenetic analysis, in which Tom36/Tom46 and ATOM69 form two separate branches that are interleaved by a large bacterial group (Fig 11B).
In spite of the fundamental role of mitochondrial translocases for the function and evolution of the eukaryotic cell, our experimental knowledge of the TOM complex is limited to a few model organisms, and direct visualisation of the TOM complex has only been achieved in two fungi, S. cerevisiae and N. crassa [14,15]. To extend our knowledge on TOM diversity in eukaryotes, we isolated and characterised the TOM complex from hydrogenosomes, an anaerobic form of mitochondria in T. vaginalis. In the present study, we have demonstrated the function of a highly divergent pore-forming TvTom40-2 and identified a protein that has limited homology with Tom22. The other components of TvTOM include three TA proteins with no orthologues in the fungal TOM complex. Furthermore, TvTOM seems to be tightly associated with Sam50 for a more efficient β-barrel biogenesis.
Electron microscopic visualisation of the TvTOM complex revealed interesting similarities and differences when compared with the TOM complex in fungi. Most observed TvTOM particles displayed two pores, which in fungi represent the TOM core complex, or particles with three pores, corresponding to the holo complex. The distance between two pore centers, the inner pore diameter, the single translocation channel diameter, and the size of the particles with two pores are similar to those determined for the TOM complex in fungi [15,20]. The appearance of single-pore particles could more likely be either a result of the dissociation of holo complexes during experimental procedures [20,44] or stable assembly intermediates. A striking deviation from known TOM models is the presence of an extra density in the single-, double-, and triple-pore particles, providing a skull-like shape to the TvTOM holo complex. Based on coIP-MS analysis, it can be speculated that the extra mass may contain the identified β-barrel proteins Sam50 or Hmp35. In yeast, the TOM and SAM complexes form a labile supercomplex that allows coupling of the translocation of the Tom40 precursor through TOM and its insertion into the OMM via SAM [17]. It has been suggested that Sam50 may account for the third pore in the yeast triplet-pore complex [15]. Cryo electron microscopy (Cryo-EM) has shown that the Sam50 monomer measures 50 Å [45], which is consistent with the size of the additional mass observed in TvTOM. BN-PAGE analysis showed that HA-tagged Sam50 migrated with the high–molecular-weight complex of TvTOM or as a monomer. The enrichment of TOM subunits, as well as Sam50 in the reciprocal coIPs, supports a tight TOM-Sam50 association in hydrogenosomes. Formation of the supercomplex in yeast is mediated by the N-terminal cytosolic domain of Tom22 and Sam37 [17,18]. In trichomonads, Sam37 has not been identified [28], and Tom22 has a short cytosolic domain. Therefore, if the observed association of TvTOM and Sam50 represents a functional complex, different protein–protein interactions are to be expected. Hmp35 is a β-barrel protein in the hydrogenosomal membrane with an unknown function that exists in a stable 300 kDa complex of Hmp35 oligomers [27]. This complex is too large to imply the formation of a complex with TvTOM.
The presence of a TOM complex with three pores observed in T. vaginalis strongly indicates that triplet-pore complex is the generic form of TOM in eukaryotes that was inherited from LECA. It has been proposed that the ancient TOM complex contained—in addition to Tom40—Tom22, which tethers Tom40s using its TMD, and a regulatory subunit Tom7 [4,12,16,46]. The Excavata group includes two major lineages, Metamonada and Discoba, represented by T. vaginalis and T. brucei, respectively. Investigations of T. brucei TOM complex initially suggested that Tom40 in kinetoplastids (ATOM40) might be a homologue of the bacterial Omp85-like protein [13]. However, profile-sequence searches found that ATOM40 belongs to the eukaryotic porin family [12,47]. Our analysis, with an extended sampling of Excavata—which included a Tom40 orthologue in E. gracilis, which shares a common ancestry with kinetoplastids—confirmed this view.
Previous sequence searches implied the absence of Tom22 in some excavates with reduced forms of mitochondria, including the hydrogenosomes of T. vaginalis [12]. However, due to its short sequence and low conservation [4,12,32], the identification of Tom22 might have been beyond the sensitivity of most search tools. Our sensitive, structure-based HMM search identified a short 6 kDa Tom22-like protein as a potential candidate. This protein is tightly associated with TvTom40-2 in the hydrogenosomal outer membrane and is present in both high molecular weight complexes (570 and 330 kDa). Tom22-like protein contains a conserved TMD motif, including invariable tryptophan and proline residues, and a short cytosolic N-terminal (cis) domain similar to the 9 kDa Tom22 orthologue, Tom9 in higher plants, the 8 kDa apicomplexan Tom22, and the kinetoplastid Tom22 orthologue, ATOM14 [4,31,32]. The long acidic extension of the cis domain evolved only in opisthokonts that interacts with lineage-specific Tom20 and Tom70 [4], and therfore its absence in Tom22-like protein is not surprising. Most Tom22s contain an IMS-localised acidic (trans) domain that interacts with the substrate and enhances its transfer to Tim50 in the TIM23 complex [19]. Tom22-like protein identified here lacks the trans domain, which may reflect the absence of Tim50 in T. vaginalis [28]. In addition to T. vaginalis and T. brucei, we retrieved Tom22 orthologues from members of both Metamonada and Discoba in support of its presence in Excavata common ancestor.
Tom7 has not been identified in parabasalids, diplomonads, and in kinetoplastids. A fusion protein with limited sequence similarity to Tom7 and Tom22 has been reported in Naegleria species [12]. Importantly, Tom7 orthologues appears to be present in free-living members of both Excavata lineages, in C. membranifera (Metamonada), and E. gracilis and S. incarcerata (Discoba). This suggests that the absence of Tom7 might be a result of a secondary loss, and if so, it happened independently in certain lineages of both Metamonada and Discoba. However, failure to identify small Toms—Tom7 as well as Tom5 and Tom6—needs to be tread with caution. Their sequences are very short and might be highly divergent, particularly in parasitic lineages and those with reduced forms of mitochondria, which can hamper their identification. Collectively, our results suggest that the triplet-pore form of the TOM complex constituted the ancestral form of TOM in LECA.
Functional studies of TvTom40-2 using a DHFR-methotrexate system demonstrated that hydrogenosomal preprotein binds to TvTom40-2 and subsequently is imported into the hydrogenosomal matrix in an unfolded or loosely folded state, a feature that is conserved in mitochondria [33]. Of note, T. vaginalis has at least seven TvTom40 paralogues that are all expressed [10]. CoIP-MS analysis revealed that TvTom40-2 is associated with five other paralogues, and therefore various combinations of TvTom40 paralogues appear to be present in a single TOM complex, as observed in the rat TOM complex, in which two Tom40 isoforms interact with each other [48]. Further, we asked whether the hydrogenosomal TvTom40-2 could be integrated and can function in the yeast OMM. Despite low amino acid sequence conservation between TvTom40-2 and yeast orthologue, heterologous expression of TvTom40-2 in yeast resulted in its localisation in the OMM and the formation of a 230 kDa complex. This finding is consistent with the recent investigation of the targeting signal in β-barrel proteins, wherein the signal appears not to be encoded in a conserved linear amino acid sequence but is embedded in the structure of a β-hairpin motif [49]. Such a targeting signal was likely inherited from bacterial β-barrel proteins and remains conserved across all eukaryotic lineages, as supported by our experiment. As observed via protease protection assay, the topology of TvTom40-2 both in hydrogenosomes and mitochondria was similar. Interestingly, TvTom40-2 was able to very partially substitute yeast Tom40, indicating that at least some proteins were imported into yeast mitochondria through TvTom40-2. It is of note that some yeast mitochondrial proteins were imported into hydrogenosomes of T. vaginalis regardless of the presence or absence of NTS [26]. Based on this, it was proposed that the hydrogenosomal Tom40 is able to recognise unspecified ITSs conserved in the proteins of mitochondrial ancestry [26].
The key question is whether the TvTOM complex in hydrogenosomes consists of only core subunits or whether there any peripheral TOM subunit(s) that contribute to the import of proteins. This is expected because both NTS- and ITS-dependent protein targeting to hydrogenosomes have been demonstrated [24–26]. However, our HMM searches confirmed the absence of known TOM receptors Tom20 and Tom70 in excavates. These receptors either evolved only in certain eukaryotic lineages (Tom20) or were present in LECA (Tom70) as hypothesised here and by others [41]. To identify yet unknown peripheral TvTOM subunits, we performed proteomic analyses of the isolated TvTOM complex that indicated the presence of three TA proteins, in addition to Tom22-like protein. Two of them, Tom36 and Tom46, possess Hsp20-TPR-TMD architecture, which is similar to T. brucei receptor ATOM69. Indeed, we observed that Tom36 and Tom46 could bind to two hydrogenosomal preproteins, frataxin and αSCS, through binding assay. Tom36, Tom46, and ATOM69 are similar to yeast Tom70 with respect to the presence of TPR domains. The proximal TPR set of Tom70 interacts with Hsp90 [50] and may have an analogous function with the Hsp20 domain in Tom36, Tom46, and ATOM69 [11]. Of note, only Tom36 was tightly associated with TvTom40-2 and was detected in both high–molecular-weight complexes, whereas Tom46 appears to be loosely associated because it appeared only in the 330 kDa complex. This is similar to the loose association of Tom70 with the TOM complex that was reported in N. crassa [20] and the absence of Tom70 in the 550 kDa TOM complex in S. cerevisiae [14]. The third protein, Homp19, is unique to T. vaginalis, and neither HHpred nor PfamA searches identified any known functional domains.
It is tempting to speculate that the subunits with similar Hsp20-TPR-TMD architecture in both T. vaginalis and T. brucei evolved from a common excavate ancestor. However, our phylogenetic profiling of Hsp20-TPR-TMD proteins revealed that they were present exclusively in parabasalids and kinetoplastids but absent in the basal lineages, S. incarcerata (Discoba), Naegleria gruberi (Discoba), and C. membranifera (Metamonada). Therefore, such a distribution is more consistent with independent gains in parabasalid and kinetoplastid lineages. This is also supported by our cluster analysis and phylogeny of TPR domains, in which Tom36/Tom46 and ATOM69 displayed a polyphyletic origin. This finding is interesting considering the recent phylogenetic studies that challenged the monophyletic origin of Excavata [35,36]. Although the phylogenetic analysis of Excavata—including long-branch members such as trichomonads—placed Metamonada as a sister group of Discoba, when long-branch representatives were excluded, these two groups separated [35]. Regardless of whether the origin of Excavata is monophyletic or polyphyletic, Tom36/Tom46 and ATOM69 most likely represent an example of convergent evolution rather than a diversification of a common ancestor.
In spite of the presence of Tom40 and Tom22 homologues, the hydrogenosomal TvTOM complex revealed considerable differences compared with the mitochondrial TOM complex. There are several constraints to be considered for the specific shaping of TvTOM. Hydrogenosomes are adapted to operate under anaerobic conditions, which resulted in a vast reduction of mitochondrial functions and, consequently, a reduction in the proteome from 1,000–1,500 proteins in mitochondria [51–53] to approximately 600 proteins in T. vaginalis hydrogenosomes [10,54]. In yeast, the positively charged NTS, forming an amphipathic α-helix, interacts with Tom20, the cis and trans domains of Tom22, and the presequence-binding groove of the Tim50 receptor during translocation across the OMM [55]. The positive charge of the NTS contributes to the membrane potential (Δψ)-driven import step through TIM23 [56]. However, hydrogenosomes have lost the inner-membrane–associated respiratory chain that generates Δψ, and this loss has possibly triggered the positive net charge of NTS to become dispensable. Indeed, most hydrogenosomal NTSs possess only a single positively charged residue [57], are considerably shorter, are not essential for preprotein import, and—in a number of matrix proteins—are not present. Thus, the import of these proteins is based on recognition of poorly understood ITSs [25,26,57]. These changes in the targeting signals are likely reflected by the modifications in TOM receptors, the loss of both Tom22 trans domain and Tim50, and the divergence of downstream import machinery [10]. Collectively, the adaptation to anaerobiosis and the loss of Δψ were critical constraints that may have allowed mutation, leading to the divergence of the TvTOM complex. Another reason for the divergence of TvTOM could be different evolutionary history of the lineage. Our finding of trichomonad Tom36 and Tom46 in Parabasalia and the phylogenomic profiling of TOM components supports the notion that the peripheral TOM subunits were added to the core components after the separation of the main eukaryotic lineages.
T. vaginalis strain T1 (J. H. Tai, Institute of Biomedical Sciences, Taipei, Taiwan) and the recombinant strains were grown in Tryptone-Yeast extract-Maltose medium (TYM; pH 6.2) with 10% (v/v) heat-inactivated horse serum, without or with 200 μg/mL Geneticin 418 (Single transfectant), or with both 200 μg/mL Geneticin 418 and 40 μg/mL Puromycin (Double transfectant) at 37 °C. Recombinant E. coli strains were grown on Luria-Bertani medium with 100 μg/mL of Ampicillin at 37 °C. The yeast strains were grown either in liquid medium (SD-Leucine or SLac-Leucine) or on solid medium (SD-Leucine or YPG) at 30 °C. For drop dilution assays, cells were cultured to an OD600 of 1.0 and diluted 5-fold, followed by spotting 5 μL of each dilution on SD-Leu, SD-Leu supplemented with 2 μg/mL Dox, YPG, or YPG supplemented with 2 μg/mL Dox.
The genes encoding TvTom40-2 (TVAG_332970) and Sam50 (TVAG_178100) were cloned into a pTagVag2 vector fused to a 2×HA tag at the C-terminus [58]. The genes encoding Tom36 (TVAG_277930), Tom46 (TVAG_137270), Homp38 (TVAG_190830), Homp19 (TVAG_283120), and Tom22-like protein (TVAG_076160) were cloned into a pTagVagV5 vector fused to a 2×V5 tag at the C-terminus [59]. The plasmids were transfected by electroporation [58] into either the WT strain or the strain expressing HA-tagged TvTom40-2. For studies in yeast, TvTom40-2 was cloned into a pYX142 vector (Novagen) fused to an HA tag at the C-terminus. The plasmid with no insert or plasmid encoding either HA-tagged TvTom40-2 or ScTom40 was transformed into yeast cells (WT strain W303α, tet-TOM40, tom40-25, and tom40-34) by lithium acetate method. The tet-TOM40 yeast strain was constructed by inserting the tetracycline operator into the genome of WT strain, YMK120, upstream of TOM40 ORF by homologous recombination, using an insertion cassette amplified from the plasmid pMK632 as described previously [60]. Yeast strains carrying temperature-sensitive alleles of TOM40, tom40-25, and tom40-34 were obtained from elsewhere [61]. The oligonucleotides used are listed in S4 Table.
Tom40-like protein sequences from T. vaginalis were searched against the NCBI Conserved Domains database and the S. cerevisiae proteome or against Protein Data Bank (PDB) using the HHpred tool [62]. A Tom40-specific HMM was built using the HMMER3 hmmbuild module [63], with a set of 24 well-annotated Tom40 sequences (S1 Data) and was scanned against the T. vaginalis protein database on the HMMER3 jackhmmer tool with the default settings [64]. Human Tom22 and Tom7 sequences were searched against the NCBI nonredundant protein database using three PSI–Basic Local Alignment Search Tool (BLAST) iterations from different eukaryotic organisms. The alignments for Tom22 and Tom7 were constructed using MAFFT [65] with 447 (S6 Data) and 349 (S7 Data) sequences, which were used to build Tom22-specific and Tom7-specific HMMs, respectively, and were searched against the Trichomonas proteome database (www.trichdb.org) using HMMER3 [64].
The homologues of 14 TOM subunits were searched against the predicted proteomes of selected eukaryotes using HHsearch. The query alignments and their sources are given in S8 Data. The best hits were then checked for conserved domains using HHpred (https://toolkit.tuebingen.mpg.de/#/tools/hhpred) and were searched against the NCBI nonredundant protein database using BLAST. The transmembrane helices were predicted using TMHMM server version 2.0 (http://www.cbs.dtu.dk/services/TMHMM/) with a relaxed cutoff of 0.3. For CLANS [43], an extensive data set of Tom36 and ATOM69 homologues was prepared. Tom36 and ATOM69 protein sequences were used as queries to search against the NCBI nonredundant protein database using PSI–BLAST with two iterations, and the sequences with an e-value less than 0.1 were selected. Altogether, 1,114 sequences were used for CLANS, which was run with 10,000 iterations. The obtained 2D clustering data were processed to color-code taxonomies. The TMD was predicted using TMHMM with a relaxed cutoff of 0.3. A subset of 418 sequences from the data set was selected for the phylogenetic analysis of their TPR domains. The TPR domains were detected using HHsearch with TPR domains from the COG database (COG0790) as a query. Multiple sequence alignment was created with MAFFT [65], and the alignment was trimmed with BMGE [66], which resulted in 179 sites. The phylogenetic tree was constructed with IQ-TREE [67] using the LG + I + G4 model and 10,000 ultra-fast bootstrap replicates.
The model of TvTom40-2 was built using the N. crassa Tom40 structure (PDB ID 5o8o) as a template. The alignment was based on 140 Tom40 and VDAC sequences from a wide spectrum of eukaryotic organisms (S9 Data). The alignment was constructed by MAFFT, using the local pair alignment settings and 100 iterations [65] and later manually edited to reflect the secondary structure prediction of TvTom40-2 made by PSIPRED [68]. The three-dimensional (3D) structure model of TvTom40-2 was built using MODELLER 9v17 [69]. The quality of the final model was verified using ModFOLD 6 [70,71]. The electrostatic potential on the solvent-accessible surface of TvTom40-2 was calculated using APBS tool2 [72].
Trichomonas cells from a 1 liter culture were harvested and homogenised by sonication, and the subcellular fractions were isolated by differential centrifugation, as described previously [10]. Isolated hydrogenosomes (protein concentration 1 mg/mL) carrying either HA-tagged or V5-tagged proteins were washed to remove protease inhibitors and incubated for 30 minutes at 37 °C in isolation buffer (225 mM sucrose, 10 mM KH2PO4, 20 mM HEPES, 0.5 mM KCl, 5 mM MgCl2, and 1 mM EDTA [pH 7.2]) supplemented with either 100 μg/mL proteinase K enzyme (Roche Holding AG, Basel, Switzerland) or proteinase K with 0.5% Triton X-100. The incubation was terminated using 1 mM of phenylmethylsulfonyl fluoride (PMSF, Sigma Aldrich). Then, samples were analysed by immunoblotting using α-HA, α-V5, α-Fdx1, α-cytosolic malic enzyme, or α-αSCS antibody, followed by either α-mouse or α-rabbit antibody conjugated to peroxidase. The blot was developed using Amersham imager 600. Subcellular fractionation for yeast strains, and alkaline carbonate extraction and protease protection assay with isolated mitochondria were performed as described previously [73]. Proteins were separated by SDS-PAGE; immunoblotted with α-HA, α-HK, α-Fis1, or α-Aco antibody; and developed using an ECL system.
The cells for immunofluorescence microscopy were processed as previously described [74]. Recombinant proteins were visualised using mouse α-HA and rabbit α-V5 antibodies, and Alexa Fluor 488 donkey α-mouse and Alexa Fluor 594 donkey α-rabbit antibodies (Thermo Fisher Scientific). The hydrogenosomal marker malic enzyme was detected by rabbit polyclonal antibody. The slides were mounted using Vectashield containing DAPI (4',6-diamidino-2-phenylindole) (Vector laboratories). The cells were examined with an Olympus Cell-R IX-81 microscope, and the images were processed using ImageJ. For STED, Abberior STAR 580 α-mouse and Abberior STAR 635p α-rabbit antibodies, along with Abberior TDE mounting medium, were used. STED images were acquired on a commercial Abberior STED 775 QUAD Scanning microscope (Abberior Instruments) equipped with a Nikon CFI Plan Apo Lambda objective (60× Oil, NA 1.40). Abberior STAR580- and STAR 635P-labeled proteins were illuminated by pulsed 561 nm and 640 nm lasers and depleted by a pulsed 775 nm STED depletion laser of the 2D donut. Fluorescence signal was filtered (Emission bandpasses: 605–625 nm and 650–720 nm; pinhole 40 μm) and detected on single photon counting modules, with time gates set to 0.8–8.8 ns. Images were scanned with a pixel size of 20 nm × 20 nm, with a 10 μs dwell time and in-line interleaved acquisition mode using the Imspector software. All images were deconvolved with Huygens Professional version software 17.04 using the Classic Maximum Likelihood Estimation algorithm.
Isolated hydrogenosomes from the recombinant strains expressing tagged proteins were lysed with the native sample buffer (Life Technologies) containing either varying concentrations (1%–3%) of digitonin or 1% digitonin. The clarified extracts were electrophoresed on 3%–12% or 4%–16% NativePAGE bis-tris gel (Thermo Fisher Scientific), immunoblotted with either α-HA or α-V5 antibody, and developed by chemiluminescence. For BN-PAGE with yeast cells, isolated mitochondria from the strain with empty plasmid, or from strain expressing HA-tagged TvTom40-2, were lysed with lysis buffer containing 1% digitonin, and the clarified samples were electrophoresed on a 6%–13% native gel, immunoblotted with either α-HA or α-ScTom40 antibody, and developed using an ECL system.
CoIPs were performed for the HA-tagged TvTom40-2 either with or without crosslinker using isolated hydrogenosomes from both WT and recombinant strains. For crosslinking, interacting proteins in hydrogenosomes (protein concentration 1 mg/mL) were crosslinked with 1 mM DSP (dithiobis(succinimidyl propionate); Thermo Scientific) for 30 minutes at 25 °C, excess DSP was quenched with 50 mM Tris (pH 7.5), and the hydrogenosomes were washed twice with isolation buffer. For coIP, the hydrogenosomes (protein concentration 1 mg/mL) were solubilised in MKG buffer (10 mM MOPS [3-(N-morpholino)propanesulfonic acid; pH 7], 50 mM potassium acetate, 10% glycerol, and EDTA-free cOmplete protease inhibitor cocktail [Roche]) containing 1% digitonin (Merck Millipore), and the clarified extract was incubated with Dynabeads (Thermo Fisher Scientific) coupled with α-HA antibody for 90 minutes on an overhead rotator at room temperature. The beads were washed thrice before elution with either SDS-PAGE buffer for crosslinking coIPs or elution buffer (MKG buffer with 0.25% digitonin and 1 mg/mL HA peptide, Thermo Fisher Scientific) for native coIPs. The coupling of α-HA antibody to the Dynabeads was performed according to the manufacturer’s instructions.
LFQ-MS was performed according to standard procedures as described previously [59]. To remove SDS from the crosslinking coIP eluates and to remove HA peptides from the native coIP eluates, samples were resuspended in 8 M urea and processed using a Filter Aided Sample Preparation (FASP) protocol, according to Wisniewski et al. [75]. The samples were digested with trypsin and the peptides obtained were subjected to liquid chromatography-MS. The MS/MS spectra obtained were searched against the T. vaginalis database (downloaded from Trichomonas Genome Resource [TrichDB; www.trichdb.org] containing 59,862 entries), the quantifications were performed with the label-free algorithms, and the data analysis was performed using Perseus 1.5.2.4 software. The MS data have been deposited to the ProteomeXchange consortium via the PRIDE [76] partner repository. The MS data were obtained from four independent coIP experiments for each immunoprecipitated protein.
The TvTOM complex was purified under native conditions from hydrogenosomes isolated from the recombinant strain expressing C-terminal HA-tagged TvTom40-2 as described earlier. Five microliters of purified TvTOM complexes in solution was applied to copper electron microscopy grids (EMS200-Cu) covered with a 20 nm carbon film, which were glow discharged for 40 seconds with a 5 mA current prior to specimen application. Excess sample was removed after 1 minute by blotting (Whatman no. 1 filter paper) for 1 to 2 seconds, and the grid was immediately stained with 5 μL of 2% phosphotungstic acid for 1 minute 40 seconds and blotted to remove excess stain. A large data set of optimised, negatively stained specimen grids was acquired with a Tecnai F20 microscope (Thermo Fisher Scientific) operating at an accelerating voltage of 200 kV, with a FEI Eagle 4K CCD camera, at a magnification of 78,000× and a pixel size of 1.79 Å. Altogether, 1,000 images were acquired with defocus ranging from 2 to 5 μm. After quality inspection and determination of Contrast Transfer Function (CTF) parameters with the GCTF program [77], 650 micrographs were subjected to particle picking. Approximately 6,000 particles were manually picked from the first 200 micrographs with the e2boxer.py routine of the EMAN2 program [78] and subjected to three rounds of class averaging in Relion 1.4 [79], with 200, 150, and 100 classes, respectively. The box size was set to 192 pixels to accommodate higher-order multimers. This analysis resulted in a set of three representative class averages, which were low-pass filtered to 30 Å and used as templates for automated particle selection of the preselected set of 650 micrographs with the Gautomatch program. Altogether, 71,834 identified particles were subjected to five rounds of 2D classification in Relion with 200 classes, which reduced the data set to 10,038 particles. All 2D classifications comprised 40 iterations. The presented resolution of the class averages corresponds to the lowest SSNR value ≥1 indicated in the *model.star file resulting from the last iteration of the final 2D classification. The number of particles contributing to the class averages was also found in the *model.star files.
The gene encoding Ferredoxin1 (TVAG_003900) was cloned into NEB PURExpress control vector fused to the DHFR gene (E. coli) at the C-terminus. Radiolabeled TvFdx1-DHFR was synthesised in vitro in the presence of L-[35S] methionine (MGP spol sro) according to the manufacturer’s instructions (NEB PURExpress in vitro protein synthesis kit). Cytoplasmic extract was prepared from the T. vaginalis strain T1 as described elsewhere [24]. For the time course experiment, the import assay was conducted in a 500 μL reaction volume, and the mixture contained 500 μg of hydrogenosomes (protein concentration) carrying both TvTom40-2-HA and Tom36-V5, import buffer (250 mM sucrose, 10 mM MOPS-KOH [pH 7.2], 3% BSA, 80 mM KCl, 7 mM MgCl2, and 10 mM ATP), 125 μL cytosolic extract, and 25 μL radiolabeled precursors at 37 °C. At each time point, 100 μL was removed and shifted to ice, and the hydrogenosomes were re-isolated and washed twice with import buffer. For the import-arrest experiment, the import assay was performed either in the presence or absence of 10 μM methotrexate (Sigma Aldrich) and 1 mM NADPH. Wherever indicated, the hydrogenosomes were treated with 50 μg/mL of proteinase K. For the import-arrest and coIP assay, the import assay was performed either in the presence or absence of 10 μM methotrexate, the hydrogenosomes obtained were subjected to crosslinking, and the HA-tagged protein was immunoprecipitated as described earlier except that 0.5% Triton X-100 was used to lyse the organelles instead of digitonin. The samples were electrophoresed, and the gel was vacuum dried. The gel was exposed for 4 to 5 days prior to phosphorimaging with Typhoon TLA 7000 scanner.
The gene encoding for the cytosolic domain of Tom36 and Tom46 (Tom36cd and Tom46cd) were cloned into pET42b vector tagged to polyhistidine at the C-terminus. The genes encoding for cytochrome b5 (TVAG_063210), frataxin (TVAG_182610), and αSCS (TVAG_165340; αSCS was fused to DHFR to the C-terminus) were subcloned into NEB PURExpress control plasmid, and the radiolabeled precursors were synthesised in the presence of L-[35S] methionine as described earlier. The recombinant His-tagged proteins were expressed in E. coli BL21 (DE3) strain at 37 °C for 3 hours following the induction with 0.5 mM IPTG. The cells from a 10 mL culture of E. coli (negative control) and strains expressing His-tagged proteins were harvested, resuspended in 4.5 mL lysis buffer (50 mM NaH2PO4, 300 mM NaCl, 10 mM imidazole, 1 mg/mL lysozyme, and EDTA-free cOmplete protease inhibitor cocktail), incubated on ice for 45 minutes, and lysed using QSonica sonicator. The homogenised extract was clarified at 9,000 rcf for 30 minutes at 4 °C. Aliquots of supernatant and pellet were used for immunoblotting to test the solubility of the proteins. The supernatant obtained was split into three equal parts and was incubated with 50 μL of Ni-NTA agarose resin (Qiagen) on an overhead rotator for 2 hours at room temperature. The resin collected was washed five times using 10 volumes of wash buffer (50 mM NaH2PO4, 300 mM NaCl, 20 mM imidazole, and EDTA-free cOmplete protease inhibitor cocktail). To block, the beads were washed thrice with wash buffer II (50 mM NaH2PO4, 300 mM NaCl, 20 mM imidazole, 3% BSA, and EDTA-free cOmplete protease inhibitor cocktail). To the mock-treated beads or beads bound with His-tagged protein, binding buffer (50 mM Tris, 150 mM NaCl [pH 7.4]), 50 μL of Trichomonas cytosolic extract, and 10 μL of radiolabeled precursors were added and incubated for 1 hour at 37 °C with gentle shaking. The beads were washed three times with the binding buffer, and the proteins were eluted with the elution buffer (50 mM NaH2PO4, 300 mM NaCl, 500 mM imidazole, and EDTA-free cOmplete protease inhibitor cocktail). The samples were electrophoresed, and the gel was vacuum dried. The gel was exposed for 4 to 5 days prior to phosphorimaging with Typhoon TLA 7000. The oligonucleotides used for cloning are listed in S4 Table.
The gene encoding Sam50 was cloned into pET42b fused to a C-terminal His tag. The protein was expressed in E. coli BL21 (DE3) strain following an induction with 1 mM IPTG, and the His-tagged Sam50 was purified using affinity chromatography under denaturing conditions. The purified antigen was separated via SDS-PAGE, and the Coomassie-stained band was used to generate polyclonal antibody in rat.
|
10.1371/journal.pgen.1006249 | Multiple Independent Retroelement Insertions in the Promoter of a Stress Response Gene Have Variable Molecular and Functional Effects in Drosophila | Promoters are structurally and functionally diverse gene regulatory regions. The presence or absence of sequence motifs and the spacing between the motifs defines the properties of promoters. Recent alternative promoter usage analyses in Drosophila melanogaster revealed that transposable elements significantly contribute to promote diversity. In this work, we analyzed in detail one of the transposable element insertions, named FBti0019985, that has been co-opted to drive expression of CG18446, a candidate stress response gene. We analyzed strains from different natural populations and we found that besides FBti0019985, there are another eight independent transposable elements inserted in the proximal promoter region of CG18446. All nine insertions are solo-LTRs that belong to the roo family. We analyzed the sequence of the nine roo insertions and we investigated whether the different insertions were functionally equivalent by performing 5’-RACE, gene expression, and cold-stress survival experiments. We found that different insertions have different molecular and functional consequences. The exact position where the transposable elements are inserted matters, as they all showed highly conserved sequences but only two of the analyzed insertions provided alternative transcription start sites, and only the FBti0019985 insertion consistently affects CG18446 expression. The phenotypic consequences of the different insertions also vary: only FBti0019985 was associated with cold-stress tolerance. Interestingly, the only previous report of transposable elements inserting repeatedly and independently in a promoter region in D. melanogaster, were also located upstream of a stress response gene. Our results suggest that functional validation of individual structural variants is needed to resolve the complexity of insertion clusters.
| The presence of several transposable element insertions in the promoter region of a Drosophila melanogaster gene has only been described in heat shock protein genes. In this work, we have discovered and characterized in detail several naturally occurring independent transposable element insertions in the promoter region of a cold-stress response gene in the fruitfly Drosophila melanogaster. The nine transposable element insertions described are clustered in a small 368 bp region and all belong to the same family of transposable elements: the roo family. Each individual insertion is present at relatively low population frequencies, ranging from 1% to 17%. However, the majority of strains analyzed contain one of these nine roo insertions suggesting that this region might be evolving under positive selection. Although the sequence of these insertions is highly similar, their molecular and functional consequences are different. Only one of them, FBti0019985, is associated with increased viability in nonstress and in cold-stress conditions.
| Promoters are crucial regions for the transcriptional regulation of gene expression. Recent computational and experimental advances in functional genomics techniques have allowed defining the promoter architecture to an unprecedented level. Several core promoter motifs such as the Initiator (Inr) and the Downstream core Promoter Element (DPE) have been described, and it is likely that many others remain to be discovered. The presence or absence of the core promoter motifs influences enhancer-promoter communication and thus gene regulation [1]. Promoter regions also harbour transcription factor binding motifs, which are another important component in the regulation of gene expression [2]. Besides cis-regulatory elements that influence the temporal and spatial expression patterns of genes, proximal promoters often contain alternative transcription start sites (TSSs) [1, 3]. Rather than being “biological noise” from imprecise binding of the transcription initiation machinery, genome-wide analyses of TSSs usage showed that alternative TSSs play an important role in the diversification of gene expression patterns [4–8].
Transposable elements (TEs), long proposed to play an important role in gene regulation [9, 10], have recently been found to provide at least 1,300 alternative TSSs in the Drosophila melanogaster genome [8]. TEs can also add Transcription Factor Binding Sites (TFBSs) to the promoter of genes as has been recently shown in Drosophila and humans [11–13]. As a result of adding particular sequence elements, many TEs confer their intrinsic regulatory properties to nearby genes demonstrating that they distribute cis-regulatory modules [8]. Finally, TEs inserted in promoter regions can also influence gene expression by disrupting the promoter architecture. This is the case, for example, of naturally occurring P-element insertions in the promoter of heat shock protein (hsp) genes [14].
One of the TEs identified as providing an alternative TSS by Batut et al (2013) [8], named FBti0019985, was previously reported in a screening designed to identify putatively adaptive TE insertions in D. melanogaster [15]. However, this particular TE was not further studied because its population frequency could not be accurately determined [15]. FBti0019985 is a roo solo-LTR inserted in the 5’-UTR of CG18446 gene, which is nested in the first intron of crossbronx (cbx) (Fig 1). TEs from the roo family have long been proposed to affect the expression of nearby genes by adding and distributing cis-regulatory regions [16–19]. Specifically, roo LTRs contain several TFBSs and the Inr sequence characteristic of core promoters [8, 20].
Interestingly, CG18446 has been identified as a candidate gene for cold resistance: it is upregulated in fly strains that have been selected for increased cold resistance compared with control strains that were not subjected to cold-stress [21]. Cold resistance is an ecologically and evolutionarily relevant trait because it influences the ability of the species to adapt to different climatic conditions and thus, their geographical distribution [22, 23]. There is good evidence suggesting that D. melanogaster adapts to cold environments and a growing list of candidate genes involved in this thermotolerance phenotype is being identified [21, 24–28]. However, the molecular variants responsible for the adaptive cold-stress resistance phenotype remain elusive [29].
In this work, we further analyzed the presence/absence of FBti0019985 in four natural populations of D. melanogaster. We found that besides FBti0019985, eight other roo elements have inserted in a 368 bp region around CG18446 transcript start site. These roo elements differ in the insertion site and in their orientation. On the other hand, all elements have the same size and show high sequence conservation: all cis-regulatory elements previously described in roo LTRs are highly conserved [8, 30]. We further investigated whether these different insertions were functionally equivalent by performing 5’-RACE, gene expression, and phenotypic analyses. Our results showed that the functional consequences of the different roo insertions depend on the particular position where the element is inserted. Among the nine different roo solo-LTR insertions, only FBti0019985 is consistently associated with increased viability in nonstress and cold-stress conditions across genetic backgrounds.
We first aimed at estimating the frequency of FBti0019985 in non-African natural D. melanogaster populations. Thus, we checked using PCR whether this insertion was present, polymorphic, or absent in 28 strains from a natural population collected in North Carolina (North America, DGRP strains [31, 32]) and in 15 strains from a natural population collected in Bari (Italy, Europe [33]) (Table 1). We obtained PCR results for 39 of the 43 strains tested: nine strains produced PCR bands consistent with FBti0019985 being present, five strains appeared as heterozygous, 13 strains showed unexpected band patterns, and 12 strains appeared as absent (Table 1) (see Material and Methods). To verify these results, we sequenced 32 of the 39 strains including all the strains that showed some evidence of presence (Table 1).
Only four of the nine strains classified as present, according to the PCR results, had the FBti0019985 insertion. For the rest of this work, we considered the position where FBti0019985 is inserted as the "reference position". The other five present strains, the five heterozygous strains, and 12 of the 13 strains that gave unexpected PCR bands contained different roo solo-LTR insertions (Table 1). Overall, besides FBti0019985, we found eight other 428 bp roo solo-LTRs inserted in eight different positions (Fig 2). Three roo insertions are located downstream of the reference position: roo+7, roo+175, and roo+278 (Fig 2). Two of the four strains carrying roo+7 have a duplication of the 95 bp region located immediately upstream of the insertion (Table 1). roo+175 element is inserted in the 5’-UTR region, and roo+278 is inserted in the first exon of CG18446 gene. Both roo+175 and roo+278 have a conserved Inr motif. If transcription starts in these insertions, flies carrying roo+175 would have a 100 bp shorter 5'-UTR, and flies carrying roo+278 would have a 35 amino acids shorter CG18446 protein. The other five roo insertions are located upstream of the reference position: roo-19, roo-28, roo-44, roo-68, roo-90 (Fig 2). Four of them, roo-19, roo-28, roo-44, and roo-68, are inserted in reverse orientation.
We used Tlex-2 software to further analyze the frequency of the nine roo insertions in 21 additional DGRP strains, in 26 strains from a Swedish natural population, and in 42 strains from a population collected in the ancestral range of the species, Zambia (Fig 2 and S1 Table) (see Material and Methods) [34]. Overall, we found that 67 strains, out of the 128 strains analyzed, contained one of the nine roo solo-LTR insertions. The two most common roo insertion in out-of-Africa populations are roo-90 and FBti0019985 present in 13% and 10% of the strains tested, respectively (Fig 2). Besides, some insertions are only present in the North Carolina natural population while others are specific to the Italian natural population (Fig 2). Only three of the nine insertions described in North Carolina and Italian populations are present in the Swedish population. However, we did not perform de novo discovery of TEs in this population. Thus, it could be that other private insertions are present in the Swedish population. Finally, all the nine insertions were present in the African population although most of them were present at very low frequencies (Fig 2).
In summary, we have found that besides the FBti0019985 insertion annotated in the reference genome, eight other 428 bp roo solo-LTRs are inserted nearby CG18446 TSS in natural populations of D. melanogaster (Fig 2) [35]. Each one of the strains analyzed contains a single solo-LTR roo insertion and most of the analyzed strains contain one of the nine solo-LTR roo insertions.
We identified the Target Site Duplications (TSD) of the nine different roo insertions using data from the 26 present strains sequenced in this work (Table 1). We could identify the TSD for all roo insertions except for roo+278. We found that six of the eight TSDs identified are five nucleotides long as has been previously described for this family [36] (Fig 2). However, the TSD sequences did not match the proposed TSD consensus sequence [34, 36, 37]. We thus used all the available roo TSD sequences to build a new consensus (S1 Fig). The different roo solo-LTR insertions had different TSDs suggesting that they are independent insertions (Fig 2). Furthermore, all the roo elements located in a given insertion site have the same exact TSD and are inserted in the same orientation suggesting that each one of them is a unique insertion event (Fig 2).
To test whether these nine insertion events were the result of a burst of transposition, we constructed a phylogenetic tree. We included the nine roo insertions sequenced in this work and 115 other roo insertions present in the D. melanogaster genome (S2 Fig and S1 Text). We found that not all the newly described roo insertions clustered together suggesting that they did not insert at the same time (S2 Fig and S1 Text).
All the TEs identified in CG18446 proximal promoter region belong to the roo family. Thus, we also investigated whether roo elements annotated in the reference genome are preferentially inserted into gene proximal promoter regions as has been previously described for other TE families [38, 39]. We analyzed the 138 insertions belonging to the roo family annotated in the D. melanogaster reference genome (v5). We found 21 roo insertions located in the 1 kb region upstream of a gene or overlapping the 5’-end of a gene. Thus, only 15.2% of the roo elements in the D. melanogaster genome are located in gene promoters and/or 5’-UTRs.
In summary, TSD analyses of the nine insertions characterized in this work suggested that they are independent insertions, and confirmed the length but not the sequence previously reported as the TSD consensus for this family. Our results are not consistent with the nine roo insertions being the result of a single burst of transposition. Finally, our analyses also suggested that roo elements do not preferentially insert in 5’ gene regions.
We analyzed multiple sequence alignments of all the roo insertions located nearby CG18446. We identified TFBSs using the JASPAR database (see Material and Methods). We also specifically looked for conservation of the regulatory regions previously described in the roo family [8, 30], and for conserved core promoter motifs [1] (Fig 3A and S2A Table). Overall, there was very little diversity among the nine solo-LTRs (S3A Fig). The five TFBSs and the Inr sequence previously identified in the consensus sequence of roo LTRs are conserved in all the roo copies located in the proximal promoter of CG18446 [8]. Additionally, we found another four TFBSs that are also highly conserved in all the copies (Fig 3A and S3A Fig). The nine transcription factors are involved in developmental processes. Additionally, Deaf1 and Nub are also involved in immune response [40, 41]. Finally, three previously identified Matrix Associated Regions (MARs) in LTRs from the roo family are also highly conserved in the nine insertions (Fig 3A and S3B Fig) [30]. These results suggest that these roo solo-LTR insertions are introducing the same cis-regulatory regions in the CG18446 proximal promoter region. Still, the functional effect of these insertions might be different because they are located in different positions and have different orientations (Fig 2).
We analyzed the proximal promoter region of CG18446 in the 30 strains sequenced in this work. We could not identify the TATA box suggesting that CG18446 has a DPE promoter [1]. We identified eight TFBSs in the proximal promoter of CG18446 (Fig 3B and S2C Table). These eight TFBSs are highly conserved in all the strains analyzed (S3C Fig). The different roo insertions characterized in this work do not disrupt any of the identified core promoter motifs or TFBSs (Fig 3B). However, they do affect the spacing between the different regulatory motifs, which might affect the protein-protein interaction at the CG18446 promoter and thus the expression level of this gene (Fig 3B) [14].
Besides affecting the spacing of transcription factor binding site, another mechanism by which roo insertions could be affecting CG18446 expression is by recruiting piRNAs that would lead to heterochromatin formation [42, 43]. We mapped piRNA reads from three different available libraries to a 1.4 kb region including FBti0019985 (Fig 4A) (see Material and Methods) [44–46]. We found that most of the piRNAs mapping to the insertion were sense reads, suggesting that FBti0019985 is not acting as a target for heterochromatin assembly [42].
We also looked for evidence of HP1a binding to FBti0019985 using modENCODE data (see Material and Methods) [47]. HP1a is a structural chromosomal protein that mediates both gene expression and gene silencing [48]. We did find evidence of HP1a reads binding to FBti0019985 (Fig 4B). Thus, by recruiting HP1a, FBti0019985 could be affecting the expression of CG18446. The same results were obtained for the other eight roo solo-LTR insertions: most of the piRNAs mapping to the insertions were sense reads and we found evidence of HP1a binding to all of them (S3 Table). Overall, our results are suggestive but not conclusive of HP1a binding to the nine roo insertions described in this work.
To further investigate the possible functional consequences of the roo insertions, we focused on the five insertions present at higher population frequencies in out-of-Africa populations: FBti0019985, roo+7, roo-44, roo-90, and roo-68 (Fig 2).
We investigated whether roo insertions could be providing an alternative TSS to CG18446. Batut et al (2013) [8] reported that the TSS of CG18446 is located inside FBti0019985. However, this finding was obtained using RAMPAGE and was not further validated using 5’-RACE. For this reason, we performed a 5’-RACE with the RAL-810 strain that carries FBti0019985 and with the RAL-783 strain that does not carry any of the nine roo solo-LTR insertions. As expected, we found that the TSS of CG18446 is inside the TE: the first 50 bp of the 276 bp 5’-UTR correspond to FBti0019985 (Fig 5). Additionally, flies with the insertion have also a shorter transcript, with a 201 bp 5’-UTR, that does not start in FBti0019985 (Fig 5). Most of the sequenced transcripts start in the FBti0019985 insertion (14 out of 20 transcripts analyzed). Flies without the FBti0019985 insertion only have the 201 bp 5’-UTR transcript (Fig 5).
We then checked whether roo+7, located only 7 bp downstream of FBti0019985, roo-90, which is the most distal insertion, and roo-44, which is inserted in reversed orientation, also provide an alternative TSS to CG18446. We found that roo+7 affects the TSS of CG18446 (Fig 5). Indeed, the TSS in roo+7 is in the same nucleotide position as in FBti0019985. Thus, CG18446 transcript in flies with roo+7 is 7 bp shorter compared with the transcript in flies with FBti0019985. Similarly to FBti0019985, most of the sequenced transcripts started in the roo+7 insertion (18 out of 22 transcripts analyzed). On the other hand, we did not find evidence of a TSS inside roo-90, which might indicate that the distance of the TE to the nearby gene affects its ability to provide an alternative TSS (Fig 5). Finally, we analyzed two different strains carrying the roo-44 insertion in the same position and we could not find evidence for a transcript with the TSS in roo-44 (Fig 5).
Overall, we found that only FBti0019985 and roo+7 insertions modify the length of CG18446 transcript. These two roo insertions are located a few nucleotides from the gene and both are inserted in 5’ to 3’ orientation.
We further analyzed whether different roo insertions were associated with changes in CG18446 expression in embryos, where this gene is highly expressed [49]. For FBti0019985, we analyzed the expression of CG18446 in flies with four different genetic backgrounds. In three of the four backgrounds, FBti0019985 is associated with upregulation of CG18446 (Fig 6A). This result is significant in two genetic backgrounds, RAL-810 and IV68, and marginally significant in a third background, RAL-639 (t-test p-value = 0.045, p-value = 0.005 and p-value = 0.062, respectively) (Fig 6A). On the other hand, only in one of the three genetic backgrounds analyzed for roo+7, the insertion is associated with downregulation of this gene (t-test p-value = 0.015 for RAL-405) (Fig 6B).
We also checked the expression of CG18446 in flies with two roo solo-LTR insertions that do not provide an alternative TSS to this gene: roo-90 and roo-44. We found that roo-90 is only associated with CG18446 upregulation in one of the three backgrounds analyzed (p-value = 0.001, for RAL-21) (Fig 6C). Two different strains with the roo-44 solo-LTR insertion did not show differences in the level of expression of CG18446 compared with strains without the insertion (p-values > 0.05 in both cases) (Fig 6D).
Overall, we found that FBti0019985 is associated with CG18446 upregulation in three of the four backgrounds analyzed (Fig 6A). In the majority of strains, roo+7, roo-90, and roo-44 are not associated with changes in CG18446 expression level (Fig 6B–6D). However, we can not discard that the presence of these insertions is associated with changes in the expression of CG18446 in other developmental stages and/or in tissues not analyzed in this work.
We have shown that FBti0019985 affects the transcript length and it is associated with upregulation of CG18446 in most of the genetic backgrounds analyzed (Figs 5 and 6A). Because CG18446 has been previously identified as a cold-stress candidate gene, we tested whether flies with and without FBti0019985 differed in their sensitivity to cold-stress [21]. We first compared RAL-810, which carries FBti0019985, with RAL-783, which does not carry any of the nine roo insertions (Fig 7A). We performed three biological replicates. ANOVA analyses showed that the experimental condition (nonstress or cold-stress) and the insertion genotype (presence or absence of FBti0019985) were significant (Table 2). Flies with FBti0019985 had a higher viability than flies without this insertion in both nonstress and cold-stress conditions. Furthermore, the interaction between these two factors was also significant suggesting that the effect of the insertion is larger in cold-stress conditions (Fig 7A and Table 2).
We repeated the experiment using flies with different genetic backgrounds: RAL-802 that carries FBti0019985 and RAL-908 that does not carry this insertion (Fig 7B). ANOVA analyses showed that the experimental condition and the insertion genotype are significant while the interaction between these two factors was not significant (Table 2). RAL-802 flies had a higher egg-to-adult viability in nonstress and in cold-stress conditions compared with flies without FBti0019985.
Finally, we tested whether flies from a different population, IV68 carrying FBti0019985 and IV22 without this particular insertion both collected in Italy, also showed significantly increased viability in nonstress and in cold-stress conditions (Fig 7C and Table 2). We found that IV68 flies had a higher viability than flies without the FBti0019985 insertion in both nonstress and cold-stress conditions (Table 2).
Overall, we found consistent results, across genetic backgrounds from two different natural populations, suggesting that flies with the FBti0019985 insertion are associated with increased viability compared to flies without this insertion in nonstress and in cold-stress conditions. In all cases, the effect of the presence of the insertion was either medium or large (Table 2). In one of the genetic backgrounds, the effect was larger under cold-stress conditions (Fig 7A) while no interaction between experimental condition and insertion genotype was found in the other two backgrounds (Fig 7B and 7C).
We further checked whether another four roo solo-LTR insertions described in this work are associated with cold-stress phenotypes. For each insertion, we compared the egg-to-adult viability of flies with two different genetic backgrounds with the egg-to-adult viability of RAL-783 that does not carry any of these insertions (Fig 8). In all cases, we performed ANOVA analyses to check whether the experimental conditions, insertion genotype, and/or the interaction between these two factors were significant (Table 2).
We found that the experimental condition had a significant effect on egg-to-adult viability in all the strains tested (Table 2). On the other hand, the effect of the insertion was only significant in some of the genetic backgrounds (Table 2). Among strains that carry the roo+7 insertion, the insertion genotype had an effect only in one of the two backgrounds tested (Fig 8A and 8B and Table 2). RAL-405 flies with roo+7 insertion showed decreased viability (Fig 8A and Table 2). The presence/ absence of roo-90 did not have a significant effect on egg-to-adult viability (Fig 8C and 8D and Table 2). For roo-44, while the insertion genotype had a significant effect on the two backgrounds tested, results were not consistent. In one background, the presence of the insertion is associated with increased viability under cold-stress conditions and the interaction between the treatment and the insertion genotype is significant (Fig 8E and Table 2), while in the other background the presence of roo-44 is associated with decreased viability (Fig 8F and Table 2). Finally, the presence of roo-68 significantly affected viability in only one of the two backgrounds tested: RAL-716 flies carrying roo-68 showed decreased viability (Fig 8H and Table 2).
Overall our results suggested that the presence of roo+7, roo-90, roo-44, and roo-68 solo-LTR insertions reported in this work was not consistently associated with cold-stress phenotypes (Fig 8). These other insertions could have no phenotypic effect or could be involved in phenotypes not analyzed in this work.
We looked for evidence of positive selection in the 2 kb region flanking the FBti0019985 insertion. We analyzed the number of segregating sites (S) in this region and estimated Tajima´s D, iHS, nSL, H12 and XP-EHH (see Material and Methods). We found reduced diversity in the strains with FBti0019985: the number of segregating sites in this region is significantly smaller than the number of segregating sites found in 2 kb regions of chromosome 2R, where the FBti0019985 insertion is located (p-value = 0.015) (S4 Table). We also found that Tajima’s D was significantly negative in the 2 kb region where FBti0019985 is inserted, as expected if this region is under positive selection (p-value = 0.009) (S4 Fig and S4 Table). Finally, we also found significant values of iHS and H12 in the region flanking the FBti0019985 insertion (p-value = 0.048 and p-value = 0.023, respectively) (S5 Fig and S4 Table).
We also looked for evidence of selection taking into account not only the strains in which FBti0019985 is inserted, but all the strains that contain one of the nine roo insertions described in this work. In this case, only iHS showed a marginally significant value (p-value = 0.049) (S6 Fig).
Overall, our results suggest that the strains carrying FBti0019985 might be evolving under positive selection while the evidence for positive selection taking into account all the strains with one of the nine roo solo-LTRs, was only marginally significant.
Besides FBti0019985, we have discovered eight other roo solo-LTR elements inserted in the 368 bp region nearby the TSS of the cold-stress response gene CG18446 (Fig 2) [21]. Each strain contained a single roo insertion and the population frequency of the different individual insertions varies from 1% to 17% (Fig 2). Full-length elements from the roo family are 8.7 kb long. Such long insertions in the proximal promoter of CG18446 located in the first intron of cbx, might be deleterious, which could explain why all the identified insertions were solo-LTR elements. In D. melanogaster, repeated insertions of TEs have only been described in the proximal promoters of a particular gene class: hsp genes [50]. The susceptibility of hsp genes to TE insertions was attributed to their peculiar chromatin architecture: constitutively decondensed chromatin and nucleosome-free regions [51, 52]. However, promoter regions of non-hsp genes with similar chromatin architecture are not targets for TE insertions suggesting that chromatin accessibility is not sufficient to explain the susceptibility of hsp genes to TE insertions [50]. From a functional point of view, the presence of TEs in the promoter regions of hsp genes has been suggested to allow a rapid gene expression response to unpredictable temperature changes [50]. Similarly, the presence of roo insertions in the promoter of CG18446 could also be enhancing the ability of this gene to respond to environmental challenges, although only one of the nine roo insertions was associated with cold-stress tolerance (see below). Interestingly, almost 100% of the insertions described in heat-shock genes are P-element insertions, and all the insertions described here are roo elements. P-elements preferentially insert in the 5' end of genes where they recognize a structural motif rather than a sequence motif [38, 39]. While 81% of P-elements insert in 5’ gene regions, our results showed that only 15.2% of the roo elements annotated in the reference genome are inserted in 5’ gene regions. Thus, with the data currently available, roo insertions do not seem to preferentially insert into 5’ gene regions although analyses of de novo insertions should shed more light on this issue.
Our results showed that the different roo elements inserted in the proximal promoter of CG18446 differ in their molecular and functional effects (Table 3). We found that the two insertions that are more closely located to CG18446, FBti0019985 and roo+7, provided an alternative TSS to this gene (Fig 5 and Table 3). However, only FBti0019985 is associated with upregulation of CG18446 expression (Fig 6 and Table 3). Besides providing an alternative TSS, the effect of the FBti0019985 insertion on CG18446 expression could be due to the addition of new regulatory regions (Fig 3A), to the disruption of the spacing of pre-existing ones (Fig 3B), and/or to the recruitment of HP1a protein that could also lead to changes in the expression of CG18446 (Fig 4B). Finally, we cannot discard that polymorphisms other than the presence/absence of the FBti0019985 insertion also affect the expression of CG18446.
We found that the FBti0019985 insertion, which is associated with increased CG18446 expression, is consistently associated with increased viability in nonstress and in cold-stress conditions (Fig 7 and Table 3). Although we cannot exclude that other variants linked to FBti0019985 contribute to the increased viability phenotypes, we argue that it is unlikely that the association between the FBti0019985 insertion and increased viability in three different genetic backgrounds from two different natural populations would occur spuriously [53]. These results also suggest that CG18446 is likely to play a role in cold tolerance as was previously suggested based on cold-stress selection experiments in which this gene was found to be overexpressed [21]. However, FBti0019985 is present in only 10% of the out-of-Africa natural strains analyzed in this work. Our screening was focused on three out-of-Africa populations, thus we cannot discard that FBti0019985 is present at higher frequencies in other populations. Alternatively, it is also possible that the relatively low frequency of FBti0019985 is due to negative fitness effects of this insertion on other phenotypes. Cold-stress resistance has been associated with decreased starvation resistance [54, 55] and reduced fecundity [56, 57]. Therefore, the benefit of flies carrying FBti0019985 in cold-stress conditions might be a cost, for example, when food resources are scarce.
While FBti0019985 has a consistent cold-stress tolerance phenotype, four other roo insertions also located on the proximal promoter of CG18446 did not (Fig 8 and Table 3). The insertion that is present at higher frequencies in out-of-Africa populations is roo-90 (Fig 2). However, this insertion is not associated with changes of expression of CG18446 in embryos (Fig 6) and was not found to be associated with cold-stress tolerance phenotypes (Fig 8C and 8D and Table 3). It could be that this insertion has no phenotypic effect. Alternatively, roo-90 could be affecting a phenotype other than cold tolerance. A recent update in FlyBase revealed that CG18446 is also an ethanol-regulated gene that could contribute to ethanol sensitivity or tolerance [58]. Another possibility is that roo-90 affects cbx. As the other roo insertion described in this work and CG18446 gene, roo-90 is inserted in the first intron of cbx which has been functionally classified as a defense response to bacterium and spermatogenesis gene [59] (Fig 1). Elucidating whether roo-90 has an adaptive effect is beyond the scope of this paper.
Overall, we did not find evidence of positive selection at the DNA level in the region where the nine roo solo-LTR elements are inserted. We did find evidence of reduced diversity in this region when only the strains containing FBti0019985 were considered (S4–S6 Figs and S4 Table). Further analyses with a bigger dataset of strains is needed in order to determine whether this region shows signals of positive selection at the DNA level.
In summary, our results showed that different TE insertions in the same gene promoter region might have different molecular and functional consequences. Thus, the description of complex regions, as the one reported in this work, should be followed by functional analysis of the structural variants if we want to elucidate which ones are functionally relevant.
We used inbred strains from the Drosophila Genetic Reference Panel (DGRP [31, 32]) and isofemale strains from an Italian population collected in Castellana Grotte (Bari, Italy [33]) to perform the molecular and phenotypic assays.
We used a PCR approach to check for presence/ absence of FBti0019985 in 28 strains from the North Carolina population and in 15 strains from Italy. The primers used were FBti0019985_FL (5’-GGCATCATAAAACCGTTGAACAC-3’), FBti0019985_L (5’-AGTCCCTTAGTGGGAGACCACAG-3’) and FBti0019985_R (5’-CGTAGGATCAGTGGGTGAAAATG-3’) (Fig 1). Primers FBti0019985_L and FBti0019985_R are expected to give a 616 bp band when the TE is present. Primers FBti0019985_FL and FBti0019985_R are expected to give a 638 bp band when the TE is absent and a 1066 bp band when the TE is present. All PCR bands giving evidence of presence and some of the PCR bands giving evidence of absence were cloned using TOPO TA Cloning Kit for Sequencing (Invitrogen) following the manufacturer’s instructions and Sanger-sequenced using M13 forward and/or M13 reverse primers to verify the results. Sequences have been deposited in GenBank under accession numbers KU672690-KU672720.
We estimated the frequencies of the nine roo solo-LTR insertions described in this work using T-lex2 software [34]. Because T-lex2 works only for annotated TEs, we constructed eight new reference sequences including each one of the newly described roo solo-LTR insertions. The new reference sequences included 500 bp at each side of the TE and the TSD of each insertion.
We run T-lex2 in strains from three different populations: 50 strains from North Carolina (DGRP [31, 32]), 27 strains from a population collected in Stockholm, Sweden [33], and 67 strains from a population collected in Siavonga, Zambia [60]. As a control, we also run T-lex2 in the strains for which we have PCR results (S1 Table). We obtained results for 21 out of 50 DGRP strains, 26 out of 27 Swedish strains and 42 out of 67 Zambian strains. In some of the strains, T-lex2 detects more than one insertion per strain. However, PCR analyses of these strains revealed that only one insertion was present. These results suggest that T-lex2 cannot accurately estimate the frequency of insertion when they are closely located to each other. We thus discarded T-lex2 results indicating the presence of more than one insertion per strain. Other factors such as the quality of the reads and the coverage of the different strains could also be affecting T-lex2 results.
Target site motifs were constructed in WebLogo (http://weblogo.berkeley.edu) using six TSDs sequences obtained in this work and 41 TSDs sequences predicted with T-lex2 software [34].
For each roo solo-LTR insertion, we constructed a consensus sequence taking into account the 26 strains sequenced in this work using Sequencher 5.0 software. We aligned the nine roo insertion consensus sequences with 115 of the 137 other roo insertions present in the D. melanogaster genome using the multiple sequence aligner program MAFFT [61]. The quality sequence of the other 22 roo insertions was too low to include them in the alignment. A maximum likelihood tree was inferred using RAxML Version 8 [62] under the general time-reversible nucleotide model and a gamma distribution of evolutionary rates. We use the ETE toolkit Python framework for the analysis and visualization of trees [63].
We looked for conservation of the Transcription Factor Binding Sites (TFBSs) previously described in the roo family [8] in all the roo solo-LTRs characterized in this work. First, we downloaded from FlyBase version r6.06 (http://flybase.org) the fasta file of FBti0019985 sequence (genome region 2R: 9,871,090–9,871,523). We also searched for TFBSs in the roo insertions and in the CG18446 promoter regions using all the available JASPAR CORE Insecta matrices (http://jaspar.genereg.net). Only those sites predicted with a relative score higher than 0.995 were considered. We identified four new TFBS in FBti0019985 sequence: Deaf1, ara, mir, and caup. We then look for conservation of the identified motifs in all the roo solo-LTR sequences described in this work. For some strains, we used the information available in http://popdrowser.uab.cat [64].
We used three piRNA libraries [44–46] to map piRNA reads to a 1.4 kb region including FBti0019985 and to all the roo insertions described in this work following the methodology described in Ullastres et al (2015) [33]. Briefly, we used BWA-MEM package version 0.7.5 a-r405 [65] to align the reads and then we used SamTools and BamTools [66] to index and filter by sense/antisense reads. The total read density was obtained using R (Rstudio v0.98.507) [67].
We used modENCODE ChIP-Seq data [47] to map HP1a reads to a 1.4 kb region including FBti0019985 and to all the roo insertions described in this work following the methodology described in Ullastres et al (2015) [33]. We aligned the reads using BWA-MEM package version 0.7.5 a-r405 [65]. The total read density was obtained using R (Rstudio v0.98.507) [67].
5-to-7 day-old flies were placed in a fly cage with egg-laying medium (2% agar with apple juice and a piece of fresh yeast) during 4 hours. Then, adult flies were separated and embryos were collected following the suspension method described in Schou (2013) [68]. Embryo dechorionation was done by bleach (50%) immersion. Total RNA was extracted using TRIzol Plus RNA Purification Kit (Ambion). RNA was then treated on-column with DNase I (Thermo) during purification, and then treated once more after purification. 5’-RACE was performed with FirstChoice RLM-RACE Kit and using Small-scale reaction RNA processing with RNA samples of RAL-783 (roo-), RAL-810 (FBti0019985), RAL-405 (roo+7), RAL-21 (roo-90), RAL-383 (roo-44) and RAL-195 (roo-44). The gene specific outer primer was 5’-GACACTCTTCGGTTGGTGGA-3’ and the gene specific inner primer was 5’-ACAACTGTTCTGTAGGATCGC-3’. The control primer was 5’-TAGTCCGCAGAGAAACGTCG-3’. Inner PCR products were then cloned and Sanger-sequenced as mentioned above. Sequences have been deposited in GenBank under accession numbers KU672721-KU672722.
Embryo collection and RNA extraction was performed as described before. Reverse transcription was carried out using 500 ng of total RNA using Transcriptor First Strand cDNA Synthesis Kit (Roche). The cDNA was then used in a 1/50 dilution for qRT-PCR with SYBR green master-mix (Bio-Rad) on an iQ5 Thermal cycler. CG18446 expression was measured using specific primers (5’-GAGCAGTTGGAATCGGGTTTTAC-3’ and 5’-GTATGAATCGCAGTCCAGCCATA-3’) spanning 99 bp cDNA in the exon 1/exon 2 junction of CG18446. The primer pair efficiency was 99,1% (r2 larger than 0.99). CG18446 expression was normalized with Act5C expression levels (5’-GCGCCCTTACTCTTTCACCA-3’ and 5’-ATGTCACGGACGATTTCACG-3’).
Embryo collection was performed as mentioned above. Embryos were put into 50 ml fresh food vials. When embryos were 4–8 hour-old, they were kept at 1 C for 14 hours and then they were kept at room temperature (22–25 C). Simultaneously, control vials were always kept at room temperature (22–25 C) and never exposed to cold-stress. A total of 8–20 vials were analyzed per experiment. The same number of embryos per vial, 30 or 50, were used for all the replicates of a given experiment. Percentage viability was calculated based on the number of emerged flies to the total number of embryos placed in each vial.
Statistical significance was calculated performing two-way ANOVA using SPSS v21. We combined all the data into a full model: experimental condition (stress and nonstress), insertion genotype (presence/absence of the insertion) and interaction between these two factors. For those experiments in which more than one replicate was performed, the replicate effect was also taken into account. Because our dependent variable was a proportion, we used the arcsine transformation of the data before performing statistical analysis. We tested whether the data was normally distributed using Kolmogorov-Smirnov test. When the data was not normally distributed after the arcsine transformation, we applied the rank transformation. When the statistical test was significant, we estimated partial eta-squared values as a measure of the effect size (0.01 small effect, 0.06 medium effect, and 0.14 large effect).
We estimated the number of segregating sites (S), Tajima´s D, iHS, nSL and XP-EHH in the 2 kb region flanking the FBti0019985 insertion (chromosome 2R: 5758000–5760000) in 10 DGRP strains containing this insertion, in the 23 DGRP strains containing one of the roo insertions described in this work, and in the 15 strains that do not contain any insertion in the promoter region of CG18446. Note that the coordinates of FBti0019985 in the r5 of the D. melanogaster genome used by the DGRP project to generate the vcf files are 2R: 5,758,595–5,759,028. S and Tajima´s D are standard mesures of neutrality. iHS and nSL tests identify hard sweeps although they have some power to detect soft sweeps as well [69, 70]. H12 tests for positive selection on new variation and standing genetic variation within a population, that is, it searches both for soft and hard sweeps in a population [71]. Finally, XP-EHH is a statistical test of positive selection in one population that uses between populations comparisons to increase power in regions near fixation in the selected population [72].
We have used vcftools to calculate the number of segregating sites, and Tajima´s D using parameters –maf 1/(2n), where n is the sample size, and –remove-indels. We have obtained iHS, nSL, and XP-EHH using the selscan software with default parameters [73]. Finally, we have calculated H12 with ad hoc scripts. The four latter statistics require phased data. Thus, chromosome 2R of the 205 DGRP strains were phased together using ShapeIt [74].
To calculate the significance for the number of segregating sites, we resampled at random the same number of strains from the 205 DGRP strains available and calculated the distribution of segregating sites in the same 2 kb region. To calculate the significance of Tajima´s D, iHS, nSL and XP-EHH, we have used the empirical distributions of these statistics obtained from chromosome 2R.
|
10.1371/journal.ppat.1005678 | Bacillus anthracis Spore Surface Protein BclA Mediates Complement Factor H Binding to Spores and Promotes Spore Persistence | Spores of Bacillus anthracis, the causative agent of anthrax, are known to persist in the host lungs for prolonged periods of time, however the underlying mechanism is poorly understood. In this study, we demonstrated that BclA, a major surface protein of B. anthracis spores, mediated direct binding of complement factor H (CFH) to spores. The surface bound CFH retained its regulatory cofactor activity resulting in C3 degradation and inhibition of downstream complement activation. By comparing results from wild type C57BL/6 mice and complement deficient mice, we further showed that BclA significantly contributed to spore persistence in the mouse lungs and dampened antibody responses to spores in a complement C3-dependent manner. In addition, prior exposure to BclA deletion spores (ΔbclA) provided significant protection against lethal challenges by B. anthracis, whereas the isogenic parent spores did not, indicating that BclA may also impair protective immunity. These results describe for the first time an immune inhibition mechanism of B. anthracis mediated by BclA and CFH that promotes spore persistence in vivo. The findings also suggested an important role of complement in persistent infections and thus have broad implications.
| We discovered an immune modulatory mechanism of Bacillus anthracis mediated by the spore surface protein BclA. We showed for the first time that BclA mediated the binding of complement factor H, a major negative regulator of complement, to the surface of spores. The binding led to the down-regulation of complement activities in vitro and in an animal model. Using mice deficient in complement components, we further showed that BclA promoted spore persistence in the mouse lungs and impaired antibody responses against spores in a complement-dependent manner. We further provided evidence suggesting a role of BclA in the development of protective immunity against lethal B. anthracis challenges. These findings draw attention to a previously understudied aspect of the complement system. They suggest that in addition to conferring resistance to complement-mediated killing and phagocytosis, complement inhibition by pathogens have long-term consequences with respect to persistent infections and development of protective immunity. Considering a growing list of microbial pathogens capable of modulating complement activities, our findings have broad implications.
| Persistent colonization of the host by microbial pathogens can cause chronic infections, which are often difficult to treat with conventional antibiotics. It is recognized that persistent infection is a unique phase often involving specific virulence factors and pathogenic mechanisms [1]. Identifying and understanding these persistent mechanisms is key to developing new strategies to more effectively combat chronic infections.
Bacillus anthracis is a spore forming, Gram-positive bacterium that causes anthrax. Infections are initiated by entry of spores into the host via the respiratory system, the gastrointestinal tract, or cuts/wounds in the skin. Among the three forms of anthrax infections, inhalational anthrax has the highest mortality rate. One of the characteristic features of inhalational anthrax is the ability of spores to persist in the host lungs for prolonged periods of time [2–7]. Viable spores can be recovered from the lungs of exposed animals including non-human primates weeks or even months after the initial exposure. In addition, incubation periods of up to 43 days have been observed in humans [6]. This led to the 60-day antibiotic regimen recommended by the Centers for Disease Control and Prevention for people with pulmonary exposure to B. anthracis spores [7].
The mechanism underlying B. anthracis spore persistence is poorly understood. Mechanisms used by other bacterial pathogens for persistent infections include biofilm formation [8–12], residing in intracellular niches [13–15], suppression of innate and adaptive immune responses [13, 16–18], and changes in bacterial physiology and metabolism that favor persistent colonization [19–21]. B. anthracis spores are metabolically inactive and resistant to microbicidal effectors present in vivo. It was originally thought that the dormancy and resilience of spores were responsible for their ability to persist in the host. However, in a mouse model for spore persistence, B. anthracis spores were found to be significantly better at persisting in the lungs than Bacillus subtilis spores, suggesting the existence of persistence-promoting mechanism(s) beyond spore dormancy and resilience [4]. B. anthracis spores were also observed to be distributed throughout the lungs as single spores with the majority being extracellularly located [4], suggesting that biofilm formation or hiding in an intracellular niche is unlikely to be the major underlying mechanism. It is known that pulmonary exposure to B. anthracis spores does not elicit robust inflammatory immune responses in the lungs. Although the spore surface lacks typical pathogen-associated molecular patterns such as lipopolysaccharides, lipotechoic acid, and flagellin [22], spores have been shown to be capable of activating Toll-like receptor 2 and MyD88-dependent signaling [23], triggering inflammatory cytokine production [24, 25], and activating natural killer cells [26, 27]. Therefore the subdued immune response is likely due to an active immune evasion/suppression mechanism rather than a passive inactivity of the spores. The anthrax toxins are known to inhibit host immune responses. However, spores of a B. anthracis strain devoid of the anthrax toxins persisted as well as the parent toxin-producing strain [4]. This speaks against the possibility that low levels of anthrax toxins produced by a small amount of germinated spores in vivo may inhibit the overall immune response in the lungs and contribute to spore persistence. These observations provide support for a spore-mediated mechanism of immune suppression that has yet to be identified.
Bacillus collagen-like protein of anthracis (BclA) is the most abundant protein on the exosporium, the outermost layer of B. anthracis spores. It is the structural component of the hair-like nap on the exosporium [28]. Because of this spatial localization, BclA sits at the forefront with respect to interactions with host factors upon entry into the host. A number of studies have shown that BclA mediates spore uptake by macrophages and epithelial cells in both complement-dependent and–independent manners [29–33]. However despite its abundance, localization and interactions with host cells, the precise role of BclA in B. anthracis pathogenesis remains unclear. In animal models of acute anthrax infections BclA did not appear to contribute to virulence [29, 34].
In this study, the ability of BclA to manipulate the complement system and its role in spore survival and persistence in vivo was investigated. We found that BclA mediated the recruitment of complement factor H (CFH), the major inhibitor of the alternative pathway, to the spore surface where it facilitated C3 degradation; thereby inhibiting downstream complement activation. We further showed that BclA significantly promoted spore persistence in the mouse lungs and dampened antibody responses to spores in a complement-dependent manner. Finally we showed that BclA impaired protective immunity against lethal B. anthracis challenges. These findings have important implications in B. anthracis pathogenesis, bacterial manipulation of complement and persistent infections in general.
Spores of B. anthracis Sterne strain 7702 and the isogenic BclA deletion mutant (ΔbclA) were incubated with purified human CFH. Spore-CFH interaction was analyzed using flow cytometry (Fig 1A), solid phase binding assays (Fig 1B) and spore pull down assays (Fig 1C). In all three different assays, deletion of BclA led to significantly reduced CFH binding compared to 7702 spores. Complementation of the deletion with the full-length bclA gene (ΔbclA/BclA) restored CFH binding (Fig 1A–1C). Surface expression of BclA in the complemented strain was confirmed by immunofluorescence microscopy and flow cytometry (S1 Fig).
We next investigated if BclA could mediate recruitment of CFH from human and mouse serum, and mouse bronchial alveolar lavage (BAL) fluids. In order to distinguish between direct CFH binding and indirect binding through C3 fragments deposited on the spore surface, the binding assays were performed using heat-treated serum and BAL fluids so that the complement system was inactivated while CFH remained functional [35]. 7702 spores were able to recruit more CFH from normal human serum (NHS), mouse serum and mouse BAL fluids, compared to ΔbclA and B. subtilis spores, respectively (S2 Fig).
To further determine if BclA was sufficient to mediate CFH binding to spores, we expressed BclA on the surface of B. subtilis spores, which do not contain any BclA-encoding genes. Surface expression was verified by immunofluorescence microscopy and flow cytometry (S1 Fig). We observed that expression of BclA significantly enhanced the binding of purified CFH and CFH in human serum, mouse serum and mouse BAL fluids to B. subtilis spores (Fig 1A–1C and S2 Fig). BclA was further expressed as a His-tag recombinant protein (rBclA). Results from ELISAs showed that rBclA bound to CFH in a concentration-dependent and saturable manner, with an apparent KD of 0.91±0.45 μM (Fig 1D). Taken together, the results described above indicated that B. anthracis spore surface protein BclA mediated direct binding of human and mouse CFH to spores.
One of the principal functions of CFH is to act as a co-factor for complement factor I (CFI) to cleave C3b to the inactive iC3b, which disrupts the formation of the alternative complement pathway (ACP) C3 convertase. We first investigated the effect of BclA-mediated CFH recruitment on C3b cleavage to iC3b on the spore surface using purified complement components C3b, CFI and CFH. The results showed that the iC3b/C3b ratio on ΔbclA spores was significantly lower than that on 7702 and ΔbclA/BclA spores (Fig 2A and 2B). We further incubated the different spores with NHS for various length of time. The rate of iC3b accumulation on ΔbclA spores was significantly slower compared to that on 7702 and ΔbclA/BclA spores (Fig 2C and 2D). These results indicated that BclA-mediated CFH recruitment significantly promoted the cleavage of C3b to iC3b on the spore surface.
The increased cleavage of C3b to iC3b in the presence of BclA could potentially reduce the available C3b necessary for efficient C3 convertase formation, thereby reducing further C3 activation. We therefore determined if BclA-mediated CFH recruitment affected C3a production in NHS incubated with the different spores. The results showed that C3a concentration was significantly higher in samples incubated with ΔbclA spores compared to those incubated with 7702 or ΔbclA/BclA spores (Fig 2E, no antibody), suggesting that C3 cleavage was inhibited in the presence of BclA-expressing spores. To further determine whether the inhibition was due to CFH, we tested the effect of a CFH functional blocking antibody (OX24) [36]. Pre-treatment of NHS with OX24 increased the C3a concentration in samples incubated with 7702 or ΔbclA/BclA spores to a similar level as that seen in those with ΔbclA spores; whereas pre-treatment with the isotype control antibody (mouse IgG1) showed a similar pattern as that seen in the no antibody control (Fig 2E). Taken together, these results suggested that BclA-mediated CFH recruitment significantly reduced further activation of C3.
Cleavage of C3b to iC3b prevents the formation of C5 convertase complexes that cleave C5 to C5a and C5b and the downstream formation of the membrane attack complex. Therefore, we next investigated the effect of BclA-mediated CFH recruitment on downstream complement activation. We first performed an indirect complement hemolytic activity assay to measure terminal stage complement activation [37]. NHS was preincubated with 7702, ΔbclA or ΔbclA/BclA spores and centrifuged. The supernatants were used as the source of complement for hemolysis assays using opsonized sheep erythrocytes (EA-SRBC) as the target. If BclA led to inhibition of downstream complement activation, 7702 or ΔbclA/BclA pre-incubated serum should contain more intact complement components than ΔbclA pre-incubated serum, and thus cause more hemolysis. We observed ~ 100% hemolytic killing of EA-SRBC in sera pre-incubated with 7702 and ΔbclA/BclA spores respectively, but only 20% in serum pre-incubated with ΔbclA spores (p < 0.0001) (Fig 3A).
We further measured the level of C5a in serum incubated with the different spores as a direct method to evaluate downstream complement activation. The results showed that the level of C5a was significantly higher in samples incubated with ΔbclA spores compared to those incubated with 7702 or ΔbclA/BclA spores (Fig 3B, no antibody). To determine if the inhibition was due to CFH, C5a assays were performed using OX24 or control antibody pre-treated serum. The results showed that pre-treatment with OX24 increased the C5a concentration in samples incubated with 7702 and ΔbclA/BclA spores, respectively, to a similar level as that seen in those incubated with ΔbclA spores (Fig 3B). In contrast, mouse IgG1 had no effect on the level of C5a in any of the samples. These results indicated that CFH was responsible for the apparent effect of BclA on C5a.
We next investigated the effect of BclA on C5a level in vivo. Mice were intranasally (i.n.) inoculated with the different spores. BAL fluids were then collected by lavaging the lungs with sterile PBS containing EDTA, which stops complement activation. We observed that C5a concentration in the BAL fluids from mice infected with 7702 or ΔbclA/BclA spores was significantly lower than that from mice infected with ΔbclA spores (Fig 3C). Taken together, the results described above indicated that BclA-CFH interaction led to reduced C5 cleavage both in vitro and in vivo.
We first investigated if BclA was important for spore persistence. C57BL/6 mice were i.n. inoculated with sub-lethal doses of spores. Total bacteria and spore load in the lungs at two and four weeks post inoculation was determined. At both time points, C57BL/6 mice inoculated with 7702 spores harbored significantly more total bacteria and spores in the lungs than those inoculated with ΔbclA spores (Fig 4A and S5A Fig). Complementation of BclA in the ΔbclA background significantly increased the spore counts in the lungs at both time points. We tested the germination efficiency of 7702, ΔbclA and ΔbclA/BclA spores in three different media: a chemically defined germination media, LB and 100% NHS (S3 Fig). We did not observe any difference in the germination efficiency between the spores in any of the media. We also tested bacterial dissemination to distal organs such as the spleen and found no significant difference in bacterial burden in the spleen of mice inoculated with the different spores (S4A Fig). Hematoxylin and Eosin (H&E) staining was performed on lung sections from C57BL/6 mice collected at two weeks post i.n. inoculation with either 7702 or ΔbclA spores. Minimum pathology was observed in sections from both groups (S7 Fig), consistent with a previous report [4]. The alveolar and small airway epithelium appeared intact in both groups and lymphocyte infiltration was only occasionally observed. Overall, we did not see obvious differences in inflammatory responses in the lungs between the two groups.
To determine if BclA alone was sufficient to promote spore persistence in the lungs, we further examined B. subtilis spores expressing BclA. The results showed that expression of BclA on the surface of B. subtilis spores significantly increased both total bacteria and spore burden in the lungs compared to the vector control (Fig 4B). Together, these results suggested that BclA significantly promoted spore persistence in vivo.
We next investigated the role of complement in spore persistence. CFH deficiency in mice caused uncontrolled complement activation resulting in C3 consumption [38]. Therefore, we compared spore persistence in C3-/- mice, as C3 is where the three complement pathways converge. We reasoned that if BclA-mediated inhibition of complement was responsible for the increased spore persistence in the lungs, we should see no difference in spore persistence between 7702 and ΔbclA in C3-/- mice. Indeed, no significant difference in either total bacteria or spore burden was observed in C3-/- mice between 7702 and ΔbclA-infected groups at 2 or 4 weeks post inoculation (Fig 4C and S5B Fig), suggesting that BclA-mediated promotion of spore persistence was C3-dependent. It was previously reported that BclA bound complement component C1q. The binding leads to internalization of spores by epithelial cells through integrin α2β1 and opsonophagocytosis of spores by macrophages [30, 32]. We examined spore persistence in C1q-deficient (C1q-/-) mice. The results from C1q-/- mice mirrored those from wild type C57BL/6 mice (S6A Fig), suggesting that BclA-C1q interaction was not important for spore persistence in the mouse lungs. Taken together, the results suggested that spore persistence was promoted by BclA-mediated inhibition of complement activation.
To test if BclA contributes to virulence in acute infections, C57BL/6 and C3-/- mice were injected with lethal doses of 7702 or ΔbclA spores by intraperitoneal injection (i.p.). No significant difference in mouse survival was observed between 7702 and ΔbclA-infected groups in either mouse strains (Fig 4D and 4E). We next compared the total bacteria and spore burden in the lungs and spleen of C57BL/6 and C3-/- mice 48 hours post inoculation. We did not observe any significant difference in total bacteria or spore load in the lungs or the spleen between mice challenged with 7702 and those with ΔbclA spores (Fig 4F and 4G). These results indicate that BclA does not contribute to virulence in this lethal challenge model. This is consistent with results from previous studies using lethal infection models [29, 34]
The complement system not only shapes the innate immune responses, but also guides the adaptive immune responses [39–43]. We examined the effect of BclA on host antibody responses against spores in the persistence model. Anti-spore IgG antibodies in serum from infected mice were detected using ELISA. Both 7702 and ΔbclA spores elicited specific antibody responses in C57BL/6 (Fig 5A), C3-/- (Fig 5B) and C1q-/- (S6B Fig) mice, respectively, compared to the saline control. However, C57BL/6 and C1q-/- mice exposed to 7702 spores had significantly lower antibody titers compared to those exposed to ΔbclA spores (Fig 5A and S6B Fig), suggesting that BclA dampened antibody responses and that BclA-C1q interaction was not important in this process. The difference in anti-spore IgG titers between the 7702- and ΔbclA-infected groups was not detectable in C3-/- mice (Fig 5B), suggesting that BclA dampened antibody responses against spores through downregulating C3 activation.
Because of the significant difference in anti-spore antibody levels, we investigated if prior exposure to 7702 or ΔbclA spores triggered different protection against lethal B. anthracis challenges. In one set of experiment, C57BL/6 mice were i.n. inoculated with a sub-lethal dose of 7702 or ΔbclA spores and then challenged with a lethal dose of 7702 spores by intraperitoneal (i.p.) injection two weeks later. In another set, C57BL/6 mice were i.n. inoculated with sub-lethal doses of 7702 or ΔbclA spores at 0 and 2 weeks and then challenged with a lethal dose of 7702 spores by i.p. injection at four weeks. The results showed that for mice with one prior exposure (Fig 6A), those pre-exposed to 7702 spores succumbed to lethal challenges within two days, similar to those pre-exposed to saline only, whereas those pre-exposed to ΔbclA spores had a significantly better survival rate (p = 0.0308 vs. the saline control) with a median survival time of 4 days. For mice with two prior exposures (Fig 6B), the difference was even more pronounced (p = 0.0002 vs. the control group, p = 0.0069 vs. 7702 pre-exposed group). Taken together, these results suggested that BclA impaired protective immunity against lethal B. anthracis infections.
In this study we discovered a novel function for the major B. anthracis spore surface protein BclA. We demonstrated that BclA mediated recruitment of CFH to spores, facilitated C3b degradation on the spore surface, inhibited further C3 activation, and reduced C5 cleavage both in vitro and in vivo. We further showed that BclA promoted spore persistence in the host lungs and inhibited antibody responses against spores in a C3-dependent manner. Furthermore, BclA impaired protective immunity against lethal B. anthracis challenges. These results describe for the first time a spore-mediated immune modulatory mechanism through inhibition of complement. The results also suggested an important role of complement in persistent infections, an aspect of pathogen-complement interaction that is poorly understood.
The ability of BclA to mediate CFH binding was demonstrated by 1) ΔbclA spores bound significantly less CFH than the parent spores, and the defect was restored by complementing BclA, 2) BclA expressed on the surface of B. subtilis spores was sufficient to promote CFH binding, and 3) recombinant BclA protein bound to purified CFH in a concentration-dependent manner. We observed weaker CFH binding by ΔbclA and B. subtilis control spores. It is possible that there is another unknown low-affinity CFH binding protein on these spores or non-specific binding of CFH to spores. Our results also suggest that recognition of CFH by BclA is not human specific, i.e., BclA can bind both human and murine CFH, unlike some other CFH binding proteins such as the CFH-binding protein (fHbp) of Neisseria meningitidis [44] and PspC of Streptococcus pneumoniae [45]. A group A streptococcal collagen-like protein (Scl1) was reported to bind CFH via the C-terminal variable region of Scl1 [46]. While BclA is also a collagen-like protein, sequence comparison indicated no significant sequence similarities between the two proteins beyond the GXY triplet-repeating motif. BclA also did not show any significant sequence similarities to other reported microbial CFH binding proteins. Thus BclA is a novel CFH binding protein. BclA-bound CFH retained its co-factor activity, as shown by increased C3b degradation on the surface of parent and complemented spores compared with ΔbclA spores. BclA-CFH interaction inhibited further C3 activation, and decreased C5 activation as shown by C5a ELISA and hemolytic assays. The finding that C5 cleavage was also reduced in the mouse lungs in the presence of BclA further suggested that this effect was relevant in vivo. In addition, CFH functional blocking antibodies completely abolished the complement inhibitory activity of BclA, suggesting that the BclA-CFH interaction was responsible for this activity.
It has been known for decades that B. anthracis spores were able to persist in the host lungs for prolonged periods of time. This capability was thought to be due to the dormancy and resilience of spores. The results from this study describe for the first time a specific persistence-promoting mechanism mediated by the spore surface protein BclA. The observation that the difference in spore load in the lungs between 7702 and ΔbclA-infected mice disappeared in C3-/- mice suggests BclA promotes spore persistence in the lungs by inhibiting complement activities. It was previously reported that BclA directly binds C1q and this interaction leads to activation of the classical complement pathway and opsonophagocytosis of spores by macrophages [30, 32]. However, the results obtained from C1q-/- mice suggest that BclA-C1q interaction is not important for spore persistence or antibody response to spores. This suggests that in this model system, inhibition of the alternative pathway plays a dominant role in promoting spore persistence. Recently it was shown that binding of CFH to the PspC protein of S. pneumoniae promoted pneumococcal nasal colonization by CFH-mediated bacterial adherence to the epithelium [47]. In our case, CFH is present in C3-/- mice, suggesting that BclA-mediated promotion of spore persistence ultimately depends on C3 and works by inhibiting complement activities. Thus the persistence colonization mechanism described here is distinct from that of S. pneumoniae. The percentage of bacteria recovered from the lungs at 2 weeks post inoculation versus the initial inoculum was ~ 0.05%. This is in the same range as reported previously in Balb/c mice (~ 0.08%) [4].
The role of BclA in pathogenesis has been controversial despite the fact that it is a dominant protein on the spore surface. Studies in lethal infection models did not show any contribution of BclA to virulence [29, 34]. In the lethal spore challenge model here, our results also show no difference in either mouse survival or bacterial burden between mice challenged with 7702 and ΔbclA spores, consistent with previous studies. The findings here suggest that the primary role of BclA in vivo may be to promote the long term survival of spores through inhibition of complement activities.
We observed that 7702 spores led to significantly lower anti-spore antibody levels compared to ΔbclA spores in C57BL/6 and C1q-/- mice. The reduced antibody response was not due to a lower bacterial burden of 7702 in vivo; on the contrary 7702 infected mice had a higher spore burden in the lungs and a similar burden in the spleen compared to ΔbclA-infected mice. The fact that there was no difference in antibody responses in C3-/- mice suggested that BclA-mediated inhibition of C3 and/or downstream complement activation was responsible for the reduced antibody response to spores. The complement system influences B cells, T cells and antigen-presenting cells, the major cell types in the adaptive immune system [48–57]. The interaction between BclA and CFH can potentially affect all these components of the adaptive immune system. It has also been reported that CFH binding led to impaired antibody responses against the corresponding CFH-binding protein [47, 58–61]. Antibodies tend to recognize epitopes outside the CFH binding sites hence do not block CFH binding, or even enhance CFH binding. It would be interesting to investigate how BclA-CFH interaction affects antibody responses against B. anthracis spores.
Finally we observed that pre-exposure to 7702 spores conferred virtually no protection against lethal challenges whereas pre-exposure to ΔbclA spores provided significant protection in our infection model. The finding that BclA not only inhibited antibody responses against spores but also impaired protective immunity against B. anthracis lethal challenges has important implications in anthrax vaccine development and in persistent infections in general. With respect to vaccine development, BclA has been pursued as a vaccine candidate together with protective antigen (PA) as a multicomponent anthrax vaccine. Vaccination with BclA either as a recombinant protein or as a DNA vaccine augmented the protective efficacy of PA [62–64]. However, vaccination with formalin killed spores showed that ΔbclA spores provided greater protection than BclA-producing spores [65]. Our findings here suggest that the latter observation may be due to the effect of BclA on complement. The fHbp of N. meningitidis was approved as a component in multicomponent vaccines against serogroup B meningococcus [66–68]. Recent studies found that fHbp mutant proteins defective in CFH binding were more immunogenic and elicited stronger protective antibody responses than wild type proteins [59, 60, 69]. This raised the possibility that perhaps BclA mutants defective in CFH binding may offer better protection against anthrax infections.
Previous studies on the effect of pathogen manipulation of complement have been primarily focused on the more immediate effects of complement such as complement-mediated killing and opsonophagocytosis in the context of acute infections. For those bacteria that are susceptible to complement-mediated killing such as Gram-negative pathogens (e.g., N. meningitidis) or spirochetes (e.g., Borrelia burgforderi), inhibition of complement activation by recruiting CFH or other mechanisms confers serum resistance to the bacteria and is important for bacterial survival and virulence in vivo [66, 70–72]. For Gram-positive bacteria which are relatively resistant to serum killing due to their thick peptidoglycan cell wall, inhibition of complement activation can hinder phagocytosis and protects bacteria from phagocytic clearance. For example, Streptococcus pyogenes was found to inhibit phagocytosis by inactivating C3b in a strain-dependent mechanism [73]. Binding to CFH or C4-binding protein by S. pyogenes led to increased mortality in mouse models [74]. In contrast, the long-term effect of complement inhibition by pathogens has not been well studied in a systematic manner. The results presented here suggest that inhibition of complement by pathogens can play an important role in promoting persistent infections. In addition, because spores are resistant to lysis by complement and to phagocytic killing [30, 75], the role of complement observed in this study was likely due to the indirect activities of the complement effectors on the innate and adaptive immune system. This finding is particularly relevant to Gram-positive, encapsulated, or spore-forming pathogens, which tend to be relatively resistant to complement-mediated or phagocytic killing. With respect to how inhibition of complement promotes spore persistence, there may be multiple mechanisms involving both the innate and adaptive immune systems. Decreased production of C3a and C5a can affect cytokine production and the activation status of phagocytes. CFH binding to spores may not only dampen antibody responses but also affect the specific antibodies produced, as found in N. meningitidis. In addition, T cell and/or B cell functions can be affected [48–57]. Further studies to elucidate the detailed mechanism underlying the role of complement in persistent infections will be important.
In conclusion, we characterized the first CFH-binding protein of B. anthracis and described for the first time a spore-mediated immune inhibition mechanism of B. anthracis. These results shed light on the role of BclA in vivo. In addition, our findings suggest that in addition to conferring resistance to complement-mediated killing and opsonophagocytosis, complement inhibition by pathogens have long-term consequences with respect to persistent infections and protective immunity. Considering a growing list of microbial pathogens capable of modulating complement activities [76–80], our findings have broad implications.
Strains and plasmids used in this study are listed in S1 Table. Spores of B. anthracis and B. subtilis were prepared by culturing in a PA broth or on LB agar plates as described [4, 30]. To inhibit spore germination, a germination inhibitor D-alanine (2.5 mM) was included in solutions for assays involving spores. Normal human serum (NHS), GVB0 buffer (Gelatin Veronal Buffer without Ca2+, Mg2+, 0.1% gelatin, 5 mM Veronal, 145 mM NaCl, 0.025% NaN3, pH 7.3), VBS++ buffer (5 mM Veronal, 145 mM NaCl, 0.025% NaN3, pH 7.3, 0.15 mM CaCl2 and 0.5 mM MgCl2), purified complement proteins and goat anti-human CFH, anti-human C1q and anti-human C3 antibodies, were from Complement Technology unless otherwise stated. Secondary antibodies were from Thermo Fisher Scientific unless otherwise stated. 2,2,2-Tribromoethanol (Avertin), bovine serum albumin (BSA), chicken ovalbumin (OVA), D-alanine, and L-alanine were purchased from Sigma. Heat inactivation of complement was carried out at 56°C for 30 min.
A DNA fragment containing the bclA gene and its upstream sequence (~ 1kb) was cloned into an E. coli—B. anthracis shuttle vector pUTE583 [81]. The construct was then introduced into ΔbclA by electroporation as described previously [82]. To express BclA on the surface of B. subtilis spores, a DNA fragment encoding amino acid residues 39–400 of BclA was fused to the C-terminus of CgeA, a protein on the outermost surface of B. subtilis spores [83]. The first 38 amino acid residues were omitted because this region was reported to be proteolytically cleaved before anchorage of BclA onto the B. anthracis spore surface [84]. CgeA-BclA fusion was cloned into pDG1662, which allows the ectopic integration at the non-essential amyE locus in the B. subtilis chromosome [85, 86]. Surface expression was evaluated by staining spores with anti-BclA antibodies and fluorescently labeled secondary antibodies followed by immunofluorescence microscopy or flow cytometry analysis, as described in S1 Text.
To detect CFH recruitment to the spore surface, ~ 5×107 spores were incubated at 37°C for 30 min in PBS containing 2.5 mM D-alanine and supplemented with one of the following: purified human CFH (10 μg/ml), bovine serum albumin (10 μg/ml), 10% (v/v) heat-inactivated NHS, 10% (v/v) heat-inactivated mouse serum (from C57BL/6), or heat-inactivated mouse BAL fluid (from C57BL/6). The spores were then washed three times with ice-cold PBS containing 2.5 mM D-alanine and resuspended in the same buffer. An aliquot of the spore suspension was used to titer the spores by dilution plating and the rest were frozen until ready for analysis. Equal amounts of spores were boiled in SDS-sample loading buffer and the supernatants were subjected to Western blot analysis using goat anti-human CFH (1:10000) or sheep anti-CFH antibody (1:2000, Abcam) followed by incubation with rabbit anti-goat antibody conjugated to horseradish peroxidase (HRP) (1:10000, Invitrogen) or HRP-conjugated rabbit anti-sheep IgG (1:10000, Invitrogen) for 1 hr.
To detect iC3b deposition, ~ 5×107 spores were incubated in PBS buffer containing 500 μg/ml C3b, 100 μg/ml CFH, 4 μg/ml CFI, 0.1% BSA, 1 mM MgCl2 and 2.5 mM D-alanine at 37°C for 10 min. Spores were washed three times with ice-cold PBS containing 2.5 mM D-alanine. Equal amounts of spores were subject to Western blot analysis following the procedure described above. C3 fragments were detected using goat-anti human C3 (1:10000) and rabbit anti-goat HRP (1:10000). Band intensities were quantified using Image J.
For CFH binding, ~ 5×107 spores were incubated in buffer containing 2.5 mM D-alanine and supplemented with either 25 μg/ml purified human CFH or 10% heat inactivated NHS at 37°C for indicated length of time. Spores were then washed and fixed with 2% paraformaldehyde for 20 min at room temperature. Bound CFH was detected using goat anti-human CFH (1:400, Santa Cruz) followed by donkey anti-goat PE (1:400, Santa Cruz). For iC3b deposition, spores were incubated in buffer containing 10% NHS and 2.5 mM D-alanine for indicated length of time. iC3b was detected using mouse monoclonal antibody to human iC3b (neoantigen) (1:400; Quidel) and donkey anti-mouse 647 (1:400; SantaCruz). Samples were analyzed in a two laser Accuri C6 analytical flow cytometer using forward and side scatter parameters to gate on at least 20,000 spores. The red laser was used to measure the mean fluorescence intensity (MFI) of PE-labeled samples and data were analyzed using CFlow Plus (Accuri Cytometers) and graphed using GraphPad Prism 6 analysis software.
Recombinant BclA (rBclA) was purified as described previously [32]. Microtiter 2HB plates were coated with 10 ug/ml purified CFH or ovalbumin (OVA) in HBS (20 mM HEPES and 50 mM NaCl, pH 7.4) overnight at 4°C. The wells were washed to remove unbound proteins by HBS with 0.05% Tween 20 (HBST), blocked in HBST with 1% OVA for 1 hr at room temperature, and incubated with increased concentrations (0.01, 0.1, 1, 5, 15 and 30 μM) of His-tagged rBclA in HBST for 2 hrs at room temperature. The wells were then washed three times with HBST and incubated with anti-His HRP (1:3000, Alpha Diagnostic Intl. Inc.) for 1 hr. Plates were developed with Sigmafast OPD and read at 450 nm. Apparent KD was determined by non-linear regression (GraphPad Prism 6).
For spore binding, purified human CFH was immobilized onto wells of 96-well plates, blocked and incubated with ~ 1×107 biotin-labeled spores suspended in the blocking buffer supplemented with 2.5mM D-alanine for 30 min at 37°C followed by washing and incubation with streptavidin-conjugated to HRP.
Approximately 5×107–1×108 spores were incubated in GVB0 buffer containing 20% NHS and 2.5 mM D-alanine at 37°C for 30–60 min. Complement activation was terminated by adding 50 mM EDTA. The samples were centrifuged to remove spores. C3a and C5a levels in the supernatants were determined using Human C3a ELISA kit (BD OptEIA™) and Human Complement Component C5a DuoSet (R&D), respectively. To determine the effect of CFH functional blocking antibody OX24, 20% NHS in GVB0 buffer was pre-incubated with OX24 (Pierce Antibody) or isotype control mouse IgG1 (Sigma) at 240 nM or 480 nM final concentration at 37°C for 30 min. The reaction mix was then incubated with spores as described above.
Spores were incubated in buffer containing 20% NHS and 2.5 mM D-alanine at 37°C for 60 min. After centrifugation, the supernatants were diluted (1:10) in VBS++ (Ca2+, Mg2+) and used as the source of complement in hemolytic assays with opsonized sheep erythrocytes (1×107 cells) following the instructions of the supplier (EA-SRBC, CompTech). %Lysis is calculated as OD540(test)−OD540(Blank)OD540(total lysis)−OD540(Blank) ×100.
C57BL/6 mice were i.n. inoculated with different spores (~1×108 spores/mouse) and BAL fluid was collected 6 hours later by lavaging the lungs with 1ml cold sterile PBS containing 50 mM EDTA. The lavage fluids were centrifuged to remove cells and bacteria. C5a level in the supernatants was measured using the Mouse Complement Component C5a DuoSet (R&D).
All animal procedures were performed according to protocols approved by the Institutional Animal Care and Use Committee, Texas A&M Health Science Center (TAMHSC). C57BL/6 (originally purchased from the Jackson Laboratories), C1q-/- [87] and C3-/- [88] mice were maintained at the animal facility at TAMHSC. Mice were euthanized by i.p. injection with an overdose of 2,2,2-Tribromoethanol (Avertin) followed by terminal bleed. Intranasal inoculation was performed as previously described [4]. Briefly, 6–12 week old mice were anesthetized with Avertin (0.3 mg/g body weight) and then inoculated with 20 μls of an indicated sub-lethal dose of spores. For lethal challenge experiments, mice were inoculated with a lethal dose of spores by i.p. injection. Mice were monitored for survival and other symptoms daily. Both male and female mice were used in the experiments in a sex-matched manner.
Lungs and spleens were homogenized in 1 ml sterile ice-cold PBS containing 2.5 mM D-alanine, and either directly dilution plated to determine the total bacterial counts or heated at 68°C for 60 min and dilution plated to determine the spore counts. Lungs were also fixed for histological evaluation as described in S1 Text.
Total exosporium proteins were extracted from 7702 spores as previously described [89] with slight modifications. Briefly, ~5×109 spores were resuspended in 200 μl of an extraction buffer (50 mM Tris-HCl, pH 7.4, 8 M urea, and 2% (v/v) 2-mercaptoethanol), heated for 20 min at 90°C, and centrifuged at 13,000× g for 10 min. The supernatant was then treated with 20% (v/v) ice-cold trichloroacetic acid for 30 min on ice and centrifuged at 13,000×g at 4°C for 25 min. The pellet was washed once with 1 ml ice-cold acetone, centrifuged at 7000 r.p.m for 2 min, and dissolved in 100 μl of a solution of 200 mM Tris-HCl (pH 7.4) and 0.1 M glycine.
Blood was collected either from the saphenous vein, or by terminal bleed from the posterior vena cava, at two weeks after mice were inoculated with spores. Blood was allowed to clot at room temperature for 45 min before centrifugation at 4000 r.p.m. at 4°C for 10 min. Serum was either stored immediately at -80°C or at 4°C with 0.1% sodium azide. Extracted spore antigens were immobilized onto 96-well plates at 0.5 μg/well. The plates were washed twice with PBS containing 0.1% Tween-20 (PBST), and blocked with PBST containing 3% BSA at 37°C for 1 hr. Serum samples were diluted (1:100 for serum from C57BL/6 and C1q-/-, and 1:2 for serum from C3-/- mice) with PBS containing 3% BSA and incubated at 37°C for 1 hr. The wells were washed three times with PBST. Bound IgG was detected using goat anti-mouse IgG conjugated with HRP (1:2500, Invitrogen).
Spore germination was evaluated as described in S1 Text.
Pairwise comparison was carried out using Student’s t test. Survival analysis was performed using the Log-rank test (GraphPad Prism 6).
All animal experiments were performed in accordance to procedures approved by the Institutional Animal Care and Use Committee at Texas A&M Health Science Center (IACUC# 2015-0361-IBT). The Texas A&M University Health Science Center—Institute of Biosciences and Technology is registered with the Office of Laboratory Animal Welfare per Assurance A4012-01. It is guided by the PHS Policy on Human Care and Use of Laboratory Animals (Policy), as well as all applicable provisions of the Animal Welfare Act. Mice were euthanized by intraperitoneal injection of overdosed Tribromethanol/Avertin followed by terminal bleed. Mice were anesthetized with Avertin before intranasal inoculation of spores. All efforts were made to minimize animal suffering.
|
10.1371/journal.pgen.1000901 | Distinct Roles of Hand2 in Initiating Polarity and Posterior Shh Expression during the Onset of Mouse Limb Bud Development | The polarization of nascent embryonic fields and the endowment of cells with organizer properties are key to initiation of vertebrate organogenesis. One such event is antero-posterior (AP) polarization of early limb buds and activation of morphogenetic Sonic Hedgehog (SHH) signaling in the posterior mesenchyme, which in turn promotes outgrowth and specifies the pentadactylous autopod. Inactivation of the Hand2 transcriptional regulator from the onset of mouse forelimb bud development disrupts establishment of posterior identity and Shh expression, which results in a skeletal phenotype identical to Shh deficient limb buds. In wild-type limb buds, Hand2 is part of the protein complexes containing Hoxd13, another essential regulator of Shh activation in limb buds. Chromatin immunoprecipitation shows that Hand2-containing chromatin complexes are bound to the far upstream cis-regulatory region (ZRS), which is specifically required for Shh expression in the limb bud. Cell-biochemical studies indicate that Hand2 and Hoxd13 can efficiently transactivate gene expression via the ZRS, while the Gli3 repressor isoform interferes with this positive transcriptional regulation. Indeed, analysis of mouse forelimb buds lacking both Hand2 and Gli3 reveals the complete absence of antero-posterior (AP) polarity along the entire proximo-distal axis and extreme digit polydactyly without AP identities. Our study uncovers essential components of the transcriptional machinery and key interactions that set-up limb bud asymmetry upstream of establishing the SHH signaling limb bud organizer.
| During early limb bud development, posterior mesenchymal cells are selected to express Sonic Hedgehog (Shh), which controls antero-posterior (AP) limb axis formation (axis from thumb to little finger). We generated a conditional loss-of-function Hand2 allele to inactivate Hand2 specifically in mouse limb buds. This genetic analysis reveals the pivotal role of Hand2 in setting up limb bud asymmetry as initiation of posterior identity and establishment of the Shh expression domain are completely disrupted in Hand2 deficient limb buds. The resulting loss of the ulna and digits mirror the skeletal malformations observed in Shh-deficient limbs. We show that Hand2 is part of the chromatin complexes that are bound to the cis-regulatory region that controls Shh expression specifically in limb buds. In addition, we show that Hand2 is part of a protein complex containing Hoxd13, which also participates in limb bud mesenchymal activation of Shh expression. Indeed, Hand2 and Hoxd13 stimulate ZRS–mediated transactivation in cells, while the Gli3 repressor form (Gli3R) interferes with this up-regulation. Interestingly, limb buds lacking both Hand2 and Gli3 lack AP asymmetry and are severely polydactylous. Molecular analysis reveals some of the key interactions and hierarchies that govern establishment of AP limb asymmetries upstream of SHH.
| An important step during the initiation of vertebrate organogenesis is the setting-up of morphogenetic signaling centers that coordinately control cell specification and proliferation. One paradigm model to study these processes is the developing limb bud and recent studies have revealed how morphogenetic Sonic hedgehog (SHH) signaling from the zone of polarizing activity (ZPA) and Fibroblast growth factor (FGF) signaling from the apical ectodermal ridge (AER) coordinate cell specification with proliferation along both major limb bud axes [1]. AER-FGF signaling mainly controls the establishment of the proximo-distal (PD) limb bud axis (sequence: stylopod-zeugopod-autopod) [2], while SHH signaling by the polarizing region controls antero-posterior (AP) axis formation (radius and ulna, thumb to little finger) [3],[4]. Cells receiving the SHH signal inhibit the constitutive processing of Gli3 to its repressor form (Gli3R) and upregulate the expression of the Gli1 transcriptional activator, which results in positive regulation of SHH target genes [5]–[7]. In limb buds of mouse embryos lacking Gli3, the expression of initially posteriorly restricted genes such as Hand2, 5′HoxD genes and the BMP antagonist Gremlin1 (Grem1) expands anteriorly from early stages onwards and an anterior ectopic Shh expression domain is established at late stages [8]. However, the resulting digit polydactyly arises in a SHH-independent manner, as limbs of embryos lacking both Shh and Gli3 are morphologically and molecularly identical to Gli3 deficient mouse embryos [9],[10]. These and other studies indicate that Gli3 acts initially up-stream of SHH signaling to restrict the expression of genes activated prior to Shh to the posterior limb bud [11] and that SHH-mediated inhibition of Gli3R production is subsequently required to enable distal progression of limb bud development [9].
The molecular interactions that polarize the nascent limb bud along its AP axis and activate SHH signaling in the posterior limb bud mesenchyme have only been partially identified. Previous studies implicated the basic helix-loop-helix (bHLH) transcription factor Hand2 (dHand) in these early determinative processes upstream of SHH signaling [1]. In particular, the development of fin and limb buds of Hand2 deficient mouse and zebrafish embryos arrests at an early stage and no Shh expression is detected [12],[13]. This early developmental arrest in conjunction with massive generalized apoptosis of Hand2 deficient mouse limb buds precluded an in depth analysis of the molecular circuits and signaling systems that control initiation and progression of limb bud development. Furthermore, transgene-mediated over-expression of Hand2 induces digit duplications in mouse limb buds [14]. The functional importance of Hand2 as a transcriptional regulator in these processes was further corroborated by an engineered mutation that inactivates the Hand2 DNA binding domain in mouse embryos, which results in limb bud defects resembling the Hand2 null phenotype [15]. Cell-biochemical analysis showed that Hand2 interacts with so-called Ebox DNA sequence elements most likely as a heterodimer with other bHLH transcription factors such as E12 [16],[17] and Twist1, which is also required for early limb bud development [18],[19].
Genetic analysis in mouse embryos showed that Gli3 is required to restrict Hand2 expression to the posterior limb bud mesenchyme as part of a mutually antagonistic interaction [11]. This interaction was proposed to pre-pattern the limb bud mesenchyme along its AP axis prior to activation of SHH signaling. However, the functional importance of this pre-patterning mechanism for normal progression of limb development remained unknown. Additional pathways are also required for establishment of the Shh expression domain in the posterior limb bud mesenchyme such as retinoic acid signaling from the flank and AER-FGF8 signaling [20],[21]. During the onset of limb bud development, the expression of the 5′ most members of the HoxD gene cluster is restricted to the posterior mesenchyme by Gli3 [22],[23]. During these early stages, the 5′HoxA and 5′HoxD transcriptional regulators are required to activate Shh expression in the posterior limb bud mesenchyme [24]–[26]. Consistent with this genetic analysis, the Hoxd10 and Hoxd13 proteins interact directly with the cis-regulatory region that controls Shh expression in limb buds [27]. This evolutionary conserved cis-regulatory region is called ZPA regulatory sequence (ZRS) and is located about 800 Kb up-stream of the Shh gene [28]. Genetic inactivation of the highly conserved core region of the ZRS (termed MFCS1) results in limb bud-specific loss of Shh expression and a Shh loss-of-function limb skeletal phenotype [29]. Interestingly, this limb bud specific cis-regulatory region is absent from vertebrate species that have lost their limbs during evolution [30]. Transgenic analysis in mouse embryos revealed that ZRS-LacZ transgenes recapitulate major aspects of Shh expression in limb buds [28]. However, this study did not reveal specific cis-regulatory elements or sub-regions within the ZRS that regulate transcription, but rather indicated that the entire ZRS is required for correct Shh expression. A recent study shows that the ZRS interacts directly with the Shh transcription unit in both the anterior and posterior limb bud mesenchyme [31]. However, the Shh locus loops out of its chromosomal territory only in the posterior mesenchyme, which results in initiation of transcription. The evolutionary conserved function of the ZRS is underscored by an ever increasing large number of point mutations that are scattered through large parts of ZRS region and cause congenital preaxial polydactylies (PPD) in humans and many other mammals [32]. In summary, these studies establish that the far upstream ZRS cis-regulatory region controls Shh expression in different tetrapod species and that point mutations cause PPD, while deletion of the central part of the ZRS results in limbless phenotypes.
We have generated a conditional Hand2 mouse loss-of-function allele and use it to study the requirement of Hand2 during limb bud initiation. Inactivation of Hand2 in the forelimb field mesenchyme using the Prx1-Cre transgenic mouse strain disrupts the development of posterior skeletal elements. Complete and early inactivation results in a limb skeletal phenotype identical to limbs lacking Shh. Indeed, establishment of the Shh expression domain in the posterior limb bud is disrupted and early molecular markers of posterior identity are lost, while anterior markers expand posteriorly. This reveals the early requirement of Hand2 for establishing posterior identity and activation of Shh expression. Using specific antibodies, we identify protein complexes containing both Hand2 and Hoxd13 transcriptional regulators in wild-type limb buds. Chromatin immunoprecipitation using Hand2 antibodies reveals the specific enrichment of the ZRS in comparison to adjacent non-ZRS DNA sequences in wild-type limb buds. Functional analysis of the DNA-protein interactions in cultured fibroblasts reveals that Hand2 and Hoxd13 transactivate expression of a ZRS-luciferase reporter construct, while this is partially inhibited by Gli3R, which has been previously shown to interact with 5′Hoxd proteins [33]. Indeed, mouse limb buds deficient for both Gli3 and Hand2 lack AP asymmetry along the entire PD limb axis and display severe digit polydactyly with complete loss of identities. Our study uncovers the interactions of Hand2 with the Gli3 and Hoxd13 transcriptional regulators and the far-upstream ZRS cis-regulatory region that are required to polarize the nascent limb bud mesenchyme and establish Shh expression in the posterior limb bud.
Mouse embryos lacking Hand2 die during mid-gestation due to cardiovascular defects and limb bud development arrests prior to formation of limb skeletal elements [12],[34]. Therefore, we generated a conditional Hand2 loss-of-function allele by inserting two loxP sites into the locus (“floxed” allele: Hand2f or H2f), which enables Cre-recombinase mediated deletion of the Hand2 transcription unit (Figure S1). Hand2 was inactivated in the limb bud mesenchyme (H2Δ ˜Δc; Δc: conditional inactivation of the Hand2f allele) using the Prx1-Cre transgene, which is expressed in the forelimb field mesenchyme from about E8.5 onwards (14 somites) [35],[36]. The inactivation of Hand2 was verified by monitoring the clearance of Hand2 transcripts and proteins in forelimb buds and mesenchymal cells (Figure 1A and Figure S2A, S2B, S2C). Limb bud specific inactivation of Hand2 (H2Δ ˜Δc; Figure 1A) causes distal truncations of the forelimb skeleton and loss of the autopod (Figure 1B). The skeletal phenotypes of Hand2 deficient forelimbs are variable, but the most severely affected cases (39% of all limbs, n = 80; Figure S3A, S3D) are identical to Shh deficient limbs (Figure 1B). Indeed, Shh expression and SHH signal transduction are lacking from a similar fraction of all H2Δ ˜Δc limb buds (Figure 1C and Figure S3C). Therefore, the most severely affected H2Δ ˜Δc limb buds correspond to the limb-specific complete Hand2 loss-of-function phenotype (Figure 1A–1C and Figure S3). Between two and four digits form in hypomorphic H2Δ ˜Δc limbs (Figure S3A, S3D) as a likely consequence of residual Hand2 expression, which triggers SHH signal transduction (Figure S3B, S3C).
In the most severely affected forelimb buds, cells along the entire PD axis, but in particular in the distal-anterior mesenchyme are eliminated by apoptosis (Figure 1D), which is distinct from the generalized apoptosis and developmental arrest of mouse embryos lacking Hand2 constitutively (Figure S1D, S1E) [12]. In H2Δ ˜Δc forelimb buds, cell death is limited to the core mesenchyme at embryonic day E10.0 (Figure 1D, white arrowhead). In contrast, no significant apoptosis is detected in forelimb buds of wild-type and Shh deficient limb buds at these early stages (Figure 1D, open arrowhead). Therefore, Hand2 is required for cell survival upstream of its role in activation of SHH signaling (Figure 1D, left panels). During progression of limb bud development, the apoptotic domain expands distal-anterior in H2Δ ˜Δc limb buds and becomes similar to the cell death domain observed in Shh deficient limb buds (Figure 1D, middle and right panels).
In mouse embryos, hindlimb development is delayed by ∼12 hrs and activation of the Prx1-Cre transgene in the posterior mesenchyme is delayed by ∼24 hrs in comparison to forelimb buds [35],[36]. The resulting ∼12 hrs delay in Hand2 inactivation at equivalent stages in the posterior hindlimb bud allows formation of an autopod with 4–5 digits, while the tarsal bones are always fused (Figure 2A). Furthermore, inactivation of Hand2 specifically in the distal forelimb bud mesenchyme from E10.5 onwards no longer alters skeletal development (data not shown). In agreement with the subtle skeletal alterations following Prx1-Cre-mediated Hand2 inactivation in hindlimb buds (Figure 2A) Shh remains expressed, albeit at slightly lower levels than in wild-types (Figure 2B). Taken together, these studies show that Hand2 is essential to establish Shh expression in the posterior mesenchyme during initiation of limb bud development. Subsequently, it contributes to transcriptional up-regulation of Shh expression.
Our further analysis focused on the most severe, complete Hand2 loss-of-function phenotypes in forelimb buds (Figure 1). The early essential requirement of Hand2 upstream of SHH in forelimb buds (for cell survival, Figure 1D) is further substantiated by molecular analysis, which reveals the lack of Tbx3 and Tbx2 expression [37] in the posterior mesenchyme of H2Δ ˜Δc forelimb buds. In contrast, their posterior expression is initiated but not up-regulated in ShhΔ ˜Δ forelimb buds (Figure 3A and 3B). The expression of 5′HoxD genes is activated but not propagated in Hand2 deficient limb buds (Figure S4A, S4B), likely due to the disruption of SHH signaling (Figure 1C). Concurrently, the expression of anterior genes such as Cry-μ, Alx4 and Gli3 is ectopically activated or expands to the posterior margin in H2Δ ˜Δc forelimb buds earlier and/or more prominently than in ShhΔ ˜Δ limb buds (Figure 3C–3E and Figure S4C). This loss of posterior and gain of anterior molecular markers reveal the early essential requirement of Hand2 for establishing posterior limb bud identity.
This analysis (Figure 1, Figure 2, Figure 3) led us to consider the possibility that Hand2 might directly transactivate Shh expression, possibly in conjunction with 5′Hox genes, which are essential for Shh activation in mouse limb buds [24],[26]. Chromatin immunoprecipitation (ChIP) studies showed previously that Hoxd13 containing chromatin complexes are bound to the far up-stream ZRS cis-regulatory region that controls Shh expression in limb buds [27]. In addition, Hoxd13 is able to transactivate a ZRS-luciferase reporter construct in transfected cells [27]. Therefore, the potential direct interactions of Hand2 with Hoxd13 proteins and the ZRS were assessed by luciferase transactivation assays in NIH3T3 cells, which are mouse fibroblasts commonly used to analyze the SHH pathway [38]. A luciferase reporter construct encoding the entire ZRS (ZRS-Luc) was generated by inserting the ∼1.7 kb mouse ZRS region (Figure 4A and Figure S5) [28] upstream of an adenovirus minimal promoter (for details see Text S1). The basal activity of this ZRS-Luc reporter construct was set to 1 and transfection of either Hand2 (∼3-fold) or Hoxd13 (∼6.5-fold) induced luciferase activity and their co-transfection resulted in an ∼10.5-fold increase (Figure 4B). In silico analysis revealed 6 bona fide Ebox sequence elements within the ZRS (Figure 4A and Figure S5). Inactivating point mutations in either individual or several of these Ebox elements reduce the activity of the ZRS, but not in a strictly Hand2-dependent manner as the transactivation by Hoxd13 alone is also affected (data not shown). As Hand2 and Gli3R act in a mutually antagonistic manner during initiation of limb bud development [11], the potential effects of Gli3R on transactivation were assessed. As neither the Gli3 nor Gli1 activator forms are able to activate the ZRS-Luc reporter on their own (data not shown), the ZRS likely lacks functional Gli binding sites [39], suggesting that any effects of Gli3R would be indirect. Indeed, co-expression of Gli3R results in significant inhibition of transactivation in the presence of Hoxd13 (Figure 4B), in agreement with the proposal that Gli3R can bind to and potentially antagonize Hoxd13 function [33]. In particular, Gli3R represses Hand2-Hoxd13 mediated transactivation of the ZRS-Luc reporter by ∼50% (Figure 4B).
The relevance of these interactions for limb bud development was determined by co-immunoprecipitation (Figure 4C and Figure S6) and ChIP analysis (Figure 4D and 4E). Immunoprecipitation of Hoxd13 proteins in combination with Western blotting reveals the existence of protein complexes containing both Hoxd13 and Hand2 protein in wild-type limb buds (Figure 4C). The likely direct nature of these interactions is supported by efficient co-precipitation of epitope-tagged Hand2 and Hoxd13 proteins from transfected cells (Figure S6). These experiments establish that Hand2 interacts directly with Hoxd13 but not with Gli3R (Figure S6), which is relevant with respect to their genetic interaction (see below). As the available polyclonal Hand2 antibodies specifically recognize and immunoprecipitate Hand2 proteins (Figure S2B, S2C, S2D), ChIP on wild-type mouse limb buds was performed [40] to enrich Hand2 containing chromatin complexes and the analysis of three independent, fresh chromatin preparations is shown in Figure 4D and 4E. Conventional PCR using the amplicon “c” (Figure 4A) detected this ZRS region in chromatin precipitated with anti-Hand2 antibodies (lanes α-H2, Figure 4D), while no such amplification was detected when non-specific IgGs were used (lanes α-IgG; Figure 4D). To further analyze this apparent association of Hand2 containing chromatin complexes with the ZRS, three amplicons (“b”, “c”, “d”) probing different regions of the ∼1.7 kb mouse ZRS (Figure 4A) were used for real-time PCR (Q-PCR) analysis. In addition, two amplicons located outside the mouse ZRS were chosen as likely negative controls (non-ZRS amplicons “a” and “e” in Figure 4A and 4E and Figure S5). Indeed, Q-PCR analysis revealed a minimally 14-fold enrichment of the amplicons located within the ZRS in comparison to the adjacent non-ZRS regions (Figure 4E). This enrichment is specific as ChIP using non-specific IgGs resulted in much lower Q-PCR amplification of all five regions. In particular, the enrichment of the ZRS in comparison to flanking non-ZRS regions is highly significant (amplicons “b” to “d” versus “a” and “e”; p = 0.0018), while the variability among the three ZRS amplicons is not significantly different. Interestingly, the ZRS region encompassing amplicon “b”, whose enrichment is most variable, does not encode any bona fide Ebox elements (Figure 4A and 4E). This provides additional evidence for the fact that the interaction of Hand2-containing chromatin complexes with the ZRS may not depend only on Ebox sequences. This ChIP analysis (Figure 4D and 4E) provides good evidence that the Hand2-containing chromatin complexes bind to the ZRS cis-regulatory region, but not to adjacent non-ZRS sequences.
As embryos lacking Hand2 in limb buds survive to advanced stages (Figure 1B), the functional relevance of the pre-patterning mechanism [11] can now be genetically investigated in Hand2 and Gli3 compound mutant (H2Δ/ΔcGli3Xt/Xt) embryos (Figure 5, Figure 6, Figure 7). In contrast to the Hand2 deficiency, H2Δ/ΔcGli3Xt/Xt limbs are severely polydactylous and display little phenotypic variability (Figure 5A and Figure S7A). In addition, the zeugopodal bones and elbow joints appear strikingly symmetrical (Figure 5A, white and black arrowheads in panel H2Δ ˜ΔcGli3Xt/Xt). These limb skeletal abnormalities are much more severe than the ones of Gli3Xt/Xt and ShhΔ ˜ΔGli3Xt/Xt limbs (Figure 4A, panel Gli3Xt/Xt; see also [9],[10]). While the skeletal elements of H2Δ ˜ΔcGli3Xt/Xt limbs seem to lack AP asymmetry, survival of the zeugopod and autopod progenitors is restored and the primordia are expanded in contrast to H2Δ ˜Δc limbs (Figure S7B and data not shown). Moreover, the Sox9 expression domain, which marks the pre-chondrogenic lineage [41], is expanded in H2Δ ˜ΔcGli3Xt/Xt limb buds that tend to be larger than normal (Figure 5B, panel H2Δ ˜ΔcGli3Xt/Xt). However, no significant changes in proliferation were observed in H2Δ ˜ΔcGli3Xt/Xt limb buds (data not shown). While the pre-chondrogenic condensations of all major skeletal elements are discernible by E10.75 in wild-type and Gli3 deficient limb buds, Sox9 expression remains diffuse and non-polarized in H2Δ ˜ΔcGli3Xt/Xt limb buds (Figure 5B). During autopod development, the pool of Sox9 expressing digit progenitors is significantly expanded in H2Δ ˜ΔcGli3Xt/Xt limb buds in comparison to Gli3 mutants and wild-types (Figure 5B; compare limb buds at E11.5). The apparent symmetry of in particular the zeugopod in the H2Δ ˜ΔcGli3Xt/Xt limbs contrasts with the normal AP asymmetry in Gli3Xt/Xt and ShhΔ ˜ΔGli3Xt/Xt limbs (Figure 5A) [9]. This observation indicates that Hand2 and Gli3 participate in establishment of the AP asymmetry of the proximal limb skeleton independent of SHH signaling. Indeed, the expression of Runx2, which marks proximal skeletal primordia [42], is altered in double mutant limb buds (Figure 5C). By E12.0, Runx2 is expressed in the presumptive stylopod and zeugopodal domains of wild-type limb buds, while few Runx2 positive cells are detected in Hand2 deficient limb buds (Figure 5C). In contrast, the Runx2 expression domain is expanded and lacks polarity in the proximal part of double mutant limb buds (Figure 5C, black arrowheads). Taken together, these results indicate that the skeletal phenotypes and the severe polydactyly of H2Δ ˜ΔcGli3Xt/Xt limbs arise as a consequence of disrupting AP asymmetry (proximally as indicated by Runx2) and aberrant expansion of the skeletal progenitor pools (distally as indicated by Sox9).
In H2Δ ˜ΔcGli3Xt/Xt limb buds, Shh expression is not detected by in situ hybridization (Figure 6A) and its expression is ∼10-fold lower than in wild-types (Figure 6C). Interestingly, the variability in Shh expression following Prx1-Cre mediated inactivation of Hand2 (Figure 1C, Figure S3B, S3C, S3D, and Figure 6C) is no longer observed in H2Δ ˜ΔcGli3Xt/Xt limb buds (Figure 6A and 6C), which agrees with the lack of significant variability in the resulting skeletal phenotypes (Figure 5A). This could be linked to the fact that posterior Shh expression is already reduced by ∼50% in Gli3Xt/Xt limb buds (Figure 6A and 6C). The low Shh transcript levels detected in the most severely affected H2Δ ˜Δc and H2Δ ˜ΔcGli3Xt/Xt limb buds (between 8% and 20%, Figure 6C) likely reflect basal expression not detected by in situ hybridization (Figure 1D, Figure 6A; see Discussion). BMP4-mediated up-regulation of its antagonist Grem1 in the posterior mesenchyme is essential to initiate the self-regulatory signaling system that promotes distal limb bud development [43],[44]. In H2Δ ˜Δc limb buds, Bmp4 expression appears not significantly altered, while its expression is slightly reduced in H2Δ ˜ΔcGli3Xt/Xt limb buds (panels Bmp4 in Figure 6B and 6C). In particular, the posterior expression domain in double mutant limb buds appears smaller (arrowheads, panels Bmp4 in Figure 6B), which results in rather symmetrical Bmp4 expression along the AP limb bud axis. Furthermore, Grem1 expression is activated, but not up-regulated and distal-anteriorly expanded in Hand2 deficient limb buds (panel Grem1 in Figure 6B), similar to Shh deficient limb buds [44]. In double mutant limb buds, the Grem1 expression domain appears symmetrical due to its anterior expansion. However, the rather variable Grem1 transcript levels are overall reduced in H2Δ ˜ΔcGli3Xt/Xt limb buds in comparison to wild-type and Gli3 deficient limb buds (panels Grem1 in Figure 6C). Finally, the expression of the direct BMP transcriptional target Msx2 [43] is expanded in H2Δ ˜Δc limb buds, while its expression is significantly reduced in Gli3 deficient and double mutant limb buds as a likely consequence of the alterations in Grem1 (panels Msx2 in Figure 6B and 6C). Taken together, these results corroborate the proposal that the initial phase of Grem1 expression in the posterior mesenchyme depends on BMP4 activity [43]. The rather symmetrical Grem1 expression in H2Δ ˜ΔcGli3Xt/Xt limb buds indicates that the second phase of SHH-dependent distal-anterior expansion of its expression in wild-type limb buds is a likely consequence of SHH-mediated inhibition of Gli3R activity [6].
The lack of discernible AP identities in the autopod of H2Δ ˜ΔcGli3Xt/Xt limb buds (Figure 7A) is confirmed by molecular analysis. In agreement with the rather symmetric distribution of Bmp4 and Grem1 in the distal limb bud mesenchyme (Figure 6B), Fgf4 is expressed uniformly by the AER in double mutant limb buds (Figure 7B). The distal expression domains of the Hoxd13 and Hoxa13 genes mark the presumptive autopod territory and are required for specification and expansion of the digit progenitors [45],[46]. Within the distal mesenchyme of H2Δ ˜ΔcGli3Xt/Xt forelimb buds, the expression of Hoxd13 is anteriorly expanded and appears apolar in comparison to wild-type and Gli3 mutant limb buds (Figure 7C; best seen in the apical views). In addition, the AP asymmetry of the distal Hoxa13 domain is also lost in double mutant limb buds (Figure 7D; best seen in the apical views). The expanded and apolar expression of these genes (Figure 7B–7D) together with the alterations in Sox9, Runx2 (Figure 5B and 5C), Bmp4 and Grem1 (Figure 6B) reveal the striking loss of the asymmetrical expression of molecular and cellular markers of the AP axis along the entire PD axis in limb buds lacking both Hand2 and Gli3.
In this study, we uncover the key regulatory interactions involving Hand2 that control establishment of posterior limb bud identity upstream of SHH signaling, in particular the genetic interactions with Gli3 that initiate AP axis polarity. Secondly, we reveal that Hand2, which like 5′Hox genes is essential for establishment of the Shh expressing limb bud organizer in the posterior-proximal mesenchyme, is part of the chromatin complexes bound to ZRS cis-regulatory region. The striking loss of posterior and gain of anterior molecular markers in Hand2 deficient limb buds indicates that limb field symmetry may normally be broken by Gli3R-mediated posterior restriction of Hand2 expression. This most likely parallels activation of 5′HoxD genes in the posterior mesenchyme [45]. In Hand2 deficient limb buds, the SHH dependent establishment of the late 5′HoxD expression domains is disrupted, while in limb buds lacking both Hand2 and Gli3, the late 5′HoxD expression domains expand uniformly throughout the distal autopod. Therefore, the down-regulation of 5′HoxD genes in Hand2 deficient limb buds is a likely consequence of increased Gli3R activity due to lack of SHH signaling [23]. Furthermore, Hand2 participates in transcriptional activation and/or upregulation of Tbx2/3 and Shh expression in the posterior mesenchyme and is required for anterior restriction of Gli3 and Alx4 expression. In Hand2 deficient limb buds, expression of the BMP antagonist Grem1 is activated in the posterior mesenchyme under the influence of BMP signaling (ref. 43 and this study). This previous analysis and the observed anterior expansion of Grem1 expression in H2Δ ˜ΔcGli3Xt/Xt limb buds reveals that the transcriptional activation and positioning of the Grem1 expression domain is controlled by interaction of BMP4 (positive) with GLI3R (negative). In wild-type limb buds, the Grem1 expression domain is always located distal-anterior to the Shh expressing cells and their descendents [47],[48], while it remains proximal and low due to the lack of SHH signaling in H2Δ ˜Δ limb buds (this study). Taken together, these results provide further insights into the molecular mechanism controlling spatial and temporal aspects of BMP4-mediated initiation and SHH-dependent progression of Grem1 expression, which acts as an essential node in the self-regulatory signaling system that controls limb development [1].
Our biochemical analysis of chromatin isolated from wild-type mouse limb buds reveals that Hand2-containing chromatin complexes are bound to the ZRS, which is the far upstream cis-regulatory region required for Shh expression in limb buds [28],[29]. In particular, ZRS sequences are specifically and significantly enriched in Hand2 containing chromatin complexes in contrast to flanking regions. Furthermore, Hand2 is part of Hoxd13 protein complexes in limb buds and in transfected cells, the two proteins transactivate the expression of a luciferase reporter gene in a ZRS-dependent manner. Albeit the fact that such transactivation studies are of somewhat artificial nature, the conclusions reached by this analysis completely agree with the results of our genetic analysis of Hand2 functions during mouse limb bud development. Early and complete genetic inactivation of Hand2 in limb buds disrupts establishment of the Shh expression domain in the posterior limb bud, while either incomplete or temporally delayed inactivation does no longer disrupt initiation of Shh expression (this study). This reveals the early essential requirement of Hand2 for establishment of the posterior Shh expression domain, while subsequently Hand2 appears to contribute to transcriptional up-regulation of Shh expression. This may happen as part of an auto-regulatory loop because SHH signaling in turn up-regulates Hand2 expression most likely via repressing production of the Gli3R isoform [9],[11],[49]. The low levels of Shh expression detected by Q-PCR even in the most affected H2Δ ˜Δc and H2Δ ˜ΔcGli3Xt/Xt limb buds, but not in Shh deficient limb buds (JDB and RZ, unpublished) are indicative of basal transcription of the Shh locus in the absence of Hand2, which is not detectable by in situ hybridization (this study). This basal expression may depend on Hox transcription factors [24],[26] or other regulators of Shh expression in limb buds (see below). However, our study shows that Hand2 is essential to establish and upregulate Shh expression in the posterior mesenchyme, which defines the SHH signaling limb bud organizer [1]. This Hand2-mediated transactivation of Shh expression is a likely consequence of its direct interaction with the ZRS cis-regulatory region and is possibly enhanced by formation of transcriptional complexes with Hoxd13 protein in limb buds.
Genetic and experimental manipulation of paired appendage buds in mouse, chicken and zebrafish embryos have begun to reveal the factors required in addition to Hand2 and 5′HoxD genes for Shh activation. In particular, AER-FGF and retinoic acid signaling have also been implicated in the activation of Shh expression [21],[50]. Deletion of both the HoxA and HoxD clusters in mouse embryos disrupts Shh activation and causes early arrest of limb bud development such that the limb skeleton is truncated at the level of the stylopod [24],[26]. But in contrast to Hand2, loss-of-function mutations in these genes alone or in combination do not phenocopy the Shh loss-of-function limb skeletal phenotypes [51],[52]. The Hand2 protein interacts with Hoxd13 and is part of the chromatin complexes bound to the ZRS in limb buds (this study). However, other transacting factors will likely contribute to ZRS dependent activation of Shh transcription. In fact, the overlap of the Hand2 and Hoxd13 expression domains in the posterior limb bud mesenchyme is much bigger than the initial Shh expression domain. During limb bud initiation stages, the Hand2 and Gli3 expression domains overlap significantly, but then become rapidly mutually exclusive [11]. Therefore, these early dynamic changes in the expression domains of the Hand2, Gli3 and Hoxd13 transcriptional regulators may well alter their interactions and spatially restrict the formation of transcription initiating/enhancing Hand2-Hoxd13 chromatin complexes at the ZRS to the posterior limb bud (this study). These direct interactions would restrict the up-regulation of Shh expression to the posterior limb bud mesenchyme, thereby establishing the SHH signaling limb bud organizer. A recent study shows that the distant ZRS is in close proximity to the Shh transcription unit in both the anterior and posterior limb bud mesenchyme, but only loops out of its chromosomal territory in the posterior mesenchyme [31]. Interestingly, Shh is apparently transcribed by only a fraction of all ZPA cells at one particular time point, which indicates that the chromosomal conformation dynamics control Shh expression at the cellular level [31].
It is known that Hand2 binds DNA primarily as a heterodimer with E12 and/or the bHLH transcription factor Twist1 [16],[19]. Interestingly, Twist1 is also required during early limb bud development [18] and point mutations in the human Twist1 gene alter its dimerization with Hand2, which causes congenital limb malformations [19]. Therefore, these additional factors may also participate in regulation of Shh expression. The expression of Hand2 and 5′HoxD genes is activated in parallel, but then they converge functionally on the ZRS to establish the Shh expression domain in the posterior limb bud (this study and ref. 24). Furthermore, the establishment of the posterior Tbx2 and Tbx3 expression domains is disrupted in Hand2 deficient limb buds. The cis-regulatory elements controlling their expression are currently unknown, but it has been shown that Tbx2 expression requires the overlying non-AER ectoderm [53]. Additional experimental and genetic evidence indicates that Tbx2 and Tbx3 act likely upstream of Shh to restrict its transcriptional activation to the posterior limb bud margin [53],[54]. In particular, ectopic expression of Tbx3 in early chicken limb buds induces an anterior shift of the entire limb bud together with transient anterior expansion of Hand2 expression [55]. These studies indicate that Tbx genes are part of the molecular circuits that position the limb bud, specify posterior identity and restrict activation of Shh to its posterior margin.
The genetic inactivation of the pre-patterning mechanism in H2Δ ˜ΔcGli3Xt/Xt limb buds disrupts establishment of AP asymmetry and self-regulatory limb bud signaling [43], while PD axis outgrowth and formation of all three major limb skeletal segments are the likely consequence of uniform AER-FGF signaling [2]. This results in a shortened and symmetric stylopod, zeugopod and a polydactylous autopod with highly dysmorphic digits. Similar to H2Δ ˜ΔcGli3Xt/Xt limb buds, limbs lacking 5′HoxD genes and Gli3 are also severely polydactylous but retain some polarity [56],[57]. Therefore, the loss of AP polarity along the entire proximo-distal axis is more severe than the phenotypes observed in limb buds lacking Gli3 alone or in combination with genes such as Shh, Alx4 or 5′HoxD genes [9], [56]–[58]. Over-expression of Hand2 in the entire limb bud mesenchyme results in a duplication of the anterior zeugopod (ulna) and posterior autopod (digits) [12], which indicates that disturbing the balance between Hand2 and Gli3 either by gene inactivation or over-expression alters AP polarity. Therefore, the balance of the opposing activities of Hand2 and Gli3R in concert with 5′HoxD genes may control specification of the AP limb axis independent and up-stream of SHH signaling. In mouse limb buds lacking the Plzf zinc finger protein, 5′HoxD genes are uniformly expressed from early stages onwards and AP polarity is partially lost in combination hindlimb digit polydactyly [59].
It remains unclear why the digit polydactyly in H2Δ ˜ΔcGli3Xt/Xt forelimbs is more severe than the one of Gli3Xt/Xt (and ShhΔ ˜ΔGli3Xt/Xt [9]) forelimbs. However, in H2Δ ˜ΔcGli3Xt/Xt forelimb buds, the distal expression domains of Hoxa13 and Hoxd13, which delineate the autopod territory and function in digit development (see [refs. 24],[26] for further detail) are anteriorly expanded in comparison to Gli3 deficient limb buds. Such anterior expansion may point to an enlarged pool of autopod/digit progenitors, which could underlie the more severe digit polydactyly. As discussed before, this expansion of the Hoxa/d13 expression domains and the presumptive autopod territory are a likely consequence of the early loss of AP polarity along the entire PD axis in double mutant forelimb buds in contrast to Gli3Xt/Xt mutants. In particular, the H2Δ ˜ΔcGli3Xt/Xt forelimb skeletons bear some resemblance to the primitive paired appendages of Devonian fish and the polydactylous limbs of early tetrapods [60]. We shows that these rather “primitive” limb structures develop in the absence of pre-patterning (Hand2, Gli3) and the self-regulatory signaling system that interlinks the SHH, BMP and FGF signaling pathways, which are both key to normal limb skeletal development [1]. During tetrapod evolution, the symmetry of primitive polydactylous autopods from the Devonian period [61] was likely broken by beginning to set-up the regulatory interactions described in this study as they initiate posterior polarity up-stream or in parallel to their requirement for establishment of the SHH signaling limb bud organizer. The establishment of these transcriptional regulatory network acting upstream of SHH signaling might have enabled the development of the more refined and better functional pentadactylous limbs of modern tetrapods.
All animal experiments were performed in accordance with Swiss law and have been approved by the regional veterinary and ethics authorities.
The generation of Hand2 conditional mutant mice is shown in Figure S1. Hand2 mouse strains were kept in a mixed 129SvJ/C57BL6 genetic background. For details of the generation and analysis of Hand2 mice and embryos see Text S1.
For IP, fore- and hind-limb buds from E11.0 embryos were collected in PBS and lysed in lysis buffer (Tris-HCl 10 mM pH 8.0; EDTA 1 mM; NaCl 140 mM; Triton 1%; SDS 0.1%; NaDeoxycholate 0.1%). Protein lysates (about 300 mg) were incubated overnight at 4°C with the anti-Hand2 (M-19, Santa Cruz; 1 mg) and protein G beads were added the next morning for about 5 hours at 4°C. After several washes in lysis buffer, beads were resuspended in Laemmli loading buffer and SDS-PAGE was performed under non-reducing conditions. Goat IgG antibodies were used as control. For Co-IP of endogenous embryonic proteins, 50 limb buds at E10.5 were dissected in PBS and processed as described [33]. The Hoxd13 or control rabbit IgG antibodies used for co-IPs were covalently cross-linked to G protein beads and bound proteins were detected with Hand2 antibodies (AF3876, R&D System).
ChIP was performed using wild-type fore- and hindlimb buds at E11.0 (38–42 somites). For each experiment, 85 limb buds were dissected, pooled and the freshly cross-linked chromatin divided among the starting samples. The average size of the DNA fragments in the cross-linked and sonicated chromatin was ∼500–2000 bp. Samples were processed as described [62] with the following modifications: protein G magnetic beads (Dynabeads, Invitrogen) were pre-absorbed with goat IgG (1–2 mg for 30 ml of beads for each sample) for minimally 1 hour at 4°C. After washing them with BSA-PBS (5 mg/ml), the beads were added to the chromatin extracts and gently rocked for 1 hour at 4°C. Afterwards, beads were spun down and the chromatin in the supernatant transferred to a new tube and incubated overnight with Hand2 antibodies (M-19, Santa Cruz; 1 mg) or goat IgG antibodies as control (1 mg). The following day, 25 ml of beads were added and the DNA-immunocomplexes were precipitated for 4 hours at 4°C. ChIP-enriched DNA samples were amplified by Q-PCR and conventional PCR. To compute the enrichment for a particular amplicon, its values were compared with the ones of a completely unrelated amplicon within the mouse β-actin gene that provides an additional negative control. The β-actin gene is located ∼114 Mb downstream of the ZRS on mouse chromosome 5. The fold of enrichment was then calculated as the fold of increase in the specific signal in relation to the values obtained when using non-specific goat IgGs for ChIP (values set arbitrarily at 1). All oligos used are listed in Table S1. Three ChIP experiments were performed using completely independent and fresh (i.e. non-frozen) chromatin preparations. The values obtained were analyzed and the graphs shown in Figure 4D (means ± standard error) were drawn using the Prism Graphpad Software (La Jolla, USA). The statistical significance of all results was assessed using the Mann-Whitney test as part of the Prism software package.
Mouse NIH3T3 fibroblasts were plated on 24-well plates and transfected using Lipofectamine LTX (Invitrogen) including a total of 500 ng of DNA. Reporter constructs were co-transfected with 100 ng of Hand2 and/or Hoxd13 and/or Gli3 expression constructs in combination with a Renilla luciferase vector. A detailed description of the generation of the expression constructs is available in Text S1. Cells were collected 28–30 hours post-transfection and luciferase reporter assays were performed using the Dual Luciferase Kit (Promega). Each assay was repeated at least 10 times. It is important to note that NIH3T3 cells do not express the endogenous Hand2, Hoxd13 and Gli3 genes (data not shown). For the co-immuno-precipitation assays in cells see Text S1.
|
10.1371/journal.pcbi.1002010 | Noise Contributions in an Inducible Genetic Switch: A Whole-Cell Simulation Study | Stochastic expression of genes produces heterogeneity in clonal populations of bacteria under identical conditions. We analyze and compare the behavior of the inducible lac genetic switch using well-stirred and spatially resolved simulations for Escherichia coli cells modeled under fast and slow-growth conditions. Our new kinetic model describing the switching of the lac operon from one phenotype to the other incorporates parameters obtained from recently published in vivo single-molecule fluorescence experiments along with in vitro rate constants. For the well-stirred system, investigation of the intrinsic noise in the circuit as a function of the inducer concentration and in the presence/absence of the feedback mechanism reveals that the noise peaks near the switching threshold. Applying maximum likelihood estimation, we show that the analytic two-state model of gene expression can be used to extract stochastic rates from the simulation data. The simulations also provide mRNA–protein probability landscapes, which demonstrate that switching is the result of crossing both mRNA and protein thresholds. Using cryoelectron tomography of an E. coli cell and data from proteomics studies, we construct spatial in vivo models of cells and quantify the noise contributions and effects on repressor rebinding due to cell structure and crowding in the cytoplasm. Compared to systems without spatial heterogeneity, the model for the fast-growth cells predicts a slight decrease in the overall noise and an increase in the repressors rebinding rate due to anomalous subdiffusion. The tomograms for E. coli grown under slow-growth conditions identify the positions of the ribosomes and the condensed nucleoid. The smaller slow-growth cells have increased mRNA localization and a larger internal inducer concentration, leading to a significant decrease in the lifetime of the repressor–operator complex and an increase in the frequency of transcriptional bursts.
| Expressing genes in a bacterial cell is noisy and random. A colony of bacteria grown from a single cell can show remarkable differences in the copy number per cell of a given protein after only a few generations. In this work we use computer simulations to study the variation in how individual cells in a population express a set of genes in response to an environmental signal. The modeled system is the lac genetic switch that Escherichia coli uses to find, collect, and process lactose sugar from the environment. The noise inherent in the genetic circuit controlling the cell's response determines how similar the cells are to each other and we study how the different components of the circuit affect this noise. Furthermore, an estimated 30–50% of the cell volume is taken up by a wide variety of large biomolecules. To study the response of the circuit caused by crowding, we simulate the circuit inside a three-dimensional model of an E. coli cell built using data from cryoelectron tomography reconstructions of a single cell and proteomics data. Correctly including random effects of molecular crowding will be critical to developing fully dynamic models of living cells.
| Transcriptional and translational regulatory networks control the phenotype of modern cells, regulating gene expression in response to changing environmental conditions and/or biological stimuli. It has been well established that intrinsic noise in gene regulation results from the discrete biochemical nature of the process [1]. There is also an extrinsic component to the total noise arising from cell-to-cell variation in the number of copies of the transcription and translation machinery (transcription factors, RNA polymerases, ribosomes, etc) [2]–[4]. Stochastic noise can lead to different phenotypic outcomes for a cellular population and, in certain fluctuating environments, the resulting heterogeneous population can be more optimal for growth than would be a population containing a single phenotype [5], [6].
Theoretical modeling of stochasticity in gene expression has been a topic of intense study in the last decade and has greatly increased our understanding of the effect that statistical noise has on gene regulation (for reviews see [7]–[11]). Without detailed information regarding spatial heterogeneity within a cell, models of stochastic gene expression are typically expressed in terms of the chemical master equation (CME), which describes the time evolution of the probability for a chemical system to be in a given state [12]. Various analytical methods including moment generating functions [1], [3], [13], the Langevin and Fokker-Planck equations [14], linear noise approximation [4], and many-body theory [15] are used to study such models of gene expression. Computer simulations, usually based on a variant of Gillespie's stochastic simulation algorithm (SSA) [16] are also widely employed to analyze gene network models that are too complex to be amenable to analytical modeling [17], [18].
Such theoretical studies have predicted and experimental measurements have shown [2], [19]–[23] that populations of cells can be quite heterogeneous, even when starting from an initially identical state. The large variance in the population distribution is usually ascribed to bursting in the process of gene transcription. Two models have been developed which can be used as a framework for quantitatively analyzing population distributions to infer the underlying gene expression kinetics.
The burst model (Figure 1A) of Friedman et al. [24] is based on the assumption that an mRNA's lifetime is short compared with that of its protein product. In that case, proteins will be produced in independent bursts with exponentially distributed sizes. The solution to the stationary probability distribution of protein in the continuous CME formulation of the model was shown to be the Gamma distribution where the and parameters were interpreted to be the frequency of transcriptional bursts relative to the protein lifetime and the mean number of proteins produced per burst, respectively.
Shahrezaei and Swain [25] further developed the analytical theory of gene expression, by deriving not only the time-dependent probability distribution for the burst model, but also the steady-state distribution for a two-state model of gene expression (Figure 1B; three-stage model in their nomenclature). In the two-state model a gene alternates between transcriptionally active and inactive states with constant rates. Their analytical distributions show that in addition to large variance within a population, bimodality can appear when transitions between the active and inactive states are slow. A similar model has also been used to analyze the switching behavior of a population due to rare large events versus the cumulative effect of many small events [26].
Computational modeling can greatly assist in understanding genetic systems where complexity exceeds the capacity of analytical solutions. In a model for an inducible genetic switch incorporating more of the complexity present in real biological systems (Figure 1C), the transitions between the active and inactive transcriptional states are no longer constant but depend upon an external inducer likely in a nonlinear manner. The positive feedback (PFB) loop changes the network topology by introducing an additional regulatory link. Both of these differences provide additional sources of noise in the circuit that may affect the probability distributions. Combining computer modeling of a complete genetic circuit with analysis using simplified analytical models can help to provide an overall picture of the dynamics of such a system.
Further complexity in modeling real biological systems comes from the spatial heterogeneity within a cell and molecular crowding in the in vivo environment. It is becoming apparent that the cell is not a well-stirred system [27]–[29]. Studies using cryoelectron tomography techniques [30]–[34] have revealed that individual macromolecules are not necessarily uniformly distributed inside the cell, but may be clustered in a spatially dependent manner. Spatial organization can affect reaction kinetics by increasing local concentrations of reactants and enzymes. Additionally, crowding and non-specific molecular interactions in the in vivo environment can lead to anomalous subdiffusive behavior for macromolecules, as measured experimentally [35], [36] and by computational modeling of bacterial cytoplasmic environments [37]–[39]. Accounting for spatial heterogeneity is a challenge to computational biology that must eventually be met and several such modeling studies have been undertaken [37]–[44].
Stochastic modeling of gene expression circuits in a three-dimensional bacterial cell poses several difficulties, both computational and informational in nature. Recently a “lattice microbe” method [37] was developed using GPU (graphics processing unit) computational accelerators to simulate diffusion of macromolecules within a modeled Escherichia coli cell packed with a distribution of obstacles according to reported proteomics data. It implemented a multiparticle reaction-diffusion algorithm on a three-dimensional lattice to perform simulations of cell-scale systems. With the lattice microbe method one can observe anomalous diffusion of macromolecules and track diffusive-reactive processes over the timescale of the cell cycle, with spatial resolution from 2–16 nm. On the informational side, painstaking efforts must be undertaken to obtain parameters for the models. Kinetic parameters, which are often obtained under in vitro conditions, must be validated by comparing modeling results to published experiments. Recent time-lapse fluorescence microscopy experiments have been able to track dynamic behavior for individual macromolecules in vivo [21], [45], providing an additional source for model parameters. Parameters obtained from in vivo single-molecule experiments are uniquely suited for stochastic modeling, as they provide population distributions not simply mean values from ensemble measurements. Equally importantly such parameters are measured under in vivo conditions and incorporate the effects of the cellular environment. Also, super-resolution imaging studies [46]–[49] provide further spatial information to complement the cryoelectron tomography data.
We present here a computational study of gene expression noise in the inducible genetic switch shown in Figure 1C using both well-stirred and spatially resolved models. Spatial models of E. coli cells were constructed to approximate cytoplasmic crowding under both rapid and slow growth phenotypes, with the latter being based on data from cryoelectron tomography [50]. Both spatial models were simulated using the lattice microbe method [37]. The genetic switch was based on the well-characterized E. coli lactose utilization system, parameterized using measurements from a recent series of in vivo single-molecule fluorescence studies [21], [22], [51] as well as published in vitro rate constants. We report the contributions to intrinsic noise from the regulatory elements of the inducible genetic circuit as well as the extrinsic noise due to in vivo crowding. Using the slow-growth model we investigate the effect of using experimentally determined cellular architecture in reaction-diffusion models, with implications for effects due to cell growth. Comparing the noise from the inducible genetic switch to the bursting and two-state models described above (Figure 1A,B), we consider what improvements in both modeling and experimental efforts are needed to develop stochastic models of gene expression with predictive power regarding phenotype switching and heterogeneity in cellular populations.
Since the stochastic switch model is more complex than can be solved using analytic methods, we used computational Monte Carlo methods to sample the master equation and estimate the probability distributions. Two stochastic approaches were used to simulate the lac kinetic model: a well-stirred method using the CME and a spatially resolved method based on the reaction-diffusion master equation (RDME). The RDME model of the lac circuit can be thought of as a superset of the CME model in that all of the kinetic rates used for modeling reactions in the CME based model are also used in the RDME model, but with additional parameters regarding the spatial localization of particles and their diffusion in three-dimensional space.
We analyzed the capability of the burst and two-state analytic models of gene expression to recover parameters from our stochastic simulations of an inducible switch by fitting molecular distributions. We used a maximum likelihood method to estimate the model parameters. Briefly, the likelihood of the model parameters having produced a set of observations is given bywhere is the conditional probability of observation occurring given the parameters . The parameters that maximize this likelihood function are those that describe the best fit of the model to the data, assuming a uniform prior distribution for the parameter probabilities. To find the best parameters for a model of gene expression, was calculated using the model's steady-state probability density function with the values being the protein counts from the 10,000 simulations. The parameter values that minimized the negative log of the likelihood function were then found using downhill simplex minimization as implement in the Matlab fminsearch function. We estimated the confidence intervals for different sample sizes by taking 1000 random sets of either 50 or 200 cells from the full set of 10,000 and performed maximum likelihood estimation on each of these data sets. The confidence range for each parameter was then defined by the middle 95% of the values obtained during these random resamplings.
The burst model was first expressed in terms of parameters and by Friedman et al. [24] as the Gamma distribution. However, since our stochastic simulations produced discrete protein counts, we used the discrete formulation for the steady-state probability density derived by Shahrezaei and Swain [25] in terms of a negative binomial distribution (10)with parameter being the burst frequency (bursts per mean protein lifetime) and being the burst size (proteins produced per burst).
The two-state model was fit using the steady-state probability density function derived by Shahrezaei and Swain [25]:(11)(12)
In this expression the parameters are , , (the activation rate), and (the inactivation rate), the latter two being expressed in units of mean protein lifetime. Additionally, , , , and is Gauss's hypergeometric function. Fitting with all four parameters free often resulted in convergence in a local minima, so we adopted a fitting procedure whereby we first constrained the and parameters and fit only and to obtain initial estimates of these two parameters. In the fully induced state the above probability density function reduces to a negative binomial distribution with no dependence on or , only and . Since neither nor depend on inducer concentration, it is a reasonable approximation to use the values for and in the fully induced state as initial estimates for all inducer concentrations. After obtaining an initial fit for and , we then performed another fit with and unconstrained and with and allowed to vary ±5%. This procedure resulted in convergence at a higher likelihood score than when all four parameter were fit simultaneously for all distributions except one.
Here we present the result of our study into the noise effects in the inducible lac genetic switch. The first two sections describe the fitting of model free parameters to data from single-molecule fluorescence studies on E. coli populations. The next two sections analyze noise in the well-stirred circuit due to its regulatory control elements. The final two sections report on changes to the behavior of the circuit from in vivo effects, using a model of a spatially heterogeneous, crowded cell and then an experimentally determined cell structure under an alternate growth phenotype.
In a recent in vivo single-molecule fluorescence study, Choi et al. measured the distributions of a fluorescent reporter protein under control of the lac operator in individual E. coli cells at various inducer (TMG) concentrations [22]. They performed the measurements in the absence of LacY's positive feedback by replacing its gene with that of the membrane protein Tsr in the lac operon. This enabled an accurate determination of the protein distribution produced by the circuit at a given inducer concentration without any confounding non-linear effects due to enhancement of the internal inducer concentration by LacY. In the absence of DNA looping, they were able to fit their observed distributions to a gamma distribution , where was interpreted as the frequency of transcriptional bursts relative to the protein lifetime and as the mean number of proteins produced per burst. They observed a relatively constant value for the burst frequency of 3–4 and a linearly increasing relationship between burst size and inducer concentration at low to intermediates concentrations.
To understand the origin of the linear relationship between burst size and inducer concentration and to reproduce this behavior in our model, we derived an expression for the burst size as a function of kinetic parameters in our model. As long as bursts are infrequent relative to protein degradation, i.e. once a free operator is bound with a repressor it remains bound for a significant fraction of the cell cycle, transcriptional bursting from the lac operon can be modeled as a Markov process with competition between RNA polymerase (RNAP) and the various LacI species for binding to the free operator (see Figure 4).
Transcription initiation by RNAP was modeled as a pseudo first order process (Equation 4), with a rate constant of . The two repressor states with potentially significant binding affinity were and , shown in Equations 1 & 2. Free repressor binds with free operator with a rate constant of resulting in a pseudo first order rate of . Given the current debate surrounding the binding affinity of the state to the operator, we set the rate constant to be proportional to the free repressor binding constant and analyzed the effect of varying the proportionality constant on the pseudo first order rate . This model of transcriptional bursting assumes that the binding of to the free operator is negligible at low inducer concentrations by assuming and ignoring Equation 3. In practice, this condition was satisfied when . We used the upper limit in our model, which is within the range experimentally reported [76].
Following the unbinding of a repressor from the repressor–operator complex, the probability of transcription initiation (and subsequent mRNA creation) occurring at the free operator as opposed to a repressor re-binding is(13)
The probability of a given number of consecutive transcription initiation events (the size of the mRNA burst) then follows a geometric distribution with of which the mean is . However, repressor unbinding events that produce no mRNA are not observable as a burst, therefore the mean number of mRNA produced in a transcription bursts (B) is(14)
Combining Equations 13 and 14 gives the expression for the mean transcription burst size in terms of the rate constants for transcription initiation and repressor binding(15)
Given the inducer mass balances (see Supporting Text S1) and the expression for the total number of repressor dimers , one can derive the equilibrium concentrations of the two repressor specieswhere is the inducer concentration at which half of the repressor monomers are bound to an inducer molecule. Substituting and into Equation 17 gives the expression for the transcription burst size as a function of inducer concentration(16)
From this last equation it is clear that the transcription burst size will be linear over the entire range of inducer concentrations only when . Figure 5 shows the effect of varying , of particular interest are the very low values. When , the transcription burst size does not linearly increase over the range of inducer concentrations for which this behavior has been reported (0–200 ). In the model here formulated, a linear relationship between size and inducer concentration exists only when the binding affinity of for the free operator is comparable to that of . For our simulations, we therefore chose , such that , as this value assumed no effect on the unbound repressor monomer due to a single bound inducer and gave a strictly linear relationship for all inducer concentrations.
To obtain values for the model parameters , , and , we used the distributions for LacY reported by Choi et al. [22], specifically the inferred burst frequency (bursts per cell cycle) and size parameters ( and ) from their gamma distribution fits. From Equation 18, the mean transcription burst size as a function of inducer concentration is . This equation is linear in inducer concentration and by fitting it (multiplied by the mean number of proteins produced per mRNA) to the experimental protein burst sizes, as shown in Figure 6, one can constrain the kinetic parameters. The y-intercept of the line fixes the ratio of transcription to repression in the uninduced state () and the slope can then be used to obtain = 17.6 for TMG.
The linear fit, however, only fixes the ratio between and . To recover unique values for these two rate constants, we next considered the mean duration of each transcription burst. The interpretation of the shape parameter of the gamma distribution as the burst frequency is only meaningful if the burst duration is short compared to the protein lifetime. In that case, individual exponentially sized bursts can be considered exponentially distributed in time and therefore act independently to give rise to a gamma distribution of protein abundance. In setting rate constants for the model, then, we wanted to ensure that the burst duration was appropriately short.
The burst duration is simply the mean time for a repressor to bind to a free operator. Given a constant , a linear relationship between burst size and inducer concentration also implies a linear relationship between and inducer concentration as can be seen from(17)where in the last step. For TMG, the linear relationship between burst size and inducer concentration extended to at least ∼200 , which is ∼11 times the value for TMG of 17.6 . From Figure 7 it can be seen that the interpretation of as the burst frequency begins to break down once is >5% of the protein lifetime. Using 5% of the protein lifetime as for 200 , we can compute the value for that gives the appropriate : , using a cell doubling time of 55 minutes. With this value for the repressor binding rate, a single repressor molecule in an E. coli cell would take ∼200 s to find a free operator. This is somewhat faster than the 354 s reported by Elf et al. [51]. Using the above value for and the ratio of to from the linear fit of the experimental data we obtained the value for the transcription rate = . This rate for transcription initiation resulted in a steady state concentration of ∼2500 LacY molecules per cell in the fully induced state, within a factor of two of the ∼1000–1200 reported in the literature [22], [26]. The value also falls within the range of 1000–3000 seen for other highly expressed proteins in E. coli [85]. Accurate measurements of the burst duration in the lac system, particularly in the fully induced state, would increase the accuracy of our model.
In order to reproduce a burst frequency of over the mean LacY lifetime in the model, the repressor should dissociate from the operator with a frequency , assuming that each dissociation event produces a burst and that the cell cycle. The burst frequencies inferred by Choi et al. for TMG levels ≤100 are relatively constant with a mean of ∼3 bursts. This corresponds to = . Since the dissociation of a repressor dimer is not thought to be significantly affected by the binding of a single inducer molecule, = . The affinity of a repressor dimer with two bound inducer molecules, however, is thought to be much lower, i.e., the binding of a second inducer molecule essentially knocks the repressor off of the operator. In the absence of this effect, the response to an increase in inducer concentrations would take a significant fraction of the cell cycle. Elf et al. reported a response time of <60 seconds for addition of IPTG to concentrations from 50 – 1 mM [51]. Therefore, we fit such that the response of the model to increase in IPTG agreed with the published data. The best fit value was obtained for (shown in Figure 8A).
The final kinetic rates to be defined were those regarding the binding of TMG to the repressor–operator complex (Equations 10 & 11). As discussed in Methods, we used the same dissociation rates as for IPTG, leaving only the association rates and , both of which can be derived from the value, which is the inducer concentration at which half of the repressor–operator complexes have a bound inducer. Figure 8B shows the effect of varying on the burst frequency. As approaches , the burst frequency begins to diverge from its expected value. This is due to the increasing occupancy of the O state, which can decay much more quickly into a free operator than the other repressed states; with operator free more often, there are more bursts over the lifetime of a protein. A value of 3 mM for gave the best agreement with the experimental burst frequencies for TMG.
Using the derived rates, we performed well-stirred stochastic simulations of the lac model in the absence of LacY positive feedback (NPF model), obtaining the stationary LacY distributions as a function of internal inducer concentration shown in Figure 9. Compared to the intrinsic noise of the two-state model, the NPF model contains additional noise contributions from the non-constant rates for transitioning between active and inactive transcriptional states. The distributions showed the widest cell-to-cell variability due to the intrinsic noise of the system at intermediate inducer concentrations of 50–400 . At high inducer concentrations the population migrated toward a less variable distribution, as expected. Up to 100 , the population distributions agreed well with those reported by Choi et al. but at 200 the agreement began to break down. This discrepancy at concentrations >100 was caused by two primary factors: the burst duration and the action of inducer knocking repressor off of the operator. Increasing the repressor binding rate would improve the fit by decreasing the duration of each burst, but would cause a large increase in the total number of LacY molecules in the fully induced state, which is not supported experimentally. Alternatively, one could increase the value, causing less inducer instigated dissociation of the repressor–operator complex, but this would decrease the responsiveness of the circuit to addition of inducer, which is also not supported experimentally. Clearly, in order for the model to have greater predictive power, additional features would be necessary. For example, adding a delay between production of mRNA to account for the steps of RNAP open complex formation or more detailed modeling of translation. But lacking the in vivo experimental results to validate any additional complexity, we chose to ignore these effects and analyzed the model as described.
The gene regulation function (GRF) of an genetic system describes the relation between the activity of a gene and its regulatory control elements [86]–[88]. In the steady state, protein production is balanced by protein degradation/dilution. The mean protein count as a function of the control elements provides a method to analyze a GRF. The mean number of LacY per cell as function of the TMG concentration (Figure 9D) and the fraction of time spent in the transcriptionally active state (Figure 9F) show the regulatory behavior of the NPF model. We saw a typical sigmoidal regulatory response that was well fit by a Hill equation with an inflection at 312 and a Hill coefficient of 2.11. In a stochastic system, though, the mean rate of gene expression is just one piece of information. As important for a stochastic GRF is how the distribution changes with inducer concentration. The Fano factor (variance/mean) provides a measure of the variation of the distribution. For reference, the Fano factor of a Poisson process is 1. For the NPF model (Figure 9E) the Fano factor monotonically increases until 100–200 where it peaks at a value of ∼60 and then begins to decrease ending at a lower value of relative noise than at zero inducer.
Next we investigated noise in the inducible genetic switch when the positive feedback regulatory link was active (PFB model). The lacY gene located in the lac operon codes for the integral membrane protein LacY, which actively imports inducer molecules (lactose/ co-transport) establishing a positive feedback loop as shown in Figure 1C. The presence of active LacY in the membrane creates a concentration gradient enriching the intracellular environment with inducer molecules relative to extracellular space. For a fixed concentration, the underlying GRF for the lac operon therefore operates not only at an increased inducer concentration but, since the number of LacY is different for each cell, across a distribution of internal inducer concentrations.
We calculated the population distributions for the PFB model using well-stirred stochastic simulations at various concentrations. Starting from a stationary population distribution in the absence of inducer, each population of 10,000 cells was subject to an instantaneous increase in and simulated for twenty-four hours. Above an concentration of ∼10 , cells in the population began to switch to an induced state in which LacY expression was near its maximum value (see Figure 10A and B). Above ∼25 the transition to full expression was relatively concerted throughout the population. In the range of 10–25 , though, there were two transiently stable subpopulations, one uninduced and the other induced – the overall population was bimodal for a time.
To quantify the switching behavior of the population, we classified cells at regular time intervals as uninduced with <400–600 LacY (best fit for each ) or induced with >1750 LacY. Each subpopulation was then analyzed separately. The mean and variance of the distributions (Figure 10C) show that, after an initial response phase, the distribution of the uninduced subpopulation was stable over time. This was true even as the total number of cells in the uninduced population was decreasing as cells within it were switching to the induced state. At intermediate inducer concentrations, the uninduced cell population appeared to reach a stationary distribution from which cells independently and stochastically transitioned to the induced state. In contrast, at higher inducer concentrations the population migrated as a whole in a more downhill-like manner.
Noise in a GRF can be expressed in terms of its effect on the phenotypic variance in a population under identical environmental conditions. To compare noise between the NPF and PFB models, we first mapped concentrations to mean concentrations in the uninduced and induced subpopulations (in the NPF model = ). We then compared both the mean of the LacY distributions and the Fano factor for the two models. The mean values for the LacY distributions (Figure 11) were similar but the noise in the uninduced subpopulation was significantly higher in the model with positive feedback. Since the underlying GRF is equivalent between the two models, it is the action of the GRF on the distribution of concentrations that gives rise to the increase in intrinsic noise in the PFB model.
Having established the well-stirred PFB stationary distribution, we next evaluated the effect of in vivo molecular crowding on the distributions, the PFB+IV model. One obvious reaction subject to spatial effects is the rebinding of the repressor to the operator following an unbinding event. Immediately after unbinding, a repressor is necessarily localized near the operator, i.e. it has a memory of its location. As was shown by van Zon et al. [27], this memory effect increases the probability of repressor rebinding at very short times compared to a well-stirred approximation. Previous studies only considered the effect of normal diffusion following unbinding but there is an additional effect caused by anomalous diffusion due to in vivo crowding.
To investigate repressor rebinding in an in vivo environment, we performed reaction-diffusion simulations of a volume centered on an operator immediately following unbinding of a repressor. We varied the packing density of the approximated in vivo environment to study its effect on rebinding. Figure 12A and B shows that there is an anomalous effect at short time scales (<1 ms). Repressor diffusion at very short time scales is normal at the in vitro rate, but between 1–100 there is a period of anomalous behavior, and at very long time scales repressor diffusion returns to normal diffusion behavior with a lower diffusion coefficient D. Brownian dynamics simulations of proteins in a virtual in vivo environment [39] show a similar anomalous behavior when including only steric constraints with a minimum in the time exponent of ∼0.8 for proteins slightly larger than the 75 kDa repressor dimer. When electrostatic effects are included in the Brownian dynamics simulations, however, the apparent diffusion coefficient as well as the anomalous exponent change greatly, so our results should only be considered an upper bound on the in vivo effects. Including further electrostatically driven interactions such as non-specific binding, will increase the anomalous behavior of the repressor.
The anomalous behavior of the repressor causes it to spend more time near the operator following unbinding than would be expected for purely Brownian diffusion, leading to more encounters with the operator and a potentially greater probability of rebinding. To measure the change in rebinding probability, we counted the number of repressors that rebound to the operator following unbinding versus the number that escaped into bulk solution, defined here as leaving the simulation volume. As can be seen in Figure 12C, as the density of in vivo crowding increases, the probability of rebinding goes up. Compared to an in vitro unpacked environment at 15% probability of rebinding, at 50% packing the probability of rebinding is ∼24%. The distribution of escape times also broadens (Figure 12D) with particles in general taking longer to diffuse away. The anomalous memory effect resulted in the duration of some bursts being significantly shorter than expected.
To study the effect of burst duration differences on the stationary LacY distributions in a population, we used our lattice microbe method to generate PFB+IV trajectories of spatially resolved rapid-growth E. coli cells (see Methods). Beginning with the stationary distribution from the well-stirred PFB population, 100 cells were simulated at five internal inducer concentrations for one hour, slightly longer than the duration of a cell cycle (55 minutes), see Video S1. Over the course of the simulations, distributions in the in vivo models gradually migrated to lower mean values and lower noise, as can be seen in Figure 13. Two factors caused this migration: First, the shorter burst durations due to the anomalous diffusion effect described above resulted in fewer proteins being produced per burst and more time spent in the inactive state led to more frequent bursts and less noise. Second, the effective increase in repressor due to the decreased reaction volume. In contrast to spatial effects in an in vitro environment [27], it appears that in vivo crowding lowers both the mean value and the noise in distributions of observables. Since bacterial cells such as E. coli are known to have packing density changes during different portions of the cell cycle and/or growth conditions, this presents the possibility of measuring these in vivo effects on living cells if the observable distributions can be accurately quantified as a function of the cell cycle or growth conditions.
As a first attempt at addressing how changes in the cellular environment due to growth conditions affect gene expression noise, we used CET of E. coli cells under slow growth to build a whole-cell model of an individual bacteria (Figure 14A). Under conditions of slow growth in minimal media E. coli B/r K grows as elongated cylinders with diameter ∼400 nm [89], which are amenable for CET [50]. The tomograms were used to identify the membrane-enclosed volume of an individual cell along with the three-dimensional position of ribosomes within it. The E. coli B/r K cell under slow growth had only of the volume of typical fast growing cells. A central region of the cell was devoid of ribosomes and inferred to be the location of the condensed nucleoid.
We studied the operation of the lac circuit in the slow-growth phenotype (PFB+IV+CET) using 100 random replica cells. Each replica used the same experimentally measured cellular geometry and ribosome positions, but a random distribution of other molecules including a condensed chromosome (see Methods for details). Cells were simulated using the lattice microbe method in 15 external TMG, starting with LacY and mRNA counts sampled from the uninduced stationary distribution of the well-stirred PFB model, see Video S2. Simulations were run for either one hour or until the cell had induced, whichever came first. There were clear differences between the slow- and fast-growth in vivo models. Of the 100 slow-growth cells, 11 induced within one hour whereas only a single fast-growth cell induced in the same time period. Also, the mean number of LacY molecules in the uninduced slow-growth population increased ∼15% over the course of one hour, compared to the fast-growth population which decreased ∼15%.
Analysis of the simulation trajectories revealed that the primary cause of the differences in LacY distributions between the slow- and fast-growth models was an increased mean inducer concentration in the smaller cells, 100 versus 42 . For a given number of LacY proteins, the cells with the smaller volume had an increased internal inducer concentration. The increased levels of inducer caused a slight lengthening of the mean duration of free operator events, 68 seconds versus 64 seconds, and a corresponding larger burst size. A bigger change was observed in the mean lifetime of the repressor–operator complex, which decreased to 430 seconds from 730 seconds (Figure 14B,C). The decrease effected an increase in the mean number of transcription bursts per hour, to 4.3 from 2.6.
The slow-growth model provides a first approximation as to the effect of differences in cellular architecture on stochastic gene expression. The model assumed the same number of repressor molecules for smaller cells, which may not be accurate as repressor is known to regulate its own expression. However, since the largest effect was due to an increased rate of repressor unbinding due to elevated inducer levels, which is independent of repressor concentration, we consider the general result of increased burst frequency and rate of induction in smaller cells to be intriguing. It implies that there might be a difference in the switching properties during the first part of the cell cycle following division when a large burst of LacY would have an increased influence on switching due to the reduced cellular volume. Such an effect could potentially be measured using cell synchronization techniques. Although specific ribosome placement likely also influenced repressor rebinding in the slow-growth model, any differences were overshadowed by the effect of the cell volume change. Nevertheless, in a situation where the placed macromolecules are involved in the reaction kinetics, we anticipate accurate (non-uniform) placement will take on much greater importance.
Another large difference between the slow- and fast-growth models arose due to the presence of a condensed nucleoid coupled with the smaller cell diameter. In the fast-growth cells the chromosome was assumed to be diffuse and not an obstacle for mRNA diffusion. In the slow-growth cells, the chromosome was randomly placed in the ribosome-excluded region observed in the tomograms and it represented an obstruction for mRNA diffusion. Additionally, the operator was positioned in the center of the fast-growth cells and at the edge of the nucleoid in the slow-growth cells. As can be seen in Figure 14D there was a dramatic increase in localization of mRNA in the slow-growth cells as a result of this arrangement. A recent report of mRNA localization in bacteria [90] suggests that the relative locations of transcription and translation in bacteria may indeed be correlated. If that is generally true, then in systems where the location of protein synthesis affects the reaction kinetics it will be important to know the actual position of the gene in the cell and measurement of the dispersion of the transcripts might be one way to quantify whether the gene is physically located near the site of translation and translocation.
Fitting protein population distributions to gene expression models will be a key step in developing simulations of other stochastic cellular systems with predictive power. Parameters obtained from fitting the distributions will drive the computations. Our stochastic simulations of the inducible lac switch provide an opportunity to test the process of extracting parameters from a population distribution arising from a complex gene expression system using simplified but analytically tractable models. To do so, we fit the stationary population distributions from our simulations to both the burst and two-state models (Figure 1A & B) and evaluated their capability to recover the stochastic rate constants used in the simulations (e.g. , , etc). The analysis was performed for each of the different noise variations described above, corresponding to the NPF, PFB, and PFB+IV simulations. We excluded the PFB+IV+CET simulations from this study as they were not performed over a range of inducer conditions. The best fit parameter values were obtained by maximum likelihood estimation using the stationary probability density function (PDF) for the burst and two-state models, Equations 12 & 14 in Methods. Fits were performed using 10,000 cells for NPF and PFB simulations and 100 cells for PFB+IV simulations.
Figure 15A & B show parameter estimates obtained from fitting using the burst model's gamma distribution PDF (Equation 12). The and parameters (the B subscript indicates parameters for the burst model) reliably recover the burst frequency and burst size, respectively, in the NPF simulations at low inducer concentrations, but diverge from the simulation values above ∼100 . This is as expected as the model is only valid when the duration of each burst is short enough that sequential bursts can be considered as occurring independently, <5% of the protein lifetime as shown in Results. In particular the divergence occurs near the switching threshold, making this model most suitable for analyzing the system in the uninduced state with low expression levels. However, the clearness of the biological interpretation for the model parameters as the burst frequency and size make the model extremely valuable over the regime it is valid.
Fitting the NPF simulation data to the stationary PDF of the two-state model (Figure 1B; Equation 14) provides good parameter estimates over a wider range of inducer concentrations. The fits are shown in Figure 15C–F for the parameters (; the TS subscript indicates two-state), (), (the rate constant for operator activation), and (the rate constant for operator inactivation), respectively. As the inducer concentration increases, though, many more cells are required to obtain reliable estimates. Using even 10,000 cells, we were unable to obtain good fits for the highest expression levels. At these inducer levels so little time is spent in the inactive state that the difference in likelihood values for different switching rates is insufficient to find a unique maximum using 10,000 samples. However, as the time spent in the inactive state approaches zero () the probability distribution approaches a negative binomial distribution without dependence on or , so it is possible to estimate the and parameters in the fully induced state by fitting to a negative binomial.
The two-state model therefore appears to be a reasonable method for fitting the NPF simulations. Using the fitting parameters (along with known or estimated mRNA and protein degradation rates), one can readily recover the transcription and translation rates as well as the rates of the operator switching between active and inactive states at a given inducer concentration. Even though switching between active and inactive states in the lac switch is not a first order process – it is controlled by 14 reactions – at a given inducer concentration the steady state switching times are reasonably well-approximated by a single exponential. A further improvement in the two-state model would allow and to depend on the inducer concentration using, e.g., a Hill function. An analytic solution to such a model would allow extraction of parameters from a multivariate fit using data across all inducer concentrations. However, to the best of our knowledge, the analytic form of such a model has not been derived.
Using the steady state distributions from the PFB simulations, neither model achieves good fits. For the two-state model, the and parameters are recovered correctly, but the fits for the and parameters are lower than expected. The poor fit for these parameters is due to noise in the switching rates of the cell population caused by differences in internal inducer concentrations. With positive feedback, it will be very difficult to reliably estimate model parameters from population distributions due to non-linear noise. Fitting to experimental data should be done in the absence of positive feedback, such as by using gene knock-outs to eliminate circuit components responsible for positive feedback. However, if an analytic model were developed including positive feedback effects, comparison of systems with and without these effects could provide estimates of positive feedback parameters, e.g. inducer transport rates.
Fits to the PFB+IV simulations as well show deviations from the expected values; in vivo crowding noise changes the parameter fits. For these simulations, an additional source of discrepancies with the models is the non-Poissonian behavior of repressor rebinding – there is a positional memory in the system for a short time following unbinding. In our simulations the effect from in vivo conditions due to excluded volume is modest, but there are other in vivo factors still not accounted for in them, especially non-specific binding as recently reported by McGuffee and Elcock [39], which would have an even larger effect on repressor rebinding. Also, repressor rebinding most likely occurs via a series of 1D sliding and 3D hopping steps, the effect of which on rebinding in a crowded environment is not known. Accounting for in vivo effects when deriving parameter from experimental population distributions, which would include in vivo noise contributions, will be difficult. Possibly an iterative process of refinement may be required, starting with model estimates and proceeding through multiple rounds of spatial simulation.
Overall, it appears that fitting population distributions to the two-state model could prove to be an effective way of obtaining rate constants for stochastic simulations of gene regulation. Single-molecule in vivo fluorescence imaging provides a way to experimentally measure these distributions. Measurements over a range of regulatory conditions could then be used to build a stochastic gene regulation function, provided the actual probability distributions from single-molecule experiments were available at each condition. However, it is important to acknowledge that our simulations did not include a contribution from global extrinsic noise. Noise in our simulations under conditions of high expression approaches Poissonian, as expected from the intrinsic noise of an uncorrelated random process. A recent study has clearly shown, though, that there is a constant level of global extrinsic noise in gene expression in E. coli, maintaining population heterogeneity even at high levels of gene expression [85]. This implies that a way to correct for the global extrinsic noise will be needed in order to fit experimental population distributions at high expression levels.
The probability distribution for a stochastic biochemical system to be in a particular state represents the totality of information about the system. From it various measures of the behavior of the system such as the mean first passage time between two states or their relative population at the steady state can be obtained. For models of stochastic gene expression, two relevant reaction coordinates are the number of protein and mRNA molecules in the system. We used our stochastic simulations to reconstruct the two-dimensional probability landscapes (negative log of the PDF) of the NPF and PFB models at two external TMG concentrations (Figure 16).
The steady-state landscape of the NPF simulations at 500 inducer shows a bistable mRNA distribution that has been reported by others [25], [91]. One minima is located near 0 /1700 LacY and the other near 10 /1800 LacY. Note that the stable 10/1800 point does not imply that 180 LacY were produced per , as the degradation rates of the two molecules differ. At 500 inducer, the net time some cells stochastically spend in the inactive state is greater than the typical lifetime of the mRNA bursts. These cells then drift to a zero mRNA abundance. The higher density is caused by an accumulation of these cells near the zero mRNA level until their next mRNA burst pushes them back into a random cycle around the mean mRNA burst size. Interestingly, though, the protein distributions at the two mRNA minimum are different, with the protein abundance being slightly lower in the lower mRNA minimum. This means that the protein and mRNA probability distributions are not completely independent of each other; the joint probability distribution has cross terms. While not a large difference, it is nevertheless possible that the joint protein–mRNA distribution could be used to obtain better parameter fits for the two-state model with fewer cells, if the mRNA counts were known.
The bimodal distributions seen in the LacY distributions from the PFB simulations (Figure 10) are recapitulated in the probability landscape for switching. The two-dimensional landscape allows classification of both the uninduced and induced states in terms of their relative protein and mRNA abundances. Additionally, the landscape reveals the transition path for switching from the uninduced to the induced state. One can imagine two possible scenarios for the transition, either the gradual build-up of protein by a series of small bursts, or alternatively, by the random occurrence of a small number of larger bursts. For the lac system with DNA looping, Choi et al. [26] have persuasively argued for the random large bursts as the switching initiator. The switching mechanism of the stochastic system in the absence of looping, however, is not so clear. The probability landscape of our lac model suggests that it is actually the occurrence of several large mRNA bursts, on the order of >10 molecules, in quick succession that is responsible for putting the cell on the path to induction. Cells can spend a significant amount of time in a high LacY but low state without inducing. This behavior is apparent in the cell trajectory plotted in Figure 16B. Switching therefore is a process in which not only a protein threshold must be crossed, but also an mRNA threshold.
Our goal with this study was to go beyond previous stochastic simulations of the lac circuit by using information from single molecule protein distributions and experimentally determined cellular architecture to constrain the kinetic parameters and estimate the effect of spatial heterogeneity on the response of the switch. The kinetic model of the inducible lac genetic switch presented in this study illustrates the utility of incorporating single-molecule, single-cell data when modeling cellular biochemical systems. The model was derived using a kinetic framework reproducing a linear relationship between protein burst size and inducer concentration at low concentrations, as has been reported experimentally. Analysis of the linear relationship in terms of inducer–repressor–operator interactions suggests that the stoichiometry of repressor binding is such that repressor dimers with one bound inducer still have significant affinity for the lac operator. Furthermore, single-cell population distributions were used to obtain estimates of the effective rate constants for transcription and repression in the cell. With future increases in performance of the lattice microbe simulation method it should be possible to iteratively refine the kinetic rate constants to account for the effects of cellular architecture, such as we obtained here from CET experiments, and cytoplasmic crowding. Using such in vivo adjusted rate constants the in vivo models should then more accurately reproduce experimental population distributions, which are after all measured under in vivo conditions, than the well-stirred models.
The lac model without positive feedback provided a baseline for the noise in the regulation of the lac operon. Intrinsic noise at low gene expression was significantly higher than Poissonian and peaked when the promoter was active 10–30% of the time. The model with positive feedback produced similar mean values for a given intracellular inducer concentration, but the noise was substantially greater. We attribute this effect to the non-linear gene regulatory function operating on a distribution of intracellular inducer levels. Global extrinsic noise in the transcription/translation machinery is a large contributor to population heterogeneity at high levels of expression, but we excluded such noise from the current study.
Fitting of data from stochastic simulations of the lac switch with the burst and two-state models of gene expression showed both the potential and limitations of these models to interpret stochastic gene regulation. The burst model described the data well under conditions of low expression, when the gene was active for ≤5% of the mean protein lifetime, but diverged for increasing expression levels. The two-state model better described the data at higher levels of expression, but near full induction the error in the activation and inactivation rates became significant. Additionally, the fits provided estimates of the number of cell measurements necessary to produce reliable parameter estimates. With 50 cells the worst-case relative error was ±90%, but with 200 cells it dropped to ±32%. Fitting to joint mRNA–protein distributions might improve parameter estimation. Fits to data with positive feedback indicated that both models were unable to reliably extract parameters from populations with such feedback.
Switching of cells from the uninduced to the induced state was observed in the positive feedback model without DNA looping over a range of low inducer concentrations. During switching, the uninduced population maintained a stable stationary distribution while cells stochastically transitioned to the induced population. The probability landscape showed that both an mRNA and a protein threshold must be crossed for a cell to switch to the induced state. The probability landscape for the DNA looping case is likely different, but additional model states would be required to accurately represent DNA looping.
Finally, we have presented what we believe to be the first whole-cell simulations of stochastic gene expression using experimentally obtained cellular architecture. These simulations showed that in vivo conditions can impact the stochastic noise in biological systems. Positional memory of transcription factors following unbinding, amplified by anomalous diffusion due to molecular crowding, introduces non-Poissonian statistics. In the case of our lac model in fast-growth cells, this effect caused a decrease in the mean value of the LacY distribution by up to 10% and its noise by up to 20%, for a given environmental condition. In a slow-growth cell phenotype we saw a large increase in burst frequency due to the smaller cell size, as determined from cryoelectron tomography. From this difference we infer that changes in cellular size and/or shape during the cell cycle can have an impact on stochastic processes. Since spatial noise can vary from cell-to-cell or even during the cell cycle so we consider it a type of extrinsic noise. The necessary computational resources and experimental data are becoming available such that computational biologists should consider adding spatial degrees of freedom into physical models of cellular biochemical networks.
|
10.1371/journal.pcbi.1000978 | Networks of High Mutual Information Define the Structural Proximity of Catalytic Sites: Implications for Catalytic Residue Identification | Identification of catalytic residues (CR) is essential for the characterization of enzyme function. CR are, in general, conserved and located in the functional site of a protein in order to attain their function. However, many non-catalytic residues are highly conserved and not all CR are conserved throughout a given protein family making identification of CR a challenging task. Here, we put forward the hypothesis that CR carry a particular signature defined by networks of close proximity residues with high mutual information (MI), and that this signature can be applied to distinguish functional from other non-functional conserved residues. Using a data set of 434 Pfam families included in the catalytic site atlas (CSA) database, we tested this hypothesis and demonstrated that MI can complement amino acid conservation scores to detect CR. The Kullback-Leibler (KL) conservation measurement was shown to significantly outperform both the Shannon entropy and maximal frequency measurements. Residues in the proximity of catalytic sites were shown to be rich in shared MI. A structural proximity MI average score (termed pMI) was demonstrated to be a strong predictor for CR, thus confirming the proposed hypothesis. A structural proximity conservation average score (termed pC) was also calculated and demonstrated to carry distinct information from pMI. A catalytic likeliness score (Cls), combining the KL, pC and pMI measures, was shown to lead to significantly improved prediction accuracy. At a specificity of 0.90, the Cls method was found to have a sensitivity of 0.816. In summary, we demonstrate that networks of residues with high MI provide a distinct signature on CR and propose that such a signature should be present in other classes of functional residues where the requirement to maintain a particular function places limitations on the diversification of the structural environment along the course of evolution.
| Enzymes are responsible for several critical cellular functions. The so-called catalytic residues are fundamental to attain the enzyme function. Those residues are often highly conserved within protein families sharing similar structure and function. Characterization of catalytic residues is essential for the understanding of enzyme function. However, this is a difficult task because conservation is a poor discriminator of catalytic residues due to the fact that many non-catalytic residues are highly conserved in a given protein family. We anticipate that variations in the structural environment of a catalytic site should be highly restrained in order for the protein to maintain its function along the course of evolution, and hypothesise that catalytic residues, due to these restrains, must carry a particular signature defined by networks of proximity sharing high mutual information (MI). We validated this hypothesis on a large data set of protein sequences with known catalytic residues, and demonstrated that catalytic sites are indeed surrounded by networks of coevolved residues. Such networks should also be present in other classes of proteins and we suggest that MI networks could be a novel feature of general importance beneficial for the prediction of functional residues.
| Catalytic residues play a fundamental role in enzymes and are generally expected to be conserved and located in the functional site of proteins. Even though characterization of catalytic residues (CR) is critical for the understanding of enzyme function, their identification remains a daunting task. To guide the identification of CR, several computational approaches have been developed based on different principles. To cite some examples: catalytic site features, amino acid physicochemical character [1], conserved functional groups density [2], sequence analysis (conservation, patterns, conserved blocks along the sequence, evolution, entropy, among others) [3], [4], [5], [6], [7], [8], sequence and structure properties [9], [10], [11], evolution and 3D structure information [12], [13], [14], [15], neural networks [16], 3D structure combined with ionization properties of a residue and its vicinity in the structure [17] and combinations of several of the above mentioned [18]. Conservation is the natural and intuitive way to predict functional residues in proteins. However, many non-catalytic residues are highly conserved and conversely, not all CR are fully conserved throughout a given protein family. On the other hand, residues involved in coevolving networks have been postulated to be functionally important [19], [20], [21] and several studies have provided evidence that they are important for specificity or allosteric regulation [22], [23], [24].
The structural environment of an active site must be highly conserved in order for the protein to maintain its function during the course of evolution. This places strict limitations on the amino acid diversity in the proximity of an active site, and it therefore seems plausible to hypothesise that catalytic residues would carry a particular signature defined by a network of close proximity of residues with high mutual information.
Although earlier published methods have suggested a linkage between functionally important sites and neighbouring coevolving residue [21], [25], [26] at present, to the best of our knowledge, no method explicitly show how the presence of such coevolving residues can provide quantitative information useful for catalytic sites identification beyond what is captured by conservation. Several methods have been proposed for identifying specificity defining positions (SDP) aiming at locating positions that are specific for a given subfamily and hence potentially could define its specificity [25], [26]. These residues are suggested to be located in the proximity of the active residues in order to carry out their role of defining the substrate specificity. The signal from such evolutional signatures could at first resemble co-evolution, and the overlap between the methods predicting SDPs and the method proposed here could seem substantial. However, the subfamily specific positions may not be coevolving, in fact they might be fully conserved within each subfamily, and Gouveia-Oliveira and Pedersen have described in details that such subfamily defining residues do not carry signatures of co-evolution but rather a phylogenetic signal that mimics coevolution [27]. The methods put forward by Gouveia-Oliveira et al, [27], Dunn et al. [28], and Buslje et al. [29] all attempt to reduce this phylogenetic bias in the signal for MI calculation aiming at identifying truly coevolving residue-pairs. Moreover, the method proposed here is hypothesis-free, and can be applied without any prior functional cluster classification of the input multiple alignment.
Here, we perform a large-scale benchmark analysis aiming at testing the hypothesis that catalytic residues carry a signature defined by networks of close proximity of residues with high mutual information. An investigation on the relationship between conservation, coevolution networks and catalytic residues is carried out on a dataset of 434 families of enzymes. We introduce a new concept, Mutual Information Proximity (pMI) that characterizes the mutual information network in the proximity of a given residue and analyse whether this measurement can complement the conventional conservation score for the detection of catalytic residues. The goal of this work is two-fold. First, we aim to validate the hypothesis stated above and demonstrate that proximity residue networks of high mutual information characterize functional residues. In doing this, we also aim at addressing the issue on the correlation between residues defined as SDP and residues carrying high signals of being part of the mutual information network. Secondly, we seek to integrate this mutual information signature to create a method able to identify catalytic residues useful for guiding the identification of functional sites in proteins.
Note, that in this work, we do not suggest that the proposed method should be more accurate than the other methods developed earlier for prediction of functional residues. We merely seek to demonstrate the existence of a mutual information network signature in the proximity of functional residues, and show that this signature is complementary to the conventional sequence conservation measurement, hence most likely would benefit any functional residue prediction method.
The main focus of this work was to investigate if mutual information could contribute beyond sequence conservation to the identification of catalytic residues. The result section naturally falls in three parts. First, we investigated how different measurements of sequence conservation could be used for the identification of catalytic residues. Next, a similar analysis was performed using different measurements of mutual information, and finally the analysis was carried out using a combined measurement of conservation and mutual information. Performance details of all methods included in the analysis are shown in supplementary table S2.
As catalytic residues are highly conserved, a natural measure used to detect them is the conservation score in a MSA. Here, we investigated three conservation measurements in four different conditions leading to twelve different conservation scores (for details see material and methods). The conservation measurements are all per-residue measurements, and their predictive performance for a given protein sequence is readily measured in terms of the AUC value. The results of this analysis on the 434 CSA Pfam families are shown in table 1.
The conservation measurement with the highest predictive performance in terms of AUC was the raw KL score with an average AUC value of 0.892 and an AUC01 value of 0.485. In terms of AUC, the raw calculation excluding both sequence weighting and pseudo count correction did perform best for all three conservation measurements. In terms of AUC01, the inclusion of sequence weighting in all cases did improve the predictive performance. The Max-Freq measurement performed significantly worse than both information-based measurements (p<0.0001, binomial test excluding ties). Although the performance is very similar between the raw Shannon and raw KL scores, the difference is highly significant (p<0.005, binomial test excluding ties). The difference between the raw and sequence weighted (c) KL score is borderline significant with a p-value of 0.05 in favour of the raw KL score for AUC and in favour of KL including sequence weighting when using AUC01. In order to make the subsequent analyses as simple as possible, for the remaining part of the work we used the raw KL score as a conservation measurement.
We analysed to what degree the predictive performance of the raw KL measurement depended on the number of sequences in the multiple sequence alignment (MSA) used as the source to estimate the conservation score (see figure 1). This figure clearly demonstrates that at least 10 sequences are required in order to make any meaningful predictions using the KL conservation measurement (similar results were observed for the other two conservation measurements). Note, that the variation in performance for each bar in the histogram is large and error-bars are not included (the raw data included in the figure are available in Supplementary table S2). The difference in predictive performance between the families with less than or more than 10 sequence members is however statistically highly significant (p<0.001, t-test).
We next turned to mutual information and analysed the environment of a catalytic residue by means of the mutual information carried by the surrounding residues. We introduced a cumulative Mutual Information concept (cMI) that measures the degree of shared mutual information of a given residue (above a certain significance threshold as measured in terms of the MI Z-score, see material and methods). We noticed that residues in close proximity with CR tend to have high cMI scores (see figure 2b). Furthermore, when measuring the proximity Mutual Information (pMI), which tells about the networks of mutual information in the proximity of a residue (within a certain distance threshold), the catalytic residues were observed to have higher pMI than other conserved residues (see figure 2c for an example).
We exploited this observation on the complete Pfam benchmark dataset, and calculated the performance of the pMI measurement as a predictor of catalytic residues. Using a distance cut-off of 7.5 Å to define the structural proximity, and a Z-score threshold of 6.0 to define reliable mutual information interactions (see [29]), the average predictive performance of the pMI measurement in terms of the average AUC and AUC01 values on the 434 Pfam entries was 0.843 and 0.342, respectively which in both cases is significantly different from random (p<0.0001, binomial test excluding ties). As the number of proximity interactions is used to normalize the pMI measurement, this predictive performance does not stem from any implicit bias in the data imposed by catalytic residues being in a particular state of solvent exposure.
To investigate how the mutual information measure (cMI) proposed in this work correlates to earlier proposed measures for SDP, we compared in terms of the Spearmans rank correlation the SDR Z-score values given in the SDR database (http://paradox.harvard.edu/sdr/) [30] to the cMI values. In doing this, we obtained a mean correlation value over the 158 Pfam families covered by both methods of 0.29+/−0.20 (for details see materials and methods). Even though this correlation is significantly different from random (p<0.01, binomial test excluding ties), it is far from perfect. This highly suggests that the cMI and SDR measures carry distinct information. We next calculated the correlation between the two measures and the KL (Kullback-Leibler) conservation score. Here, we obtained an average Spearmans rank correlation values of 0.64±0.21, and −0.04±0.17 for the SDR Z-score and cMI measure respectively. These results further demonstrate that the SDP and cMI measures are different in nature, and that SDR Z-score is highly related to sequence conservation whereas the cMI score is independent of the latter. This strongly suggests that the cMI measure is more information rich compared to SDP when combined with sequence conservation.
As the active site in most cases is defined in terms of multiple catalytic residues in close proximity, it is natural to suggest that a proximity score based on sequence conservation would be a strong catalytic residue predictor. Using the same distance cut-off as for the mutual information proximity score, we find that the proximity conservation score, pC, achieves an average predictive performance of 0.854 and 0.379 in terms of AUC and AUC01, respectively. These values are greater than what was obtained using the pMI score, but for both AUC and AUC01, the difference between the two methods is not statistically significant (p<0.05, binomial test excluding ties).
We finally applied the combined catalytic likeliness score (Cls) to identify catalytic residues. The Cls is calculated as a weighted sum of the KL conservation the pMI mutual information and the pC scores. The optimal parameters defining the score were identified using 5-fold cross validation as described in Materials and Methods. The parameters Zthr, DMI, DC, WMI and WC were found to have the following optimal values Zthr = 5.5±0.2, DMI = 8.0±0.1, DC = 5.6±0.5, wMI = 0.6±0.0, and wC = 0.2±0.0. The low standard deviation value on each parameter-estimate indicates that the parameter optimization is robust across the different cross-validation data sets. The average performance in terms of the AUC and AUC01 of the Cls score to detect catalytic residues was 0.927, and 0.594, respectively. This performance is significantly higher than the KL conservation, the pMI and the pC individual scoring functions (p<0.001 in all cases using binomial test excluding ties).
To investigate the individual contribution to the performance of the Cls score of the pMI and pC measures, we next searched for optimal parameters for a combined score including only one of the two proximity measures in combination with the KL conservation score. Estimating the optimal parameters using 5 fold cross-validation as described above, we find the following results (see table 2).
The AUC values for both of these methods are significantly lower that what was obtained using the Cls score combining the conservation score with both proximity measures (p<0.01 in both cases, binomial test excluding ties) demonstrating that the two proximity measures contribute distinct information to the combined Cls score. The difference between the two scores including only one proximity measure is not statistically significant when looking at the complete data set of 434 PF families. However, when looking at the subset of 172 PFam families that are covered by more than 400 unique sequences/clusters (corresponding to the number of clusters needed to provide reliable estimates of MI as shown by Buslje et al. [29]), the combined method including proximity mutual information, pMI, achieves a performance of AUC = 0.920, and AUC01 = 0.597. These values significantly outperform the performance values AUC = 0.889 and AUC01 = 0.559 of the combined method including proximity conservation, pC (p<0.05, binomial test excluding ties). This further underlines the observation that the pMI measure contributes information not included in the conservation scores.
To further illustrate that the two proximity measures contribute different information to the combined Cls-score, we in figure 2 display the role of the four prediction measurements, KL, pMI, pC and Cls for the identification of the catalytic residues in the Pfam entry PF00890 represented by fumarate reductase of Shewanella putrefaciens MR-1 (PDB entry 1D4C). This family was chosen from the subset of 172 Pfams entries mentioned above covered by more than 400 unique sequences/clusters (similar results are obtained for most other families in this set). The function of fumarate reductase is carried out by the active cite residues His364, Arg401, His503 and Arg544 [31]. It can be seen that the KL conservation score of the catalytic residues is relatively low (figure 2a) while both the pC, and pMI scores are high in the catalytic residue proximity (figure 2b, and 2c). Comparing the figures 2b and 2c, it is evident that the two proximity measures contribute different information to the combined, Cls, prediction score. Finally, the combined catalytic likeliness score (Cls) is depicted in figure 2d. The AUC values for the four prediction measurements shown in figure 2 are 0.92, 0.94, 0.98 and 0.99 (KL, pC, pMI and Cls respectively). These values translate into a number of false positive predictions at 100% sensitivity (corresponding to the number of non-catalytic residues with a prediction score higher than the lowest score obtained by a CR) of 47 (figure 2a), 39 (figure 2b), 15 (figure 2c), and 4 (figure 2d), again underlining the strong predictive power of the Cls measurements in identifying catalytic residues and eliminating false positive predictions.
The gain in predictive performance for detecting catalytic residues is consistent for families independently on the level of conservation of the catalytic residue, however the most dramatic gain in performance when including pMI is observed for families where the conservation of the catalytic residues is poor. If we for instance take the 217 Pfam families with the lowest predictive performance when using the KL conservation score and ask how many of these families gain in performance when including the pMI score, we find that this number is significantly higher compared to the corresponding number of families in the group of 217 Pfam families with the highest predictive performance using the KL conservation score (p<0.001, binomial test excluding ties). This difference in performance gain between the two subsets of Pfam families is not imposed by a difference in data size between the two sets as the average family size in the two set is comparable (p>0.1, t-test). The catalytic environment of an active site needs to be conserved in order for a protein family to maintain its function, and one might speculate that when the conservation of a catalytic residue is weak, the catalytic environment is maintained in great measure by coevolution.
We next determined the sensitivities of the different methods at different specificity thresholds. This analysis is summarized in table 3. The analysis clearly confirms the strong improvement across the entire benchmark data set of the predictability of catalytic residues imposed by the inclusion of the pMI score in the combined catalytic likeliness score. At all specificity thresholds, the Cls method did achieve the highest sensitivity. The difference in sensitivity between the Cls and the other methods is statistically significant (p<0.05, binomial test excluding ties) for all comparisons. The Cls score threshold corresponding to a specificity of 0.90 for the 434 CSA families is 1.44±0.26. This low standard deviation of the threshold score indicates that the Cls approach is stable across the different CSA families and suggests that the method can be applied universally to any enzyme protein family independently of diversities in structure, composition and size of the MSA, as long as the number of sequences is greater than 10 (see figure 1).
Catalytic residues are in general expected to be conserved and located in the functional site of a protein in order to attain their function. However, many non-catalytic residues are highly conserved as well and conversely, not all catalytic residues are conserved throughout a given protein family, making identification of catalytic residues a big challenge. The requirement to maintain a given catalytic function during the course of evolution places great limitations on the diversity of the structural environment of an active site. Therefore, here we put forward the hypothesis that catalytic residues carry a particular signature defined by networks of close spatial proximity residues sharing high mutual information, so that this signature could be applied to differentiate functional from other non-functional conserved residues.
We tested this hypothesis using a data set of 434 Pfam families each characterized by a PDB structure and one or more catalytic residues assigned from the CSA database, and investigated whether mutual information could complement conventional amino acid conservation scores and improve the ability to detect catalytic residues. Three methods to calculate sequence conservation were considered and the KL relative entropy (KL) was shown to significantly outperform both the Shannon entropy and maximal frequency measurements. We observed that sequence-weighting and low count correction do not improve the predictive performance for any of the methods. Additionally, in order to achieve reliable predictions the number of sequences required in the MSA was found to be relatively small. Only 10 sequences in the MSA were needed to reach AUC values of 0.89.
We observed that in the proximity of a catalytic site, residues are rich in shared mutual information (calculated as the cumulative mutual information, cMI): therefore, we defined a residue specific score characterizing this fact in terms of a structural proximity average (termed pMI) score. The pMI score was demonstrated to be a strong predictor for catalytic residues, suggesting that catalytic residues indeed carry a particular signature imposed by networks of mutual information. We compared the predictive performance of the pMI measure to that of a proximity measure based on sequence conservation and demonstrated that the two measures achieved comparable predictive performance but more importantly that they carried distinct information suitable as predictor of catalytic residues. Finally, we demonstrated that the conventional KL relative entropy sequence conservation, the pC and pMI measurements are complementary and that a combined catalytic likeliness score (Cls) of the three leads to significantly improved prediction accuracy. For instance, we found that, at a specificity threshold of 0.90, the KL, pMI, pC and Cls methods have a sensitivity of 0.716, 0.560, 0.604 and 0.816, respectively.
This work thus demonstrates in direct quantitative terms (gain in predictive performance) the contribution of the coevolution signal in determining catalytic residues, and hence goes beyond earlier published papers in the field [20], [21], [25], [26] and not only describe the observation that such signals might be present near functionally important residues but in details demonstrate how such information can be applied to guide their identification.
We also analyzed to what extent the score characterizing specificity defining positions (SDPs) and the mutual information derived score defined in this work carry distinct information on the functional neighbor of catalytic residues. We used data from the Paradox database to carry out the comparison, and compared SDP and cMI scores for a set of 158 families covered by both methods. The obtained results clearly demonstrated that the SDP and cMI measures are different in nature, and that SDR Z-score is highly related to sequence conservation whereas the cMI score is independent of the latter. This observation strongly suggests that the cMI measure is more information rich for the identification of functional residues compared to SDP when combined with sequence conservation.
In summary, we have demonstrated that mutual information provides a distinct proximity signature that can be applied to determine catalytic residues. The approach outlined is general, and we suggest that the method should be applicable to the identification of other classes of functional residues where the requirement to maintain a particular function places limitations on the diversity of the structural environment along the course of evolution.
The dataset was constructed based on the CSA database (version 2.2.11, released August 2009) [32]. CSA provides catalytic site annotation for enzymes in the PDB. Catalytic residues were defined as those residues thought to be directly involved in some aspect of the reaction catalysed by an enzyme (for a detailed description of the classification see [1]). The database consists of two types of annotated sites: an original, hand annotated set and an additional homologous set, containing annotations inferred by Psi-Blast and sequence alignment to one of the original entries. CSA contains 968 original literature entries, which belong to 455 Pfam families [33]. Due to some inconsistency between CSA and PDB, a few families were eliminated, so that we ended up with a dataset of 434 protein families (each of one containing at least one PDB entry), which in turn include a total of 1212 CSA, annotated catalytic residues. For 9 of the 434 families the selected PDB representative was an NMR structure. For these PDB entries the first model was selected to represent the structure. The 434 Pfam families included in the benchmark data set cover 8 SCOP classes, 199 folds, 249 super families and 389 families.
When more than one PDB entry with catalytic site annotation was available for a given family, one reference PDB entry was selected following the criteria: highest sequence coverage of the Pfam MSA, the year of structure determination (preferably later than 2000) and resolution (Supplementary table S1 provides the Pfam family and reference PDB). In all cases, MSAs were gap trimmed to remove positions with gaps in the reference sequence. In addition, all positions with >50% gaps, as well as sequences covering <50% of the reference sequence length were removed, as described in [29]. Supplementary figure S1 shows the distribution of the number of sequences and sequence clusters in the dataset.
Conservation of each position in the MSA's was calculated with three different measurements: Shannon entropy [34], KL relative entropy [35] calculated using an amino acids background frequency distribution obtained from the Uniprot database [36] and the maximal frequency (the frequency of the most represented amino acid). Each of these measurements were calculated from the raw MSA, from the MSA corrected for sequence redundancy using sequence weighting by 62% identity clustering (c), from the MSA including pseudo-counts to correct for low counts (l) [37], [38] and from the MSA applying both clustering and pseudo-count correction (cl). The total number of conservation measurements investigated was hence twelve.
Mutual information (MI) was calculated as described in [29]. In short, the MI is calculated between pairs of columns in the MSA. The frequency for each amino acid pair is calculated using techniques of sequences weighting and low count corrections and is compared to the expected pair-frequency assuming that the amino acids are non-correlated. Next, the MI is calculated as a weighted sum of the log-ratios between the observed and expected amino acids pair frequencies. The APC method of Dunn et al. [28] was applied to reduce the background mutual information signal for each pair of positions and the MI scores were finally translated into MI Z-scores by comparing the MI values for each pair of position to a large set of MI values calculated from permutated MSA. MI gives a value for each pair of residues in a MSA. We sought a mutual information score per residue that characterizes the extent of mutual information “interactions” in its physical neighbourhood. This score was defined in two steps. First, we calculated a cumulative mutual information score (cMI) for each residue as the sum of MI values above a certain threshold for every amino acid pair where the particular residue appears. This value defines to what degree a given amino acid takes part in a mutual information network. Next, we defined a proximity average for each residue as the average of cMI of all the residues within a certain physical distance to the given amino acid. Finally, we normalized the proximity average values for a given MSA to fall in the range [0–1] to obtain the proximity MI (pMI) score. The distance between each pair of residues in the structure was calculated as the shortest distance between any two atoms different from H belonging to each of the two residues.
We define a combined catalytic likeliness score (Cls) as a weighted sum of the conservation (defined in terms of the KL relative entropy), the proximity mutual information (pMI) and the proximity conservation (pC) scores.Here, pC is the average conservation score of residues within a given proximity distance, and wC, and wMI are adjustable relative weights.
The calculation of the combined catalytic likeliness score depends on three parameters; Zthr (Z-score threshold for including an amino acids pair in the cMI score), DMI (distance threshold to include an amino acid in the pMI average score), DC (distance threshold to include an amino acid in the pC average score), and the relative weights, wMI and wC, on pMI and pC, respectively. These parameters were estimated using five-fold cross validation, where optimal values were obtained using brute force grid-sampling on 4/5 of the data set to optimize the average AUC value and the remaining 1/5 of the data was evaluated next using this set of optimal parameters. This procedure was repeated five times leading to five sets of optimal parameters and evaluation performance values for each MSA in the data set.
The predictive performance in detecting catalytic residues, by way of conservation, pMI and Cls, was evaluated in terms of the area under the ROC curve (AUC) [39] per family. The AUC measure might not be optimal if the benchmark data set has a high ratio on negative data, and a high specificity in actual number could translate into a large number of false positive. In such situations, it might be beneficial to use only the high specificity part of the ROC curve to calculate the predictive performance. Here, we hence complement the AUC measure with AUC01 calculated including only the specificity range for 1 to 0.9 when calculating the AUC. For both measures will a value of 1 indicate a perfect prediction while a value of 0.5 indicates a random prediction. Annotated catalytic residues in the CSA were taken as the positive set, and all other residues with annotated PDB-ATOM coordinates were assigned as negative. The final performance was determined as the average AUC over the 434 CSA Pfam families.
We downloaded the entire Paradox SDR database (specificity-determining residues in protein families database; http://paradox.harvard.edu/sdr/), and identified the subset of families present in our benchmark dataset where the reference sequence from the CSA database was also member of the paradox multiple sequence alignment (MSA). This gave us a set of 158 families. The Paradox database provides SDR Z-scores only for a subset of the positions in the MSA [30]. Residues with undefined SDR Z-score were assigned a Z-score of 0 to allow for complete sequence coverage. Next, we compare for each family the SDR Z-score value to our cMI (cumulative mutual information) value of each position in the alignment in terms of the Spearmans rank correlation. We also calculate the Spearmans rank correlation between KL and both SDR Z-score and cMI values of each position for each family in the dataset.
|
10.1371/journal.pgen.1007397 | Auxin production in diploid microsporocytes is necessary and sufficient for early stages of pollen development | Gametophytic development in Arabidopsis depends on nutrients and cell wall materials from sporophytic cells. However, it is not clear whether hormones and signaling molecules from sporophytic tissues are also required for gametophytic development. Herein, we show that auxin produced by the flavin monooxygenases YUC2 and YUC6 in the sporophytic microsporocytes is essential for early stages of pollen development. The first asymmetric mitotic division (PMI) of haploid microspores is the earliest event in male gametophyte development. Microspore development in yuc2yuc6 double mutants arrests before PMI and consequently yuc2yuc6 fail to produce viable pollens. Our genetic analyses reveal that YUC2 and YUC6 act as sporophytic genes for pollen formation. We further show that ectopic production of auxin in tapetum, which provides nutrients for pollen development, fails to rescue the sterile phenotypes of yuc2yuc6. In contrast, production of auxin in either microsporocytes or microspores rescued the defects of pollen development in yuc2yuc6 double mutants. Our results demonstrate that local auxin biosynthesis in sporophytic microsporocytic cells and microspore controls male gametophyte development during the generation transition from sporophyte to male gametophyte.
| Plant life cycle alternates between the diploid sporophyte generation and the haploid gametophyte generation. Understanding the molecular mechanisms governing the generation alternation impacts fundamental plant biology and plant breeding. It is known that the development of haploid generation in vascular plants requires the diploid tapetum cells to supply nutrients. Here we show that the male gametophyte (haploid) development in Arabidopsis requires auxin produced in the diploid microsporocytic cells. Moreover, we show that auxin produced in microsporocytic cells and microspore is also sufficient to support normal development of the haploid microspores. This work demonstrates that Arabidopsis uses two different diploid cell types to supply growth hormone and nutrients for the growth of the haploid generation.
| Life cycle of eukaryotes alternates between haploid and diploid generations. The alternation of generations is initiated by meiosis (2n to 1n) and gamete fusion (1n to 2n) [1]. In land plants, the multicellular diploid generation is called sporophyte, whereas the multicellular haploid organism is named gametophyte. In bryophytes (mosses and liverworts), haploid gametophyte is the dominant generation and represents the main plant. In vascular plants, including ferns, gymnosperms, and angiosperms, the diploid sporophyte generation is dominant, whereas the gametophyte generation is much reduced [1]. For example, in seed plants, both the female and male gametophytes develop within the sporophyte. Understanding the molecular mechanisms governing the generation alternation will impact fundamental plant biology and plant breeding.
Pollen grains, which are the male gametophyte in seed plants, are developed in locules encircled by four sporophytic cell layers: tapetum, middle layer, endothecium, and epidermis. Inside a locule, a diploid male meiocyte divides into a tetrad of four haploid microspores after meiosis [2, 3]. Each microspore then undergoes an asymmetric cell division (pollen mitosis I (PMI)), resulting in two structurally and functionally different daughter cells: the small generative cell and the large vegetative cell. The generative cell divides one more time (PMII) to produce two sperm cells whereas the vegetative cell no longer divides. The mature pollen grain contains two haploid sperm cells and one haploid vegetative cell [3, 4].
Genetic analyses have identified a number of genes that play important roles in pollen development [5]. These genes can be classified into two categories: gametophytic or sporophytic genes. Pollen development depends on coordinated expression of both sporophytic genes and gametophytic genes [3]. Most of the identified gametophytic genes are related to cell division and differentiation. For example, MOR1, a member of the microtubule-associated protein 215 (MAP215), is involved in the PMI asymmetric cell division [6, 7]. In the mor1/gem1 mutants, defects in microspore nucleus migration lead to altered division plane, and the formation of two equal- or similar sized cells [6, 7]. Genes including Two-in-one (TIO), GAMMA-TUBULIN 1 (TUBG1) and 2 essential for phragmoplast formation, localization and/or expansion, also affect male gametophyte development. Mutations in these genes cause incomplete cytokinesis in PMI and result in failure to produce the generative cell [8–13]. Several cell cycle factors have also been reported to be important for pollen mitosis. ICK4/KRP6 (Interactors of Cdc2 kinase 4/kip-related protein 6) is a cyclin-dependent kinase inhibitor. Timely degradation of ICK6/KRP6 by RHF1a/2a and SCFFBL17 E3 ligase is essential for the progression of pollen mitosis [14–16]. Cyclin-dependent kinase D (CDKD) was recently found to be essential for pollen mitosis. In the cdkd;1–1 cdkd;3–1 double mutants, microspore is defective in both PMI and PMII [17]. A main feature of the male gametophytic genes is that no viable mutant pollens can be generated and that no homozygous diploid mutants are available.
Sporophytic genes for pollen formation represent the contribution of sporophytic tissues including tapetum and microsporocyte for male gametophyte development. Tapetum directly provides pollen wall materials and nutrients including Magnesium for pollen development [18–22]. A series of transcription factors including DYT1, TDF1, AMS, MYB33/65, MS188/MYB80/MYB103, and MS1 have been shown to play essential roles in normal development of tapetum [23–32]. The defective tapetum caused by mutations in these genes results in pollen wall defects and leads to complete pollen abortion [23–32]. Enzymes involved in outer pollen wall formation are also expressed in tapetum and are essential for pollen development [33–41]. The pollen wall pattern is determined inside a tetrad that depends on the sporophytic genes expressed in microsporocytes, such as RPG1, NPU, NEF and DEX1 [42–45]. The sporophytic cells including tapetum and microsporocyte supply material and nutrients for pollen development and determine the pollen wall pattern. Unlike gametophytic genes, viable mutant pollen can be produced from heterozygous mutant plants, and homozygous mutant plants can be obtained. However, homozygous diploid mutants cannot produce viable pollens.
Plant hormones are essential for normal plant development. However, it is not clear whether plant hormones or other signaling molecules produced in sporophytic tissues are required for the development of male gametophyte. The plant hormone auxin plays critical roles in nearly all aspects of plant development including embryogenesis, organogenesis, gametophyte development, and various tropisms [46]. It is known that disruption of either auxin biosynthesis, or polar transport, or signaling causes defects in male gametophyte development and pollen formation. Indole-3-acetic acid (IAA), the primary natural auxin in plants, is mainly synthesized through a TAA/YUC two-step tryptophan-dependent pathway [47–51]. Simultaneously disruption of both YUC2 and YUC6 completely eliminates the production of viable pollen grains [49]. It was reported that the two atypical members of the PIN auxin efflux carriers, PIN5 and PIN8, which are believed to regulate intracellular auxin homeostasis and metabolism in pollen, also participate in the development of normal pollen morphology [52, 53]. Other auxin transporters including ATP-BINDING CASSETTE B19 (ABCB19)/MULTIDRUG RESISTANCE PROTEIN 1 (MDR1)/ P-GLYCOPROTEIN 19 (PGP19) and ABCB1/ PGP1 also play important roles in pollen development [54, 55]. The abcb1 abcb19 double mutants show precocious pollen maturation [54, 55]. Similar precocious pollen maturation takes place in tir1 afb2 afb3 triple and tir1 afb1 afb2 afb3 quadruple mutants, which are defective in auxin perception [54]. It is known that auxin biosynthetic, transport, and signaling genes are expressed during pollen development [49, 54].
Previous studies have clearly demonstrated that auxin is required for anther development and pollen formation [49, 52–55], but it was not understood the role of auxin for pollen development and in the transition from sporophytic generation to gametophyte generation. It was previously proposed that auxin produced in tapetum is required for pollen development [56]. Here we report that pollen development in the auxin biosynthetic mutants yuc2yuc6 failed to progress past the PMI, which is an early step in male gametophyte growth. Our genetic analyses demonstrated that both YUC2 and YUC6 function as sporophytic genes. We further show that auxin produced in sporophytic microsporocytes rather than tapetum is required for the early stages of pollen development, demonstrating that auxin produced in the diploid sporophytic cells plays a critical role in the haploid male gametophyte development.
We previously showed that yuc2yuc6 double mutants were male sterile, and the expression of the bacterial auxin biosynthetic gene iaaM under the control of the YUC6 promoter fully restored the fertility of yuc2yuc6, indicating that the fertility defects of yuc2yuc6 were caused by partial auxin deficiency during anther development [49]. The auxin reporter DR5-GFP/GUS has been previously detected in anther from flower stages 10 to 12 [54, 57, 58]. It has also been reported that the DR5 activity is decreased in the yuc6 single mutant, although yuc6 did not display obvious reproductive defect [58]. To better understand the distribution patterns of DR5 in anthers of yuc2yuc6, we introduced the auxin reporter DR5-GFP into yuc2yuc6 plants and compared the GFP signals in yuc2yuc6 with those in WT plants. Consistent with previous findings [54, 57, 58], GFP signals in WT plants were detected in anthers from anther stages 9 to 12 (all the stages refer to anther stages in our results) (Fig 1A). The expression pattern of DR5-GFP in yuc2 and yuc6 anthers was similar to that in WT (S1 Fig). However, no substantial signals were detected in yuc2yuc6 anthers at the same developmental stages (Fig 1B). We also introduced the DR5-GUS into the yuc2yuc6 background. Similar to the DR5-GFP patterns, GUS signals were not detected in the yuc2yuc6 anthers (S1 Fig). Therefore, it appears that YUC2 and YUC6 are the main auxin biosynthetic genes responsible for auxin production during anther development.
Because we hardly observed any DR5-GFP signal in pollen, we used DII and mDII auxin reporter lines as a proxy to determine the pattern of auxin-induced degradation of Aux/IAA repressors during pollen development [59]. We replaced the 35S promoter with the microspore-specific promoter proMSP1 to drive the DII-VENUS and mDII-VENUS expression units [60]. Our results showed that fluorescence signals in microspores/pollens of proMSP1:DII-VENUS transgenic plants were weaker than that in proMSP1:mDII-VENUS transgenic plants at stage 10 (proMSP1:DII-VENUS/proMSP1:mDII-VENUS = 0.45) and at stage 11 (proMSP1:DII-VENUS/proMSP1:mDII-VENUS = 0.54)) (Fig 2A and 2B). The results of the auxin response reporters are indicative that auxin accumulated significantly in unicellular microspores and bicellular pollens.
To investigate whether signals of the auxin response reporters are correlated with the expression patterns of YUC2 and YUC6, we generated proYUC2-GFP and proYUC6-GFP transgenic plants. We found that both YUC2 and YUC6 were weakly expressed in microsporocytes, tetrad and microspores at stage 9 (S2A, S2B, S2C, S2I, S2J and S2O Fig) and strongly expressed in microspores from stages 10 to 13 (S2D–S2G and S2K–S2N Fig). We also found that YUC2 and YUC6 were strongly expressed in somatic cell layers including tapetum cells in anther (S2H, S2I and S2O Fig).
The yuc2yuc6 double mutants showed markedly reduced fertility with few viable pollen inside the locule [49, 54]. Pollination of the yuc2yuc6 pistil with WT pollen resulted in F1 plants with normal fertility (~50 seeds produced in each silique, n = 4), indicating that the female fertility of yuc2yuc6 was unaffected. Alexander staining was performed to understand the defects of pollen development in yuc2yuc6. Similar to those in wild type (WT), the anthers of yuc2 and yuc6 single mutants contained purple-stained viable pollen grains (Fig 3A–3C). Consistent with previous reports [49], Fig 3D showed that the yuc2yuc6 anthers did not produce viable pollens (Fig 3D). We then generated anther cross sections and used transmission electron microscopy to determine in which stage(s) the anther and pollen developmental defects took place in yuc2yuc6. We noticed that a normal tetrad was produced in yuc2yuc6 (Fig 3E and 3K), suggesting that the meiotic division progressed normally. After release from the tetrad, microspore development in yuc2yuc6 appeared similar to that of WT from stages 8 to 9 (Fig 3F, 3G, 3L, 3M, 3Q, 3R, 3V and 3W). However, at stage 10, WT microspores became vacuolated and contained a polarized nucleus (Fig 3H and 3S), whereas yuc2yuc6 microspores had an irregular shape and started to degenerate (Fig 3N and 3X). From stages 11 to 12, WT microspores underwent the first mitotic division and gradually developed into mature pollen (Fig 3I, 3J and 3T). In yuc2yuc6, microspores were severely degenerated and failed to form normal pollen (Fig 3O, 3P and 3Y). The pollen wall of yuc2yuc6 appeared normal as compared with that of WT (Fig 3U and 3Z).
To obtain further insight into the microgametogenesis defects of yuc2yuc6, we stained nuclei with 4’, 6-diamidino-2-phenylindole (DAPI) in developing pollen. The microspores of yuc2yuc6 were similar to WT microspores at stages 8 to 9 (Fig 4A, 4B, 4F and 4G). However, at stage 10, some of the yuc2yuc6 microspores were degenerated, with little DAPI staining signal (Fig 4H right). At this stage, some normal yuc2yuc6 microspores with a nucleus located at one side of the microspore were still visible (Fig 4C and 4H left). Overall during the unicellular stage (from stages 8 to 10), most of the single haploid cells in both the WT (92.7%) and yuc2yuc6 (67.9%) showed a bright nucleus (Fig 4A–4C, 4F–4H and 4K), and about 30% of the microspores in yuc2yuc6 were degenerated (Fig 4H right, 4K). After PMI, 91.1% of the WT microspores divided asymmetrically, producing a large vegetative cell and a small generative cell engulfed in the cytoplasm of the vegetative cell (Fig 4D and 4K), which is called the bicellular stage. In contrast, only about 2.2% of the yuc2yuc6 microspores progressed past PMI to reach the bicellular stage (Fig 4K). Approximately 44.8% of the microspores in yuc2yuc6 remain arrested at the unicellular stage (Fig 4I left, 4K), and the rest (53%) became degenerated (Fig 4I right, 4K). At the tricellular stage, WT microspores fully developed into tricellular pollen, whereas yuc2yuc6 microspores (96%) became completely degenerated or still contained a very loosely packed DNA mass (Fig 4E, 4J and 4K). Therefore, we conclude that mutations in YUC2 and YUC6 lead to the defects in early stages including PMI of pollen development.
We performed genetic complementation to determine whether mutations in the YUC genes are responsible for the sterile phenotype of yuc2yuc6. The DNA fragment including about 2 kb upstream sequence of YUC2 and the YUC2 open reading frame (ORF) was fused with GFP (named proYUC2:YUC2-GFP) and the construct was introduced into yuc2-/- yuc6+/- plants (S3 Fig). Two independent lines of yuc2yuc6 that contained the proYUC2:YUC2-GFP transgene were identified in T1 generation. Both lines of proYUC2:YUC2-GFP in the yuc2yuc6 background had normal fertility (S3B Fig). Alexander staining and DAPI staining results indicated that YUC2 could rescue the pollen development defects of yuc2yuc6 (S3C and S3D Fig and S5 Fig). The results also demonstrated that the YUC2-GFP fusion is functional.
From our phenotypic analysis, it was clear that the yuc2yuc6 double mutants were defective in gametophyte development. Gametophyte development depends on the expression of both sporophytic and gametophytic genes. To determine whether YUC2 and YUC6 function as sporophytic or gametophytic genes, we analyzed the male transmission efficiency of yuc2yuc6. We used yuc2-/-yuc6+/- and yuc2+/-yuc6-/- plants as pollen donors to cross with WT plants. PCR genotyping revealed that approximately 50% of F1 progeny contained the yuc2yuc6 mutations (Table 1), suggesting that yuc2yuc6 microspores were transmitted normally. We next analyzed the segregation ratio of self-fertilized yuc2-/-yuc6+/-and yuc2+/-yuc6-/- plants. Consistent with the normal transmission efficiency, the segregation displayed a typical Mendelian ratio (3:1) in both cases (n>290 for each case). These results showed that YUC2 and YUC6 are sporophytic genes for pollen development, indicating that auxin produced in the sporophytic tissues by the YUCs is required for normal male gametophyte development.
It is known that YUC2/6 mRNA is expressed in meiocytes, microspores, tapetum, middle layer, and endothecium in anthers [54]. To investigate the source of auxin for pollen development, we used various promoters to drive the expression of YUC2-GFP fusion, which we have shown functional (S3 Fig). The DR5 auxin reporter line showed an extremely active auxin response in the tapetum cells during late developmental stages in Arabidopsis [54, 56, 58]. It was proposed that auxin is transported from tapetum cells into developing pollens [56]. To investigate whether the YUC genes expressed in tapetum are responsible for regulating microspore development and pollen formation, we generated transgenic lines that express YUC2-GFP in tapetum cells using specific tapetum promoters (proA9 and proATA7 (ARABIDOPSIS THALIANA ANTHER7)) in the yuc2yuc6 background [32, 61, 62] (Fig 5 and S4 Fig). We found that both proA9: YUC2-GFP (yuc2yuc6) (n = 6) and proATA7:YUC2-GFP (yuc2yuc6) (n = 7) T1 transgenic plants were still sterile (Fig 5B and S4 Fig). The pollen defects in yuc2yuc6 were not rescued by the YUC2-GFP transgene (Fig 5C, S4 and S5 Figs). We investigated whether the tapetum-specific promoters behaved as designed. RNA in situ hybridization data showed that YUC2-GFP was significantly transcribed in tapetum cells from microsporocytes stage to early microspore stage in proA9:YUC2-GFP (yuc2yuc6) plants (Fig 5D). The GFP signal in proA9:YUC2-GFP (yuc2yuc6) plants appeared in the tapetum layer at stages 8 and 9 (Fig 5E showed stage 9). Although the GFP fluorescence of proATA7:YUC2-GFP (yuc2yuc6) plants could not be detected (S4 Fig), the YUC2-GFP transcripts were detected in tapetum layer at stage 8 (S4 Fig). To rule out the possibility that the ATA7 promoter was too weak to drive adequate expression of YUC2-GFP in anther, we used real-time quantitative RT-PCR to analyze the transcript levels of YUC2 in WT and YUC2-GFP in the transgenic plants. The expression levels of YUC2-GFP in all of the analyzed transgenic plants were similar to or higher than YUC2 expression in WT (S5 Fig). These results suggest that the ATA7 and A9 promoters were able to drive YUC2-GFP expression in tapetum, but auxin production in tapetum is not sufficient to overcome the auxin deficiency in yuc2yuc6 microspores.
We next used promoters specific for microsporocytes and microspores (proARF17) [63] to drive the expression of YUC2-GFP in yuc2yuc6 plants (Fig 6). All of the 5 independent T1 proARF17:YUC2-GFP (yuc2yuc6) lines showed almost complete rescue of the sterile phenotypes of yuc2yuc6 (Fig 6A). In addition, the pollen defects were fully rescued in the transgenic plants (Fig 6A and S5 Fig). In the transgenic plants, YUC2-GFP expression was observed in microsporocytes and was significant in early stages of microspores (Fig 6A). Fig 6A also showed that the YUC2-GFP protein accumulated in early stages of microspores. Because ARF17 promoter drives gene expression in both microsporocytes and microspores, we further tested whether localized auxin biosynthesis in microspores is sufficient to rescue yuc2yuc6. We used the promoter proLAT52, which is specifically activated in male gametophyte [14, 64], to drive the expression of YUC2-GFP in yuc2yuc6 plants (Fig 6B). All of the 6 independent T1 proLat52:YUC2-GFP (yuc2yuc6) lines showed partial rescue of the fertility defects of yuc2yuc6 (Fig 6B). The proLat52:YUC2-GFP (yuc2yuc6) plants were fertile at a late reproductive development stage (Fig 6B). During late reproduction development, 30% to 70% of the pollen in the transgenic lines appeared normal in the anthers (Fig 6B and S5 Fig). At bicellular stage, 70.7% of the unicellular pollen can develop into bicellular pollen, and 18% of the microspores are arrested at unicellular stage. At tricellular stage, proLat52:YUC2-GFP (yuc2yuc6) produced 56.9% tricellular pollens, and 31.8% pollens were aborted (Fig 6B and S5 Fig). These results indicated that proLat52:YUC2-GFP could partially support the microspores development past PMI and PMII. In the proLat52:YUC2-GFP (yuc2yuc6) transgenic plants, YUC2-GFP was transcribed from late stages of microspores (Fig 6B). Meanwhile, YUC2-GFP protein was detected in pollen at stage 13 (Fig 6B). The observation that auxin produced in microspores partially rescued the sterile phenotype of yuc2yuc6 suggested that early stages of pollen development including PMI of unicellular microspores require auxin. The sterility rescue efficiencies of proARF17:YUC2-GFP (yuc2yuc6) and proYUC2:YUC2-GFP (yuc2yuc6) (S3 Fig) were significantly higher than that in proLat52:YUC2-GFP (yuc2yuc6) transgenic plants. We noticed that the expression of YUC2-GFP or GFP was at earlier stage of microspores in proARF17:YUC2-GFP (yuc2yuc6) and in proYUC2:GFP (S2 Fig) than that in proLat52:YUC2-GFP (yuc2yuc6) transgenic plants (Fig 6). Therefore, it is likely that auxin may be required at early phase of microspore development. Combined with the genetic data that YUC2 and YUC6 act as sporophytic genes, we conclude that the auxin synthesized in the sporophytic microsporocytes is essential for early stages of pollen development in plants. Therefore, auxin produced in sporophyte contributes to male gametophyte during the generation alternation in plant.
Both YUC2 and YUC6 are known required for pollen development and the yuc2yuc6 double mutants are male sterile. Here we further defined that the male sterility of yuc2yuc6 is caused by defects in early stages of pollen development including the first mitotic cell division (PMI) of microspores. Moreover, we show that early stages of microspore development and PMI require auxin produced in the diploid sporophytic microsporocytes, indicating that sporophytic cells not only provide nutrients and cell wall materials for pollen development, but also supply hormone and signaling molecules to haploid cells. Our results also demonstrate that different sporophytic cells play different roles in male gametophytic development. Tapetum cells provide nutrients, but auxin produced in tapetum cells is not sufficient to support early stages of pollen development. In contrast, auxin synthesized in sporophytic microsporocytes is necessary and sufficient for male gametophytic development.
Because yuc2yuc6 double mutants could undergo meiosis successfully and our results showed that supply of auxin after meiosis could partially rescue the pollen defects of yuc2yuc6 (Fig 6, ProLat52:YUC2-GFP (yuc2yuc6)), we conclude that auxin produced by YUC2 and YUC6 may not be required for male meiosis. Following meiosis, the microspore undergoes two rounds of mitosis to form mature pollen during anther development. PMI is the first round of mitosis for pollen formation. Several lines of evidence show that auxin produced by YUCs is essential for PMI. We showed that YUC2 and YUC6 are the main enzymes responsible for auxin synthesis in anther based on the expression of the auxin reporter DR5:GFP/GUS (Fig 1 and S1 Fig). Our results are consistent with previous studies showing DR5 activity in anthers possibly resulting from auxin synthesis [54]. Secondly, we found that the microspores of yuc2yuc6 were aborted and degenerated before PMI and failed to form mature pollen (Fig 3 and Fig 4). Consistent with this phenotype, we found significant AUX/IAA degradation, which is presumably induced by auxin, in unicellular microspores judging from the expression of the modified DII-Venus reporter (Fig 2). Lastly, we revealed that YUC2-GFP expressed in microspores at early gametophyte stages could partially rescue the yuc2yuc6 pollen defect (Fig 6, ProLat52:YUC2-GFP (yuc2yuc6)). These results demonstrated that auxin is required for early stages of pollen development. Because most of the microspores in yuc2yuc6 were arrested prior to PMI, we could not exclude the possibility that auxin is also required in subsequent steps such as PMII. Auxin regulates plant development mainly through auxin response factors (ARFs). Among the ARF genes, ARF6, ARF8, ARF10, ARF16, and ARF17 are expressed in unicellular microspores [65]. The arf17 mutant shows defective pollen formation [63], suggesting that auxin synthesized by YUCs may control male gametophyte development through ARF17. However, we noticed that pollen development defects in arf17 and yuc2yuc6 were not the same. The arf17 mutants displayed defective pollen coat formation whereas it was not the case for yuc2yuc6. Therefore, we believe that ARF17 may have unique functions in the earlier stages of pollen development that do not overlap with those of YUC2 and YUC6, such as controlling of the morphogenesis of pollen wall.
It was reported that pollen development in tir1 afb1 afb2 afb3 quadruple mutants was not significantly compromised, an observation that was quite different from the severe pollen defects found in yuc2yuc6. It is puzzling that disruption of auxin signaling and auxin biosynthesis resulted in different phenotypes. It was known that some of the tir1 afb1 afb2 afb3 plants can survive and produce seeds, but auxin transport mutants such as pin1 and auxin biosynthesis mutants including yuc1yuc4 were completely sterile [49, 66]. It is likely that other AFB proteins such as AFB4 and AFB5 may compensate the loss of tir1 afb1 afb2 afb3. In addition, we noticed that afb1 and afb3 in the quadruple mutants still produced truncated transcripts and might have produced residual protein activities [67]. The fact that tir1 afb1 afb2 afb3 were not a complete null in auxin receptor may account for the observed paradoxical results. AFB5 is expressed in bicellular pollens (S1 Table). Analysis of higher order tir1 afb mutants may clarify why auxin signaling mutants behaved differently from auxin biosynthesis and transport mutants.
Most of the known functions of auxin in flowering plants are related to sporophyte generation, which is the dominant generation. The auxin pathways are evolutionary conserved in the plant kingdom. The homologs of auxin biosynthetic genes such as TAAs and YUCs are also used for auxin biosynthesis in moss Physcomitrella patens and Liverwort Marchantia polymorpha [68, 69], whose haploid gametophyte is the dominant generation. In liverwort, auxin synthesized by YUCs is essential for normal gametophyte development and dormancy [69], implying that auxin may also play essential roles in gametophyte generation in flowering plants. However, it is difficult to study the roles of auxin in male gametophyte development because male gametophyte has been reduced to several cells within the diploid sporophyte. Our detailed analysis of the auxin biosynthetic mutants yuc2yuc6 revealed that auxin is required for male gametophyte development.
More importantly, we show that auxin produced in microsporocytes, which are sporophytic cells, is necessary for normal progression of the haploid microspores. Our genetic data suggest that YUC2 and YUC6 affect early stages of pollen development mainly via a sporophytic effect, implying that auxin required for pollen development may be synthesized by YUC2 and YUC6 in sporophytic anther tissues. These data are consistent with the transcription patterns of YUC2 and YUC6 in microsporocytes, microspores and anther somatic cell layers at premeiotic and meiotic stages [54]. In anther, tapetum cells and microsporocytes are closely related sporophyte cells involved in microspore/male gametophyte development. It was proposed that auxin could be transported to developing pollen from tapetum cells [56]. However, the expression of YUC2 in the tapetum cell of yuc2yuc6 could not lead to viable pollen (Fig 5 and S4 Fig), suggesting that the auxin from tapetum is not sufficient to support pollen development. On the other hand, the expression of YUC2-GFP in microsporocytes completely rescued the sterility phenotypes of yuc2yuc6 (Fig 6), demonstrating that auxin synthesized in microsporocytes is sufficient for pollen development.
We propose that auxin production in diploid microsporocytes is necessary and sufficient for the early stages of the development of the haploid cells during pollen development. Our conclusion is mainly based on two observations: 1) YUC2 and YUC6 are sporophytic genes and the yuc2yuc6 fail to produce viable pollen; 2) expression of YUC2 driven by specific promoters rescued yuc2yuc6 (Fig 6). The ARF17 promoter is active in both microsporocytes (sporophytic) and microspores (gametophytic) (Fig 6). Therefore, it is conceivable that the rescued yuc2yuc6 phenotypes by YUC2-GFP driven by ARF17 promoter were caused by ectopic auxin production in the haploid microspores. However, from our genetic analysis, it is clear that viable and fully functional pollen that lacks YUC2 and YUC6 can be produced from yuc2-/-yuc6+/- or yuc2+/-yuc6-/-, demonstrating that auxin does not have to be produced in microspores for the development of viable pollen. Rather auxin produced by YUCs in sporophytic cells (microsporocytes to be exact) is sufficient to guide the progressive development of microspores.
Microsporocytes developing into mature pollen through PMI and PMII is a continuing process. Production of auxin in microsporocytes is an efficient way to regulate newly formed microspores to undergo PMI and other processes of early male gametophyte development. Using two different diploid cell types (tapetum and microsporocytes) to provide nutrients and hormonal signals for male gametophyte development may also be advantageous in terms of efficiency and specificity.
All plants used in this study were in the Columbia-0 genetic background. The yuc2yuc6 mutants have been described previously [49]. All relevant primer sequences were listed in S2 Table. The DII-VENUS (N799173) and mDII-VENUS (N799174) seeds were ordered from the European Arabidopsis Stock Centre (NASC). Plants were grown under long-day conditions (16 hr light/8 hr dark) in a ~22°C growth room. The DR5:GFP and DR5:GUS reporter lines were introduced into yuc2yuc6 by genetic cross.
Plants were photographed using a Nikon digital camera (D-7000). Alexander solution was prepared as previously described [70]. Anthers were dissected and immersed in Alexander solution for 0.5 hr, and images were obtained under a microscope with an Olympus BX51 digital camera (Olympus, Japan). Plant materials for the semi-thin sections were prepared and embedded in Spurr resin as described in [25] and cut into 1-μm thick sections, stained with toluidine blue, then photographed with an Olympus BX51 digital camera. For transmission electron microscopic analysis, the same-stage anthers of wild type (WT) and yuc2yuc6 were fixed and embedded as previously described [25]. Green fluorescent protein (GFP) fluorescence in ProARF17:YUC2-GFP (yuc2yuc6), ProLAT52:YUC2-GFP (yuc2yuc6), ProMSP1:DII, and ProMSP1:mDII transgenic plants was detected under a fluorescence microscope (Olympus BX51). GFP fluorescence in ProA9:YUC2-GFP (yuc2yuc6) transgenic plants and WT or yuc2yuc6 expressing DR5:GFP was detected by confocal laser scanning microscopy (Carl Zeiss, LSM 5 PASCAL).
To generate YUC2-GFP, we amplified YUC2 cDNA without the stop codon from inflorescence mRNA by standard RT-PCR (see S2 Table for primers) for cloning into the modified pCAMBIA1300 binary vector (CAMBIA, Australia), pCAMBIA1300-GFP. The resulting construct was named pCAMBIA1300-YUC2-GFP. The promoter sequences of YUC2, YUC6, ARF17, ATA7 and A9 were PCR-amplified from genomic DNA of WT Col-0. The promoter sequence of LAT52 was PCR-amplified from genomic DNA of tomato. The primers for amplification were listed in S2 Table. The amplified sequences were first inserted into pMD19-T (Takara) for sequence verification, then subcloned into the vector pCAMBIA1300-YUC2-GFP for plant transformation. The promoter sequences for YUC2 and YUC6 were subcloned into pCAMBIA1300-GFP for plant transformation. The promoter sequence for MSP1 was PCR-amplified from genomic DNA of WT Col-0, then subcloned into the vector pCAMBIA1300 to obtain pCAMBIA1300-proMSP1. The DII-VENUS and mDII-VENUS sequences were amplified from genomic DNA from DII-VENUS and mDII-VENUS plants. The amplified sequences were inserted into pMD19-T (Takara) for verification, then subcloned into the vector pCAMBIA1300-proMSP1 for plant transformation. For plant transformation, all plasmids were introduced into the Agrobacterium strain GV3101 and transformed into plants by the floral dip method[71]. For tissue specific rescue experiments, all of the constructs were transformed into the offspring of yuc2-/-yuc6+/-. After collected the T1 seeds, we first selected transgenic plants by growing on MS media containing hygromycin. We then genotyped the transgenic plants for yuc2yuc6 double mutants[49].
The probe fragments were amplified from plasmid containing GFP (pCAMBIA1300-GFP) with primers GFP-F and GFP-R. The PCR products were cloned into the pBluescriptSK vector and confirmed by sequencing. Plasmid DNA was digested with HindIII or BamHI. The digestion products were used as templates for transcription into sense and antisense probes by T3 and T7 RNA polymerases, respectively (Roche). Oligonucleotide sequences of GFP-F and GFP-R are provided in S2 Table. Images were taken using the Olympus BX-51 microscope.
Total RNA was extracted from the inflorescences of T1 transgenic plants by the Trizol method (Invitrogen, USA) following the manufacturer’s instructions. Quantitative real-time PCR involved an ABI PRISM 7300 detection system (Applied Biosystems, USA) with SYBR Green I master mix (Toyobo, Japan). Relevant primer sequences are in S2 Table. β-Tubulin was as a constitutive expression control. Three biological repeats were used for gene expression analysis.
|
10.1371/journal.pcbi.1004597 | Distinctive Behaviors of Druggable Proteins in Cellular Networks | The interaction environment of a protein in a cellular network is important in defining the role that the protein plays in the system as a whole, and thus its potential suitability as a drug target. Despite the importance of the network environment, it is neglected during target selection for drug discovery. Here, we present the first systematic, comprehensive computational analysis of topological, community and graphical network parameters of the human interactome and identify discriminatory network patterns that strongly distinguish drug targets from the interactome as a whole. Importantly, we identify striking differences in the network behavior of targets of cancer drugs versus targets from other therapeutic areas and explore how they may relate to successful drug combinations to overcome acquired resistance to cancer drugs. We develop, computationally validate and provide the first public domain predictive algorithm for identifying druggable neighborhoods based on network parameters. We also make available full predictions for 13,345 proteins to aid target selection for drug discovery. All target predictions are available through canSAR.icr.ac.uk. Underlying data and tools are available at https://cansar.icr.ac.uk/cansar/publications/druggable_network_neighbourhoods/.
| The need for well-validated targets for drug discovery is more pressing than ever, especially in cancer in view of resistance to current therapeutics coupled with late stage drug failures. Target prioritization and selection methodologies have typically not taken the protein interaction environment into account. Here we analyze a large representation of the human interactome comprising almost 90,000 interactions between 13,345 proteins. We assess these interactions using an extensive set of topological, graphical and community parameters, and we identify behaviors that distinguish the protein interaction environments of drug targets from the general interactome. Moreover, we identify clear distinctions between the network environment of cancer-drug targets and targets from other therapeutics areas. We use these distinguishing properties to build a predictive methodology to prioritize potential drug targets based on network parameters alone and we validate our predictive models using current FDA-approved drug targets. Our models provide an objective, interactome-based target prioritization methodology to complement existing structure-based and ligand-based prioritization methods. We provide our interactome-based predictions alongside other druggability predictors within the public canSAR resource (cansar.icr.ac.uk).
| Identifying novel drug targets and prioritizing proteins for target validation and therapeutic development are essential activities in modern mechanism-driven drug discovery, and are key if we are to benefit from large-scale genomic initiatives [1]. Multiple approaches exist to estimate the ‘druggability’ or chemical tractability of a protein [2–4]. 3D structure-based assessments predict cavities in the protein structure that are capable of binding small molecules [3–5]. Alternative methods include sequence feature-based druggability [4,6] and ligand-based methods that examine the properties of compounds known to be bioactive against a protein [7–9].
While many genes have been identified as disease-causing (see for example reports on cancer [10,11] and cardiovascular disease [12]), the products of relatively few of these have become targets for approved therapeutics. The challenges facing researchers attempting to target a gene and its product proteins for clinical application lie both in validating their pathogenic role and in their technical ‘doability’. As well as possessing a pocket or interface suitable for drug binding, a potential drug target must exert an appropriate influence on the system, enabling a drug to have a selective and enduring therapeutic effect. Genetic diseases, prominently cancer, are disorders arising from deregulation or disruption of normal cellular wiring and protein communication. It is therefore essential that the network environment of a potential drug target should be incorporated into target selection rationale.
Previous studies have highlighted the importance of considering the interactome when predicting protein function [13,14], assessing drug-target interaction data and understanding polypharmacology [9,15], or predicting novel uses for drugs [16–18]. Meanwhile, recent technological advances in systems biology have generated large quantities of experimentally-derived protein interaction data [19] and networks have been applied to understand the relationships between these protein interactions and disease [20–24]. For example, relationships between protein interactions and cancer have been identified by integrating protein interaction networks with functional or gene expression data [25,26]; structural differences in the network between cancer-causing and non-cancer-causing genes have been highlighted [24–26]; and a potential core ‘diseasome’ network has been documented [27].
Tantalizingly, a number of studies have examined the distribution of some focused topological network parameters, such as degree and clustering coefficient, in drug targets versus non-drug targets [17,18,28]. Most notably, the number of first neighbors (degree) was identified as a distinguishing feature of the human ‘highly optimized tolerance’ or ‘HOT’ network [17] and was proposed as a measure to consider when selecting drug targets. This proposition was based on the assumption that inhibiting proteins with a high degree will impact widely on a biological system and thus have undesirable effects [17]. While such extrapolations may not always hold true—for example, many cancer-drug targets are major hubs yet their modulation, singularly or in combination, shows clear selectivity for cancer cells (see references [29,30] and discussed below)—these studies have highlighted the potential of network parameters to provide discriminatory patterns for identifying druggable network nodes. However, such studies imply that any patterns that may exist to distinguish drug targets from other proteins are likely to be complex. Indeed, no purely network-based discriminatory models have been described. Instead, reported models include functional or family annotation [4,6,17,18,28,31] to ensure predictive power. These functional and family annotations overshadow any network parameters due to the dominance of certain protein families (e.g. G-protein coupled receptors) or functions (e.g. enzymes) in the training set of known drug targets [7–9,32].
Thus, the true network behaviors that may distinguish the points in a network most suitable for therapeutic intervention remain elusive. Consequently, despite the fundamental role of cellular wiring in drug action and resistance, there is, to our knowledge, no network-based druggability predictor in existence in the public domain. In this article, we present a comprehensive and systematic computational analysis of 321 topological, community-based and graphical network properties of a fully-connected human interactome. Furthermore, we show how these properties relate to druggability in its more complete sense: the suitability for intervention with a molecularly targeted therapeutic agent of any type. In particular, we explore the differing network environments of cancer-drug targets and targets from other therapeutic indications and discuss the potential impact that these differing network environments may have on resistance to cancer drugs. We build and benchmark the first publicly available predictive network-based models to identify likely druggable network nodes and node clusters, and apply these models to a set of 13,345 proteins in the human interactome. To our knowledge, these are the first published models that enable a prediction of druggability based on the topological, graphical and community behavior of proteins in the interaction network. Our network method is intended to be used to complement structure-, sequence- and ligand-based druggability prediction methods in order to provide a holistic view of the likely utility of a given protein as a drug target. The results of this analysis have been implemented in the canSAR knowledgebase [10,11,33,34] and each protein’s network signature, as well as its predicted druggability score, is accessible alongside other druggability measures at http://cansar.icr.ac.uk/.
Having constructed the high quality interactome, we constructed four distinct, manually curated training sets representing: a) all targets of FDA-approved drugs, which in turn was divided into b) targets of cancer drugs and c) targets of drugs from non-cancer therapeutic areas; and finally d) cancer proteins—the products of cancer associated genes. We then trained suites of predictive models using each of these training sets to predict druggability using only network topological, community and graph features (see Methods and Fig A in S1 Text).
Our predictive models achieve a mean area-under-the-curve (AUC) value of 83% (see also Fig E in S1 Text for recall precision). The significance compared to random prediction is as follows: drug targets (all therapeutics areas), p-value < 2.0−16 compared to 0.018 for the randomized network model; cancer drug targets, p-value = < 2.0−16 compared to 0.125 for the randomized network model and non-cancer-drug targets, p-value = < 2.0−16 compared to 0.331 compared to the randomized network model. Thus, the models have high predictive power using extensive in silico validation and can, therefore, be used to enrich potential drug targets during target prioritization for drug discovery. The full results are provided in S1 Table and per-protein analyses are supplied within the canSAR resource (https://cansar.icr.ac.uk/) alongside previously described structure-based and ligand-based methodologies [4,12,34].
We found that several network parameters show distinct distributions in drug targets or targets of cancer versus non-cancer drugs (key parameters are shown in Fig C in S1 Text). On average, drug targets have a higher degree (i.e. more first neighbors) than non-drug targets. Whilst the mean degree of drug targets is 26.34, this is primarily due to cancer targets which have a mean degree of 47.21 (compared to 13.72 for targets from other therapeutic areas (TAs) and 12.65 for the background—see Table C in S1 Text). We also found that targets of cancer therapeutics have more neighbors and tend to be more hub-like than the average cancer-associated proteins (Table C in S1 Text).
In fact, considering the interactome as a whole, out of the 50 proteins with the highest number of interactions, only six (SRC, EGRF, ESR1, AR, HDAC1 and FYN) are drug targets, all of which are targeted by cancer drugs. This indicates that a large number of first neighbors are not generically associated with being a drug target, but rather the average is skewed by a few highly connected cancer-drug targets.
Network articulation points are nodes that are critical for communication within the network and their removal would disconnect the network into separate graphs or break off peripheral nodes into unconnected singletons. Our analysis shows that 15% of all drug targets, 17% of cancer-drug targets and 14% of non-cancer drug targets are articulation points as compared to 9% of the background set (Table C in S1 Text). However, this enrichment is more statistically significant for all drug targets (p-value = 0.0003) and targets of cancer drugs (p-value = 0.0026) than for targets of non-cancer drugs (p-value = 0.0192). Being an articulation point is akin to an ‘ambassador’ between regions of the network and is a property that is enriched in cancer-drug targets. Interestingly, most articulation points from the cancer-drug target set are through nuclear hormone receptors (NHRs) and receptor tyrosine kinases (RTKs), which are logical gateways for signaling. We also found that cancer targets are more embedded in their local environment than targets from other therapeutic areas (using Burt’s network constraints [13,14,35] and closeness centrality [9,15,36]; Fig C in S1 Text). In summary, our analysis of 28 topological parameters indicates that there are distinguishing patterns of behavior between 1) cancer-drug targets; 2) targets of non-cancer drugs; and 3) the background interactome as a whole. This indicates that the topological parameters can be used as useful features in a predictive model for ab initio identification of drug targets for cancer or for non-cancer drugs.
A community within a network is defined a set of nodes that are densely connected within subsets of the full interactome (see Methods) but may not all interact directly with each other [16–18,37]. Proteins in a community may be linked together via a function, such as belonging to a particular cellular process. Previous studies have shown the relationship between biological function and network communities [19,37,38]. Drug targets from non-cancer therapeutic areas tend to be members of smaller communities compared to other proteins (Fig C in S1 Text). Interestingly, cancer-associated proteins participate in significantly larger communities, indicating the far-reaching effects of biological malfunctions in this class. Furthermore, cancer-drug targets differ from non-cancer-drug targets when considering their community pattern of interactions. To assess the type of community interactions that a protein is involved in we developed a vertex modularity score based on the proportion of interacting neighbors that are in the same community (see Methods). We find that non-cancer-drug targets tend to interact intra-community, whereas cancer-drug targets, interact both intra- and inter-community. This indicates that while targets of non-cancer drugs address specific functions and defined processes, cancer-drug targets may have wider reaching effects on different cellular functions. This pattern holds equally true for targets of classical cytotoxic cancer drugs, such as tubulin, as for the modern class of cancer genome-targeted cancer drugs, such as kinase inhibitors (Fig D in S1 Text).
Complex networks can be divided into smaller sub-graphs, or graphlets, of increasing complexity [39], (see Methods and Fig B in S1 Text). We find a striking difference in the behavior of cancer-drug targets as compared with targets of non-cancer drugs (Fig 1). Not only are cancer-drug targets significantly more active in graphlets (on average involved in 368 million target-graphlet activities compared to 121 million activities for non-cancer targets), but they are also more commonly seen in complex graphlets (such as G26, G27, G28 and G29) than can be expected at random (Fig 1). Importantly, while targets of cancer therapeutics are enriched in these graphlets when normalized against the interactome background, we find that targets of other therapeutic areas are, in contrast, slightly depleted or show little change from background. On average cancer-drug targets have more publications per node than non-cancer drug targets (with 39 versus 11 publications per node) and both sets are better studied than the background interactome (which has an average of 7 publications per node). Approved cancer drugs primarily target a single functional sites on a single type of protein (such as the catalytic site of a kinase and cytochrome P450s or the hormone binding sites on a hormone receptor). Currently most drugs that target heteromeric complexes fall in non-cancer areas (such as the ligand-gated ion channel blockers used to treat disorders of the central nervous system). The graphlet enrichment pattern that we have described here may be due to cancer targets being members of more transient signaling cascades or transcriptional complexes.
Thus far, we have discussed some of the parameters that most obviously differentiate the target sets (cancer targets, non-cancer targets, the background interactome). In order to uncover which network features play the most important role we examined the feature contributions for each of the models generated (see S1 File). We find that no single feature is sufficient for discrimination between the target sets. For example, for the GBM models, the range of maximum relative information carried by any one feature was between 4.78% for the all-drug-target model and 5.86% for the non-cancer target model. Similarly, the maximal mean standard error (%MSE) effect of any one feature for the random forest models was 0.02% across all models. Interestingly, although the top features reported by the different models vary, community and graphlet-based features dominate the list of top 20 highest features produced by all the models, while topological features rank lower.
Of the 343 drug targets in our interactome, 310 (90%) ranked in the top 25% for druggability according to the overall drug target model. This is a 3.6-fold enrichment of drug targets compared with what might be expected if proteins were ranked at random. In addition to the 343 targets of already FDA approved drugs, a further 3,026 currently undrugged proteins in this top-quartile most druggable set (see S1 Table), highlighting them as potentially suitable for drug discovery. The lowest ranking was CYP17A1, the molecular target of abiraterone with a rank of 49%. Additionally, we examined how our network-based assessment of druggability performed in relation to several targets of drugs that are under clinical investigation. We examined targets from different protein families and molecular classes. We found that, despite not having representatives of their families, TLR7, BCL2, EZH2, and MDM2 for example, scored highly using the druggability models (74% or higher using the all-drug-target model; and 85% or higher using the cancer-drug-target model). Full results for seven targets of investigational therapeutics are shown in Table F in S1 Text.
To examine the persistence of the signal we investigated the predictive models and target coverage across different datasets. First, we explored the effect of utilizing large-scale Yeast2Hybrid (Y2H) data instead of compiling all high-quality binary interaction data from different sources. Although the Y2H technique is unbiased, we found many interactions were missing from the interactome. This is probably because the full matrix of bait-and-prey proteins has not yet been fully examined. We describe this analysis in detail under the section ‘Defining the interactome’ in S1 Text. In summary, we compiled a large Y2H interactome by collecting Y2H data from 5,537 publications, including 30 publications reporting at least 70 proteins. This resulted in an interactome containing 10,998 proteins and 47,994 interactions–covering 256 of the 345 drug targets in the training set. To complement this we compiled a more comprehensive interactome containing all Y2H studies and high quality data from multiple sources (see Methods). This resulted in an interactome containing 13,345 proteins and 89,691 interactions and covered 343 of the 345 drug targets in the training set. As well as missing 26% of the drug target training set, the large Y2H interactome was missing many known interactions which should have been detected using the methodology. For example, despite MTOR and its complex components such as FKBP1A and DEPTOR being nodes in the Y2H studies, no interactions between them have been reported so far, despite these interactions being experimentally validated outside of Y2H studies [40]. We found that the predictive power of the models is stronger when network properties were calculated using the full interactome rather than the Y2H interactome (See Fig G in S1 Text). We detail all models and prediction results in the Supporting Information.
Additionally, we defined non-redundant versions of the training sets based on protein sequence similarity, drug chemical similarity and therapeutic class similarity (see Methods). Again, we found that models built using these training sets underperformed in comparison with models built on the full training set (Fig H in S1 Text and section ‘Further Information’).
To identify potential novel drug targets, we collated the top 20 proteins, that are not targets of approved pharmaceuticals and are predicted to be druggable using each of our three models (removing any duplicates). This resulted in a set of 49 proteins shown in S2 Table with their rank from each model and any known link to a disease (using the Online Mendelian Inheritance in Man, OMIM [41] and the Cancer Gene Census [27]). The distinctions in the prediction ranks illustrate the significant differences between drug targets for cancer and drug targets for non-cancer diseases. The 49 proteins fall into 28 protein families. Despite not including any functional or family annotation in the training descriptors, and focusing only on network parameters, we find enrichment in a number of families and classes. The list of 49 proteins contains 18 enzymes, of which six are phosphatases and three are protein kinases. It also contains five G-protein coupled receptors (GPCRs) and five ephrin ligands of receptor tyrosine kinases. Interesting, as well as identifying targets that are druggable, the network-based method additionally identified ligands of drugged or druggable proteins. In summary, the results obtained using our predictive network-based models reflect the enrichment of these druggable target families that is seen in targets of approved pharmaceuticals [32,42].
Additionally, among the 49 top proteins are 18 cell surface proteins and several secreted growth factors (S2 Table). It is interesting that these protein classes are identified as druggable although they are not significantly represented in the training set. However, as the training sets include all targets of FDA-approved drugs, be they targets of small molecules or biotherapeutics, it is reassuring that these cell surface targets are scoring highly as they can potentially be drugged by biotherapeutics such as monoclonal antibodies. Furthermore, at least two of these cell surface or secreted proteins are known ligands of existing drug targets (S2 Table). Similarly, several of the top-scoring proteins are adaptor proteins, three of which are known to interact with existing drug targets. Overall, 23 of the 49 proteins have direct interactions with targets of FDA-approved drugs. Thus the methods seem to identify druggable neighborhoods in the interactome as well as individual druggable nodes. Several of the top-scoring proteins of the whole interactome (S1 Table) are similarly ligands or direct interactors of drug targets, indicating that the predictive models are identifying druggable connections or network neighborhoods and not just individual drug targets.
To compare our network druggability assessment with other methods of scoring druggability, we used the protein annotation tool in canSAR [17,34,42] to obtain 3D structure-based and ligand-based druggability information for our top 49 proteins (S2 Table). Approximately half of them (24 proteins) can be linked to disease using OMIM or the Cancer Gene Census. We found that 31 of the 49 proteins have 3D structures available and, of these, 16 (52%) have at least two independent structures that are predicted to possess druggable cavities [17,34,42,43]. In comparison with the coverage of the proteome for which an estimated 25% is predicted to be druggable by the same criteria [29,30,34,44], this is a 2-fold enrichment in druggability and shows a degree of concordance between the two independent network- and structure-based druggability predictions, without the bias of functional or family annotations. The overlap may increase in the future with improved coverage of 3D structures for the proteome. Twenty four of our top 49 proteins are bound by bioactive small molecules (S2 Table) at sub-micromolar concentrations, according to the medicinal chemistry literature [34,43]. Using the ligand-based chemical druggability score that ranks targets based on the drug-like properties of bioactive compounds [34,42], we find that 28 of the 49 proteins rank in the top 25% most druggable proteins in the proteome showing a 2.3-fold enrichment over what would be expected at random. Again this highlights that the output from our network-based methodology overlaps, and complements, other independent measures of druggability despite using completely different training sets and parameters. Note that many targets cannot be assessed for ligand-based druggability or have low scores because of a lack of available chemical compound bioactivity data; thus the overlap may well increase with time as more targets are chemically explored [42].
An annotated, community-correlated map of the human interactome as described in this study is shown in Fig 2A. Although at first glance the targets of FDA-approved drugs (blue and pink) appear widely distributed, detailed inspection shows that they are concentrated in certain areas, often clustered together, whereas non-cancer drug targets are more widely distributed. There are 148 communities of size greater than 4 in this network, yet 70% of all drug targets are in the top 10 communities (Fig 2B). Furthermore, one community shows a 23-fold enrichment in the number of cancer-drug targets that it contains over what would be expected at random (Fig 2C). This probably reflects historical biases where focus was on a few easier-to-drug families or on specific, well-studied disease pathways. However, there are druggable opportunities across most regions of the interactome, as shown by the proteins that are predicted to be druggable using protein 3D structural parameters and the network parameters described in this work. Comparing the output from the three orthogonal predictors of druggability (Network-based, as presented in this method, 3D structure-based and chemical/ligand-based [34]) shows significant overlap despite basing their predictions on completely independent properties (Fig I in S1 Text).
This global view highlights large numbers of potentially missed opportunities and novel target spaces that can be explored, provided that these potential targets are validated for disease causation. Chemical exploration of these barren areas of the interactome, that are predicted to be druggable by both structure- and network-based methodologies, may well yield novel targets for future drug discovery.
There is a striking difference in the behavior of the cancer-drug versus non-cancer drug targets in the key network parameters described above, such as community behaviors and graphlet structures. This poses an important question: do these, apparently inherent properties of cancer-drug targets, make it easier for the cell to adapt signaling cascades and remodel the network in response to target inhibition? Furthermore, does this contribute to the emergence of drug resistance? Acquired drug resistance through remodeling of signaling pathways is frequently encountered in cancer therapy [30,45] and one possible way of overcoming such resistance may be through the use of combinations of drugs that target proteins occupying different network environments. We compared the network parameter profiles of the targets of well-studied drug combinations (detailed in section ‘Further information’ in S1 Text) using the limited available data. Our analysis suggests that resistance to drug combinations is more likely to occur if they act on targets with similar network profiles, and which are in close proximity in a subnetwork (such as BRAF and MEK [46,47]) compared to drug combinations that act on targets with different network environments [46,48,49] (see Fig 3).
When combining drugs acting on targets with close network proximity, the inhibition of three or more of these targets seems to be required to prevent the emergence of resistance [46]. Despite these intriguing observations, there is insufficient experimental data to allow statistical examination of whether combinations targeting different network environments have a longer-lived effect than those targeting proximal and similar network nodes.
We have presented a systematic, large-scale comparison of 321 topological, community and graphical network parameters for a fully connected interactome of 13,345 proteins and almost 90,000 interactions, totaling 4.2 million calculated properties. We identified significant differences in the network environments that are occupied by cancer-drug targets, non-cancer-drug targets, and the overall interactome. We found a major difference between the degree of cancer-drug targets which tend to have a greater number of first neighbors and be more hub-like, and the degree of non-cancer-drug targets, which, have fewer first neighbors than the interactome average. We found that cancer-drug targets tend to communicate both within and across network communities unlike non-cancer-drug targets that primarily communicate within their communities. Overall, community behavior and subgraph connectivities played the most significant roles in this distinction. Indeed it takes a complex interplay of topological, graphical and community behaviors to provide discriminatory signatures that can distinguish cancer-drug targets from non-cancer-drug targets and from the interactome as a whole. These signatures led to the generation of predictive models that predict druggable network nodes and neighborhoods with an average accuracy of 83%.
As well as identifying targets of approved drugs, the network druggability prediction models described here identified both potentially druggable targets and target local neighborhoods, providing an independent and complementary method of assessing the suitability of a target for therapeutic modulation. The methods presented in this study use only network parameters and the training sets include targets of all approved therapeutics and not just small molecule drugs. Despite this, the output of our network models showed strong concordance with the output from other orthogonal methods that use 3D structural information or ligand binding data to predict druggability. To enable the research community to use our methodologies for objective and independent target prioritization, we have provided the results of our network-based predictions alongside structure-based and ligand-based druggability results within the canSAR website (https://cansar.icr.ac.uk). These models are already useful predictive tools, the predictive power of which can only improve with the elucidation of the full human interactome and the mapping of disease-specific temporal interactions.
Exploration of the network parameters of targets in several examples of resistance to cancer drugs and mechanisms for synergistic drug targets suggests that the combination of modulators of distinct environments within the cell may be a more effective approach to overcome drug resistance than modulating targets with similar network environments. As more data from systematic, large-scale drug combination screens and clinical practice becomes available, we will be able to explore the extent to which such predictions of effective drug combinations are useful and if the can provide us with an a priori systems view of selected therapies.
The global view of the interactome presented here provides insights into important, but often neglected, systems-based considerations that should be included when selecting a target for therapeutic investigation which have the potential to inform better drug combinations.
Data imbalance, redundancy and lack of clear quality measures are all problems in defining the human interactome [19,31]. The ideal solution would be the availability of a comprehensive and unbiased protein-interaction data collection. Data from Yeast-2-Hybrid (Y2H) studies (e.g. [50,51]) are making headway towards this goal, yet currently only cover a fraction of the human interactome (detailed in ‘Further information’ in S1 Text). Nonetheless, for objective comparison, we created three separate views of the human interactome: Set A) comprising only published Y2H studies from large-scale Y2H publications containing at least 1000 proteins–this interactome contains 7,722 proteins and 24,406 interactions; Set B) all Y2H data that we could identify in the public domain–this utilized 5,537 publications and includes 10,998 proteins and 47,994 interactions; and Set C) the full experimental interactome including all Y2H publications as well as other high quality interaction data. For the Set C interactome we collected the human protein-protein interaction data from the partners of the International Molecular Exchange Consortium (IMEx [19]), Phosphosite (http://www.phosphosite.org/), and structurally-determined complexes from the Protein Data Bank [52]. We removed ambiguous interactions derived from converting a protein complex into a set of binary interactions. We created a network using R igraph package [53]. In order to compensate for the differing amounts of interactions between the proteins, we removed isolated proteins and isolated small subnetworks (Fig A in S1 Text). This resulted in a single network consisting of 13,345 proteins with 89,691 interactions and no unconnected nodes or networks. Despite our stringency in data selection, this network still contains roughly 66% of the proteome and 37% of the total predicted interactome [54].
We have defined a number of target/protein classes. Firstly, the positive ‘drug target list’ is a list of manually-curated targets of FDA-approved pharmaceuticals [32,34], defined using strict criteria based on known pharmacological action and drug approval information. Thus it is strictly confined to the curated efficacy targets of the drugs rather than targets that may bind a drug without therapeutic effect. The ‘drug target’ list includes targets of both small molecule drugs and biotherapeutics. A total of 343 human drug targets were successfully mapped to the network: of these, 127 are targets of cancer therapeutics, constituting the ‘cancer target list’, while the remainder comprise the ‘drug targets, other therapeutic areas (TA)’ list (Fig A in S1 Text). Finally, we also define a fourth ‘cancer-associated’ protein list, containing proteins that contribute to the pathology of cancer, to be a superset of the cancer-drug targets and protein products of genes from the Cancer Gene Census [27]. Thus 633 of the proteins in the network are labeled as ‘cancer-associated’. For each of the four defined positive sets (e.g. all drug targets) a matching ‘background’ dataset was defined as the remainder of the 13,345 proteins in this study (Table A in S1 Text). To address the bias caused by the correlation between the degree, or number of first neighbors, and other topological descriptors (e.g. hub-score), we further classified the datasets into three categories [17] depending on the number of first neighbors: low (≤5), medium (6–30) or high (≥31) degree (Table A in S1 Text). We examined each of the network descriptors analyzed for the full datasets as well as for these, degree-dependent subclasses of each dataset.
Additionally, we created non-redundant representatives of the training sets: 1) we clustered targets based on sequence similarity using a sequence identify cut-off of 50%; BLAST [55] E-value ≤10−6 and at least 30% sequence overlap reducing the drug target training set from 343 to 246 targets, 2) we clustered targets based on the Anatomical Therapeutic Chemical Classification System (ATC) level-3 therapeutic/pharmacological subgroups, reducing the drug target training set to 82, and 3) we clustered the targets based on shared chemical scaffolds using Bemis and Murcko [56] framework definitions; this reduced the target set to 283.
We calculated a total of 321 properties that fell into three categories: topological, graph-based and community based features (detailed in Table B in S1 Text). We calculated 31 global- and local-network topological parameters using the igraph package [53] and the Disconnectivity Valuation tool DiVa [57]. We also calculated the Dice similarity coefficient [53] based on fractions of shared neighbors which we converted to a distance matrix and performed multidimensional scaling. We used the two primary dimensions V1 and V2 as part of our topological descriptor set. For community detection, we applied two types of algorithms: Random Walk [58] and Spin-Glass [59] as implemented in igraph. The function walktrap.community was applied with a random walk of length = 4 and spinglass.community was applied with a predefined number of communities set to 50. We developed a bespoke measure of protein community communication behavior, the vertex modularity (VM) computed as the number of a protein's neighbors that are in the same community divided by the total number of neighbors a protein has. Therefore, a high VM number means the protein’s neighbors are in the same community and therefore the protein favors intra-community communication while a low VM number indicates the protein favors inter-community communication. We used GraphCrunch [60] to calculate subgraphs previously described as a means of fragmenting networks into smaller graphlets [39] (Fig B in S1 Text). The nodes within these graphlets can be classified into ‘isomorphism orbits’ [39] (here referred to simply as ‘orbits’), that reflect the pattern of interactions within the graphlet.
We created four training datasets comprising the positive and background sets (Fig A in S1 Text): drug targets (all TAs), cancer-drug targets, non-cancer-drug targets, and cancer-disease associated proteins. We further split each of these sets into four degree-based subsets as described earlier (all, high degree, medium degree and low degree). This resulted in 16 datasets for modeling (Table A in S1 Text). It is important to note that some of the subsets are very small, such as the ‘Low’ cancer drug target set which contains only 16 proteins and the ‘High’ non-cancer drug targets which contains 23 proteins. These sets are too small for effective model building. We inputted the 321 descriptors, calculated for each of the 16 sets, into three distinct predictive modeling algorithms: Random Forests [61], Gradient Boosted Machines (GBM [62]) and Generalized Linear Models (GLM [63]). Since we can only label proteins as drug targets or background or unlabeled proteins (i.e. it is not possible to assign a negative training set as it is not possible to say which proteins are currently are not drugged but may become successful drug targets in future) we apply a positive-unlabeled (PU) learning paradigm (see e.g.[64]).
All models were implemented in R and the code with the specific algorithm parameters is provided at: https://cansar.icr.ac.uk/cansar/publications/druggable_network_neighbourhoods/
Using the data derived above, we constructed several models to predict: 1) general druggability (the likelihood of a protein to be a drug target for any therapeutic area); 2) cancer druggability (the likelihood of a protein to be a cancer-drug target); 2) non-cancer druggablity (the likelihood of a protein to be a drug target for a non-cancer therapeutic area); and finally 4) cancer-association (the likelihood of a protein to be a cancer-associated protein). Table 1 reports the results of a 10-fold cross validation for the All, Low and Medium datasets and a 5-fold cross validation for the High dataset due to the small minority class. In total, we built 450 models.
Predictive modeling of network data poses an interesting problem when it comes to training the model. The standard is to report the results of a k-fold cross-validation (CV). For example in 10-fold CV, the data are split into training and validation sets and the model is built using the 90% training subset and validated on the 10% subset. This process is repeated 10 times and the average accuracy of the validation is reported as the prediction accuracy. This method is widely adopted as it approximates how the model will perform on new, unseen data. However, with a network, each instance is dependent on other instances as the descriptors are based on the instance’s position in the network. Consequently, using a holdout-set is nonsensical, as there can be no new cases without generating the network data again. To overcome this problem with the 10-fold CV, we recreated random training sets that maintained the structure of the network and the number of positives, but where the positives were allocated to random proteins. We carried out a 10-fold CV on these random sets to compare to the predictive results observed from the true training sets. Another problem for the predictive modeling of the network was the imbalance of the data. The minority classes ranged from 1% to 5% and therefore regression models were built rather than 2-class classification models. As our data comprises only PU data sets, we report the results based on a ranked evaluation of area under the curve (AUC). We ranked the predictions according to their average regression output and calculated the percentile, for example, a score of 78% means that 78% of proteins had a lower rank than this protein.
There are five supplementary files provided with this document and also on the canSAR website: https://cansar.icr.ac.uk/cansar/publications/druggable_network_neighbourhoods/
|
10.1371/journal.pgen.1006268 | Smc5/6 Is a Telomere-Associated Complex that Regulates Sir4 Binding and TPE | SMC proteins constitute the core members of the Smc5/6, cohesin and condensin complexes. We demonstrate that Smc5/6 is present at telomeres throughout the cell cycle and its association with chromosome ends is dependent on Nse3, a subcomponent of the complex. Cells harboring a temperature sensitive mutant, nse3-1, are defective in Smc5/6 localization to telomeres and have slightly shorter telomeres. Nse3 interacts physically and genetically with two Rap1-binding factors, Rif2 and Sir4. Reduction in telomere-associated Smc5/6 leads to defects in telomere clustering, dispersion of the silencing factor, Sir4, and a loss in transcriptional repression for sub-telomeric genes and non-coding telomeric repeat-containing RNA (TERRA). SIR4 recovery at telomeres is reduced in cells lacking Smc5/6 functionality and vice versa. However, nse3-1/ sir4 Δ double mutants show additive defects for telomere shortening and TPE indicating the contribution of Smc5/6 to telomere homeostasis is only in partial overlap with SIR factor silencing. These findings support a role for Smc5/6 in telomere maintenance that is separate from its canonical role(s) in HR-mediated events during replication and telomere elongation.
| Structural Maintenance of Chromosome (SMC) complexes, include cohesin, condensin, and the Smc5/6 complex. These complexes are involved in many aspects of chromosome organization, with cohesin and condensin having relatively well-characterized roles. Cohesin holds newly replicated DNA strands together and condesin is critical for chromosome condensation and genome compaction as cells enter mitosis. However, a role for the Smc5/6 complex in higher-level chromosome organization has remained ill defined. The Smc5/6 complex is recovered at chromosome ends in all stages of the cell cycle and has a role in telomere biology. Smc5/6 integrity is necessary for Sir4 binding, telomere clustering, and transcriptional silencing. In all, our data suggest that Smc5/6 has a physiologically relevant role in chromatin maintenance at telomeres and telomere organization within the nucleus that are distinct of it functionality in homologous recombination.
| Structural maintenance of chromosome (SMC) protein complexes facilitate chromosome structure and organization in eukaryotes. Three SMC complexes are found in eukaryotes and each has a unique role in chromosome dynamics and metabolism. Underscoring their importance and distinct functionality, all three complexes and their individual components are essential for cell viability. Cohesin regulates sister chromatid cohesion and condensin is important for chromosome compaction by tethering different regions of the same chromosome [1–3].
The third complex, Smc5/6, contains six non-SMC proteins in addition to Smc5 and 6 including Mms21/ non-Smc element 2 (Nse2), which is an E3 SUMO ligase (Fig 1A) [4–6]. As well, Nse1 and Nse3 bind to Nse4 to form a heterotrimer, which in turn interacts with the ATPase head domain generated by the N- and C-termini of Smc5 and Smc6 [7, 8]. Nse1 is a putative ubiquitin ligase and Nse3 is a MAGE (melanoma-associated antigen gene) domain containing protein that is important for loading the complex onto chromatin [9–11]. The Smc5/6 complex functions in homologous recombination (HR) and replication, and it localizes to repetitive elements such as the rDNA and telomeres presumably to promote and resolve HR-dependent intermediates [12–14].
A telomeric function for the Smc5/6 complex in ALT has been demonstrated in both human and yeast cells [22–24]. In human ALT cells, a knockdown of components in the Smc5/6 complex inhibits recombination at telomeres, resulting in telomere shortening and senescence [22]. As well, in telomerase negative yeast cells smc6-9 and mms21-11 mutant alleles exhibited accelerated senescence attributed to the accumulation of recombination intermediates, but also to an HR–independent mechanism involving the untimely termination of DNA replication [23, 24]. The Smc5/6 complex is enriched at telomeres in telomerase positive asynchronous cultures [12, 13], however its characterization outside the ALT pathway remains limited. In telomerase positive cells, the smc6-9 allele exhibited mis-segregation of repetitive elements at telomeres which is attributed to defects in HR [12] and the mms21-11 allele was shown to have defects in telomere clustering with increased telomere position effect (TPE) [4]. Subsequent to the initial characterization of mms21-11, mms21Δsl mutants showed a loss of TPE and SIR binding [25]. Thus, allele specific variations have complicated the understanding of Mms21 and SUMO mediated events in TPE [4, 25]. Further characterization of Smc5/6 in telomere homeostasis using a mutant allele of a distinct complex component will provide additional information about the functionality of Smc5/6 at telomeres.
Telomeric DNA in S. cerevisiae contains tandem repeats of (AC1-3/TG1-3) n; n = 275–375 [26] along with two types of subtelomeric repeat elements called Y’ and X [27]. The Y’ sequence is located adjacent to the tandem repeats at many, but not all subtelomeres, whereas X-elements are found at the ends of all chromosomes [28]. Rap1 binds directly to the double-stranded TG1-3 DNA moiety and is a central regulator of telomere biology [29]. The C-terminal domain of Rap1 interacts with Rif1 and Rif2 and regulates telomere length via a counting system that involves their interaction with Rap1 [30, 31]. Telomeres are elongated in rif1Δ and rif2Δ cells via telomerase dependent and HR independent events [32, 33].
The C-terminal domain of Rap1 also binds the SIR complex, which is important for transcriptional silencing primarily via interactions with Sir4 [30, 32, 34, 35]. SIR proteins are important for telomere position effect (TPE) and the formation of heterochromatin, which nucleates at telomeres and then spreads several kilobases into subtelomeric regions [36, 37]. Subtelomeric heterochromatin is maintained by seemingly distinct events that are likely to be interrelated in vivo. For example, in budding yeast, 32 telomeres cluster together in 3–8 foci at the nuclear periphery, and this drives the sequestration of SIR complex sub-compartments within the nucleus, and promotes silencing [38]. Additionally, the SIR complex, along with Rif1 and Rif2, modulates the level of long non-coding telomeric repeat-containing RNA, TERRA, which is also an integral factor in heterochromatin formation [39–42]. TERRA levels have never been reportedly assessed in Smc5/6 compromised cells and a role for the complex in heterochromatin maintenance and transcription at telomeres remains to be clearly defined.
Here we show that the Smc5/6 complex binds telomeres, not only during late S phase when telomeres are synthesized, but also throughout the cell cycle in telomerase positive cells. Telomere clustering and full Sir4 binding is indeed dependent on the SUMO ligase activity of Mms21, however in the course of characterizing a temperature sensitive (ts) mutant of NSE3, telomere defects were observed in cells harboring the nse3-1 allele, which have not been previously reported with other alleles having compromised Smc5/6 functionality. TPE and TERRA regulation, as well as telomere length defects in nse3-1 mutants were additive with the loss of SIR4. In all, our data support a model that extends the functionality of Smc5/6 at telomeres beyond its previously reported roles in homology-mediated events in the ALT pathway [22–24].
The Smc5/6 complex has been detected at telomeres [12, 13] and stalled and collapsed replication forks [43–48]. Given that telomeres are difficult to replicate sites and prone to fork stalling, we wanted to determine if the presence of Smc5/6 at chromosome ends coincided solely with telomere duplication or if it was present at telomeres independent of replication. We monitored Smc6FLAG enrichment as a marker for the complex and performed chromatin immuno-precipitation (ChIP)–qPCR at multiple time points after cells were synchronously released from G1 into S phase. Significant enrichment of Smc6FLAG was observed at three telomeres above a late-replicating control region on Chr V (Fig 1B) [49, 50], showing the Smc5/6 complex is constitutively present at telomeres and not only during the time of telomere replication in late S phase (Fig 1B).
It was recently demonstrated that Nse3 in fission yeast is important for loading the Smc5/6 complex onto chromatin [11]. We wanted to determine the involvement of Nse3 in localizing Smc5/6 to its endogenous binding sites such as telomeres in budding yeast. As with all subcomponents of the complex (Fig 1A), NSE3 is essential precluding its deletion. Therefore, we utilized a mutant allele, nse3-1, which contains seven amino acid substitutions and was isolated from a screen for temperature sensitivity (ts) at 37°C [51] (Fig 1C and 1D).
As nse3-1 mutant cells do not synchronize efficiently with α-factor (S1A Fig), we determined Smc5/6 localization in asynchronous cultures at the semi-permissive temperature 34°C. The enrichment of Smc6FLAG was significantly reduced in nse3-1 mutant cells at telomeres and other known sites of Smc5/6 binding (Fig 1E, S1B Fig). In contrast, the level of Smc6FLAG recovered at telomeres in mms21-11 mutant cells, which are HR and SUMO ligase deficient, was similar to wild type (Fig 1E, S1B Fig). One explanation for the loss of Smc6FLAG recovery is that the complex is unstable in nse3-1 mutant cells. To address this possibility, we performed co-immunoprecipitation with two subcomponents that do not directly interact with one another, Nse6 and Smc5, as previously described [48]. In nse3-1 mutant cells, Nse6 was recovered in Smc5 pull-downs at levels comparable to wild type cells (S1C Fig), suggesting the complex does not markedly dissociate in nse3-1 mutants.
Telomeres were also slightly shorter in nse3-1 mutants compared to wild type and HR-defective smc6-9 mutant cells (Fig 1F). In contrast, slightly longer telomeres were observed in mms21-11 mutants (Fig 1F), which is consistent with its initial characterization showing that this allele had longer telomeres [4]. The changes are indeed subtle, however there is a noticeable difference in telomere length when comparing the nse3-1 to the other complex mutants, suggesting that the Smc5/6 complex might have a role at telomeres distinct from HR-mediated events.
Telomere clustering at the nuclear periphery in S. cerevisiae establishes sub-nuclear zones that sequester repressors of transcription [52, 53]. Clustering can be visualised by performing immunofluorescences and counting GFP-Rap1 foci. In haploid cells, it has been demonstrated that 32 telomeres cluster in limited number [54], and consistent with this, our quantification showed ~90% of wild type cells contained ≤ 6 foci in both G1 and S phases of the cell cycle at 34°C (Fig 2A–2C). In contrast, nse3-1 mutants had ≥ 6 foci in ~65% and ~80% of the cells in G1 and S phases respectively, with 10–20% having ≥ 9 foci (Fig 2A–2C). In a side-by-side comparison and in line with its initial characterization, a similar clustering defect was observed in mms21-11 mutants [4], but smc6-9 mutant cells were similar to wild type (Fig 2D).
Defects in clustering coincide with a disruption in SIR proteins, [55, 56]. Sir4Myc is expressed at similar levels in all strains (Fig 2E), and as measured by immunofluorescence, Sir4Myc forms discrete punctate foci in wild type cells (Fig 2F). In contrast, Sir4Myc became relatively dispersed throughout the nucleus in nse3-1 mutant cells (Fig 2F). Dispersion was also observed in mms21-11 and smc6-9 alleles, but to a lesser extent than the level observed in nse3-1 mutants (Fig 2F). Foci, albeit with reduced intensity, remained in all mutants to varying degrees, therefore as a complement to immunofluorescence and to quantify changes at telomere, we performed ChIP with Sir4Myc. The level of Sir4Myc recovered at telomeres in both nse3-1 and mms21-11 mutants was reduced to ~40% that of wild type cells (Fig 2G). In smc6-9 mutant cells, the level of Sir4Myc at telomeres was not significantly different from the amount recovered in wild type (Fig 2G). Taken together, the alleles with defects in clustering, nse3-1 and mms21-11, also showed a reduction in the level of Sir4 bound at telomeres.
Sir4 sumoylation by Siz2 was previously implicated in peripheral telomere position [57, 58]. Given that our results indicated Sir4 localization to be regulated by Mms21, we investigated if the SUMO status of Sir4 itself might provide a level of regulation. Similar to siz2Δ, the level of Sir4 sumoylation was reduced in mms21-11, however SUMO levels remained similar to WT, if not higher in nse3-1 mutants (S2 Fig). These data suggest that Sir4 localization to telomeres is not regulated by the SUMO status of Sir4 in nse3-1 cells.
To further understand the relationship between Sir4 and the Smc5/6 complex we performed co-immunoprecipitation to see if we could detect a physical interaction. Upon Smc6FLAG immunoprecipitation, we recovered Sir4Myc (Fig 3A). We had variable results with the reciprocal IP, however we found that Nse3HA associated with Sir4Myc pull downs (Fig 3B), suggesting that the Smc5/6 and SIR complexes physically associate in vivo.
The Smc5/6 complex influenced Sir4 recovery at telomeres and a physical interaction between the complexes was detected. Thus, the reverse was performed to determine if Sir4 levels impacted the localization of Smc5/6 at telomeres. ChIP was performed with Smc6FLAG and Smc5FLAG and recovery at telomeres was compared in sir4Δ and wild type cells (Fig 3C and 3D). The level of Smc6FLAG in cells lacking SIR4 decreased to ~60% the amount recovered in wild type cells (Fig 3C). Similarly, Smc5FLAG was reduced in sir4Δ mutants to ~25% that of wild type levels (Fig 3D). As Smc5 and Smc6 are present at stoichiometric levels in the complex [4, 59], the greater relative change with Smc5FLAG might result from IP variability. Nonetheless, there is a statistically significant decrease in both core factors of the Smc5/6 complex bound to telomeres in sir4Δ mutants compared to wild type cells (Fig 3C and 3D).
Sir4 is a critical factor for TPE and in the maintenance of heterochromatin near telomeres [60]. As the Smc5/6 complex interacts with Sir4, and the presence of Smc5/6 is important for Sir4 recovery at telomeres and vice versa, we assessed a role for the complex in transcriptional gene silencing regulation. TPE was determined in reporter strains where URA3 was integrated at the left arm of telomere VII [61]. Consistent with previous reports, sir4Δ cells showed defects in TPE as measured by their compromised ability to form colonies on medium containing 5-fluoroorotic acid (5-FOA) (Fig 3E) [60]. For nse3-1 mutants, TPE was disrupted but not to the level observed with sir4Δ (Fig 3E). In contrast and consistent with previous reports, TPE in mms21-11 and smc6-9 mutant cells remained intact at 25°C and 34°C (Fig 3E; [4]). This data indicated that the loss of silencing in nse3-1 mutant cells could not be solely attributed to a defect in Sir4 recruitment. This is supported by the observation that both nse3-1 and mms21-11 mutants showed a comparable defect of Sir4 recovery at telomeres and this was sufficient to silence the reporter transgene in the mms21-11 allele.
To bring insight to the functionality of Smc5/6 in transcriptional silencing at telomeres the nse3-1 allele was combined with the loss of either SIR4 and/or RIF2. Utilizing the URA3 reporter assay (Fig 4A, S3 Fig), it was difficult to observe an additive defect in silencing for nse3-1 sir4Δ double mutants because the loss of silencing is so penetrant with the loss of SIR4. Therefore, two endogenous sub-telomeric sites, YR043C and CHA1, on Tel9R and Tel3L respectively were assessed [62, 63]. Gene transcription increased in nse3-1 sir4Δ double mutants compared to sir4Δ single mutant cells (Fig 4B). Moreover, a defect in silencing was also observed in nse3-1 mutants at VAC17, a gene adjacent to CHA1 and previously determined to be silenced independently of Sir4 (Fig 4B; [63]). An additive loss of silencing was not observed when smc6-9 was combined with sir4Δ (S4 Fig), suggesting that HR-regulated functions involving the Smc5/6 complex are separable from its function in transcriptional silencing. In rif2Δ cells, silencing remains and even increases presumably through increased binding of Sir4 to Rap1 at telomeres (Fig 4C) [30, 64]. The nse3-1 rif2Δ double mutants exhibited a loss of silencing that was similar to nse3-1 single mutant cells (Fig 4C), however this was difficult to observe when measuring TPE from the URA3 reporter unless cell concentrations were low (S3 Fig).
Rap1 binds both Sir4 and Rif1/2 [30, 32, 34, 65], and given the interactions nse3-1 had with these factors it was important to assess Rap1 binding to telomeres in nse3-1 mutants. By ChIP, we observed no significant difference in the level of Rap1Myc bound at telomeres in nse3-1 mutants compared to the levels in wild type cells (S5A Fig). These data also support the interpretation that the increased number of Rap1 foci we measured in nse3-1 cells resulted from a disruption in telomere clustering rather than a disruption of Rap1 binding to telomeres (Fig 2A–2C).
Nse3 was previously reported to interact with Rif2 in a high-throughput yeast two-hybrid (Y2H) screen [66]. We verified the Rif2-Nse3 interaction and determined it was reduced when nse3-1 was expressed (S6 Fig), however, in contrast to Sir4Myc, the levels of Rif1Myc and Rif2Myc at telomeres in nse3-1 were similar to wild type (Fig 4D, S5B Fig), and no significant change with Smc6FLAG was measured at telomeres in cells lacking RIF1 or RIF2 (Fig 4E). In all, these data suggest that the physical association between Nse3 and Rif2 is not driving the recruitment of either factor/complex to telomeres.
Cells carrying the nse3-1 allele exhibit slightly shorter telomeres (Figs 1F and 4F), which is opposite to cells lacking RIF1 or RIF2, which are negative regulators of telomerase [33]. Telomere length was determined when nse3-1 was combined with rif1Δ and rif2Δ. The nse3-1 rif2Δ double mutant cells exhibited a partial reversion in the telomere length phenotype (lanes 5 and 6; Fig 4F). However, when nse3-1 was combined with rif1Δ, telomere length looked indistinguishable from rif1Δ single mutants (lanes 3 and 4; Fig 4F). These data suggest the nse3-1 mutation does not counteract telomere elongation as a general mechanism per se and support the model that Rif1 and Rif2 having non-overlapping roles in telomere maintenance even though they interact with each other and with Rap1 [67–70].
As the Smc5/6 complex is implicated in HR and the ALT pathway, we also investigated if the partial reversion of long telomeres in nse3-1 rif2Δ was regulated by HR events. Upon disruption of RAD52, no detectable changes were observed, as telomeres for nse3-1 rad52Δ and nse3-1 rif2Δ rad52Δ mutants were similar in size to nse3-1 and nse3-1 rif2Δ mutants respectively (lanes 2 and 8; lanes 6 and 10; Fig 4F). Moreover, telomere shortening was not observed when the loss of RIF2 was combined with the HR-deficient smc6-9 allele (S7 Fig). Taken together, these data provide additional support for Smc5/6 having a role at telomeres distinct of its functionality in HR-mediate events.
In addition to the transcription of gene-coding regions, RNA polymerase II also transcribes TERRA at telomeres [40]. There are reported correlations between non-physiological increases and decreases in TERRA levels with telomeric abnormalities [39, 71]. Moreover, TERRA expression was previously demonstrated to be regulated by Rap1, the SIR complex, and Rif1/2 proteins, with the role of Rif2 being minimal and only at a subset of telomeres [42]. As the nse3-1 mutation results in a loss of silencing at subtelomeric genes and showed interactions with Rif2 and Sir4 we measured TERRA expression in cell carrying the nse3-1 allele.
Compared to wild type, there was a significant de-repression in TERRA expressed from both X only and Y’ telomeres in nse3-1 mutants at both 28°C and 34°C (red, Fig 5A and 5B, S8 Fig). Consistent with previous reports [42], sir4Δ mutants showed substantial TERRA expression from X only telomeres (purple, Fig 5A and 5B), and we observed no distinguishable increase in TERRA levels in cells lacking RIF2 at TEL1R, 6R, or Y’ (aqua, Fig 5A and 5B, S8 Fig). TERRA levels in nse3-1 and nse3-1 rif2Δ were similar and significantly higher than the level measured in rif2Δ mutant cells (red, green, and aqua; Figs 5A and 5B and S8). Interestingly, and consistent with the TPE reporter assay, TERRA levels in sir4Δ rif2Δ cells (light grey) were silenced to levels not statistically different from wild type (dark grey), and similar to rif2Δ (aqua, Fig 5A and 5B). There was a 2- and 4- fold increase in the level of TERRA from Y’ and X-only telomeres respectively in nse3-1 sir4Δ cells (blue) compared to cells lacking SIR4 (purple) at 28°C (Fig 5A, S8A Fig). The same trend was observed at 34°C, however variability between experiments resulted in p values > 0.05 (Fig 5B, S8B Fig).
Both nse3-1 and sir4Δ mutants have slightly shorter telomeres (Figs 1F and 5C) [53]. As well, transcription and TERRA levels increased in nse3-1 and these phenotypes were additive with sir4Δ. Given the correlations between increased TERRA levels and induced transcription with telomere shortening [40, 72] we proceeded to assess telomere length in nse3-1 sir4Δ double mutants. Telomeres shorten further in double mutants compared to cells harboring either nse3-1 or sir4Δ single mutant alone (Fig 5C). Highlighting the difference again between nse3-1 and smc6-9, the level of TERRA expression was not additive in smc6-9 sir4Δ double mutant cells (S9A and S9B Fig) and in contrast to nse3-1, telomere length in smc6-9 did not result in additive shortening when combined with sir4Δ. (S9C Fig). Taken together, our data support a model whereby Smc5/6 has a role in transcriptional silencing and telomere length maintenance that is different from its involvement in HR dependent events at telomeres and underscore the value of characterizing various ts alleles of the complex.
We report a previously uncharacterized function for the Smc5/6 complex with links to transcriptional silencing and demonstrate a role for the complex in telomere homeostasis. In cells carrying the nse3-1 allele, Smc5/6 complex levels are markedly reduced at telomeres. This was true for cells grown at 25°C or 34°C, the temperature we used in many of our measurements, indicating that higher temperature did not introduce confounding defects to the complex in this mutant background (S10 Fig). Utilizing nse3-1, we show that Smc5/6 is critical for 1. Maintaining proper telomere length, 2. Telomere clustering, 3. SIR complex recovery at telomeres, 4. TPE, and 5. Regulating TERRA levels.
Telomere defects involving mutations in the Smc5/6 complex were first reported with the mms21-11 allele; however, the loss of SUMO ligase activity did not appear to impact TPE, as expression from a URA3 reporter construct integrated at Tel5R remained silent [4]. Upon characterization of the nse3-1 allele, we also observed a loss of clustering, but unlike mms21-11, TPE was disrupted as shown by an increase in expression of sub-telomeric genes and URA3 reporter expression. Further characterization of nse3-1 and mms21-11 alleles demonstrated that a decrease in Sir4 binding at telomeres was common to both alleles (a summary of phenotypes can be found in S3 Table). In agreement with previous reports (Zhao & Blobel, 2005), and in side-by-side comparison with nse3-1 and wild type, we find silencing at sub-telomeres remained intact for mms21-11 and smc6-9 mutants (Fig 3E), suggesting that the partial reduction in Sir4 at telomeres in mms21-11 and nse3-1 mutants was not sufficient to abrogate silencing. These data also raise the possibility that the complex might have additional functions, which are disrupted in nse3-1, that are important for silencing. Our data also suggest a partial interdependency between the Smc5/6 complex and Sir proteins at telomeres. Indeed, a physical interaction is detected between the Smc5/6 complex and Sir4 (Fig 3A and 3B) and in the absence of SIR4 there is a moderate but statistically significant ~30% reduction in the levels of Smc6FLAG recovered at telomeres, however for comparison, Smc6FLAG was reduced further in nse3-1 mutant cells by ~60% the levels of wild type (S10B Fig). Even though Smc5/6 and Sir4 contribute to the stability of one another at telomeres, the defects in TPE and TERRA expression associated with the loss of Smc5/6 at telomeres are additive with the loss of SIR4.
Live-cell imaging at the single-cell level demonstrated that when telomeres become critically short, TERRA is transcribed, and this recruits telomerase to the TERRA-expressing telomere to promote elongation [73]. Increased TERRA levels above physiologically important levels likely have an inhibitory affect on telomere length maintenance. TERRA levels in nse3-1 mutants are above wild type and when combined with sir4Δ, the double mutants show an even greater increase in TERRA compared to the levels measured in cells lacking SIR4 only. The elevated transcription and loss of TPE in nse3-1 is likely to have a direct effect on TERRA expression and supports the model that Smc5/6 functionality is important for silencing, and when deregulated, transcription lead to increases in TERRA and telomere loss [74]. Telomere shortening is additive in nse3-1 sir4Δ mutants. The robust expression of TERRA in nse3-1 sir4Δ cells possibly reinforces the shortening of telomeres, and vice versa. Indeed this explanation is consistent with previous work showing that when TERRA increases, telomeres shorten via telomerase inhibition [40], as well as disrupting the inhibitory effect of yKu70/80 on Exonuclease 1, leading to its increased activity at telomeres [75]. A more speculative model, that will require additional investigation, is that increases in TERRA expression might lead to increased RNA-DNA hybrids at telomeres and subsequently more aberrant replication fork structures that fail to be resolved by Smc5/6, and this results in telomere loss specifically in alleles deficient in silencing, as in nse3-1 and nse3-1 sir4Δ mutant cells. Lastly, an alterative model that we cannot exclude is that there is a more direct effect of nse3-1 on telomere length independent of TERRA, which might involve interactions of the Nse1-Nse3-Nse4 sub-complex within Smc5/6 that become altered in cells carrying the nse3-1 allele.
We also assessed the SUMO status of Sir4 and determined that sumoylation was reduced in mms21-11 mutant cells to levels similar to those previously observed in cells lacking SIZ2 (S2 Fig) [57]. However, Sir4 sumoylation remained, and was slightly higher in cells harbouring the nse3-1 allele when silencing is reduced. This is consistent with previous work showing that increased levels of Siz2, and by extension elevated sumoylation, function antagonistically to silencing [58], and also suggests there is no direct correlation between Sir4 sumoylation in telomere clustering at the periphery. These data are also consistent with the observation that a SUMO-Sir4 fusion construct could not restore anchoring in siz2Δ mutants, which suggested that sumoylation of another target, besides Sir4, is important for telomere positioning at the periphery [57].
Telomere clustering and silencing are distinguishable functions [76, 77]. Our data indicates that Smc5/6 likely contributes to both and independently of HR as smc6-9 was not distinguishable from wild type in all measures, and that Mms21 sumoylation is important for clustering, but not silencing. Determining the role of Smc5/6 in clustering at the periphery will require further investigation. Organization of telomeres at the periphery is driven by partially redundant pathways involving Sir4 binding to membrane bound Esc1 and Yku70/80 [76, 78]. First, although Sir4 sumoylation does not control clustering we have not assessed if Esc1, which is also a target of sumoylation, regulates clustering in a pathway dependent on Mms21 activity [79, 80]. Secondly, unlike Sir4, we observed that the level of YKu70 at telomeres in nse3-1 mutant cells was not statistically different from wild type cells (S11 Fig). However, determining if Mms21 dependent sumoylation of yKu70 at telomeres is critical for Smc5/6 mediated anchoring will provide an additional level of understanding as both Yku70 and Yku80 sumoylation are important for perinuclear positioning [57], and while Yku80 sumoylation is markedly reduced in siz2Δ mutants, Yku70-sumoylation is primarily dependent on Mms21 [4, 57]. Our data support a model where the Smc5/6 complex, like other proteins involved in DNA repair, such as Tel1 and Mre11, contributes to transcriptional silencing via two pathways, one involving direct interactions with SIR factors and the other regulating nuclear position and association with the periphery [81].
The current study demonstrates a role for Smc5/6 complex in telomere maintenance that is distinct from its previously characterized functions in replication and HR. Our data show that the Smc5/6 complex is a bona fide telomere-binding factor that has reduced recovery in nse3-1 mutant cells (Fig 5D). Our study establishes Smc5/6 as having a physiological role in the structural maintenance of chromosome ends where its localization and integrity contribute to the stabilization of factors with well-established roles in telomere maintenance and metabolism. Consistent with a role in end protection, the localization of Smc5/6 to telomeres is critical for telomere clustering and transcriptional repression (Fig 5D). These roles for Smc5/6 together its involvement in the various aspects of HR-mediated DNA metabolism, such as replication and repair, perhaps contribute to the essential requirement of this complex for cell survival.
All strains used in this study are listed in S1 Table. The nse3-1 mutant was a kind gift from Dr. P. Hieter at Michael Smith Laboratories. In all experiments exponentially growing cells were incubated at 34°C for 2hrs before harvesting, unless indicated otherwise. Drop assays were performed by growing cells overnight, and then performing 10-fold serial dilutions where 4μl of each dilution were plated on YPAD an incubated at the indicated temperature. For repression assays, 5-fold or 10-fold dilutions from overnight cultures were plated on SC or SC + 5-FOA as described [76, 82] at the indicated temperatures.
ChIP experiments performed as described previously [83], except that cells were incubated at 34°C for 2 hours before crosslinking with formaldehyde in media where the temperature was held a 25°C to allow efficient crosslinking. Immunoprecipitates were washed once with lysis buffer (50 mm HEPES, 140 mm NaCl, 1 mm EDTA, 1% Triton X-100, 1 mM PMSF and protease inhibitor pellet (Roche)) and twice with wash buffer (100 mM Tris (pH 8), 0.5% Nonidet P-40, 1 mM EDTA, 500 mM NaCl, 250 mM LiCl, 1 mM PMSF and protease inhibitor pellet (Roche)). Real-time qPCR reactions were carried on using SYBR green method. Results shown as fold enrichment at three native subtelomeres (Tel1L, Tel6R and Tel15L) compared to a control (ctrl) late replicating region on Chromosome V (469104–469177) [49, 50]. Primer sequences are listed in S2 Table.
For Rap1-GFP foci imaging, cell were grown to the 5x106 cells/ml at 34°C for 2 hours in synthetic complete (SC) media. Images were captured immediately in 21 Z-stacks of 0.2 μm using Zeiss Axiovert 200 microscope. GFP foci per nucleus were manually counted as a representation for telomere foci. For Sir4 immunofluorescence, cell cultures were grown to the 5x106 cells/ml at 34°C for 2 hours in synthetic complete (SC) media. Cells were immediately fixed using 3.7% formaldehyde and spheroplasted in SK (0.1M KPO4/1.2M sorbitol) buffer containing 0.4 mg/ml Zymolase (US, Biological). Spheroplasted cells were fixed on poly-lysine coated coverslips as described previously [84]. Coverslips were blocked in 1% BSA in PBS for 1 hour, then incubated with primary (αMyc, ab9106-100) followed by secondary (Alexa 488; Molecular Probes, Invitrogen) antibodies each for 30 minutes. Coverslips were mounted on microscope slides using vectashield-containing DAPI (Molecular Probes, Invitrogen). Images were taken in 21 Z-stacks of 0.2 μm using Zeiss Axiovert 200 microscope and Z-stack images were flattened and presented in the figures. ImageJ (NIH, USA) was used for adjusting background in both live and immunofluorescence imaging methods.
Strains carrying HA-tagged Nse6 and Myc-tagged Smc5 were grown to the log phase at room temperature and then incubated for 2 hours at 34°C in YPAD media. Cells were lysed with zirconia beads in lysis buffer (50 mm HEPES, 140 mm NaCl, 1 mm EDTA, 1% Triton X-100, 1 mM PMSF and protease inhibitor pellet (Roche)). Cell lysates were incubated with αMyc antibody-coupled Dynabeads (Invitrogen) for 2 hours at 4°C. Immunoprecipitates were washed once with lysis buffer and twice with wash buffer (100 mM Tris (pH 8), 0.5% Nonidet P-40, 1 mM EDTA, and 400 mM NaCl, 1 mM PMSF and protease inhibitor pellet (Roche)), each for 5 minutes. Beads were resuspended in SDS loading buffer and subjected to SDS gel electrophoresis followed by western blotting by αHA (Santa Cruz, F7) and αMyc (9E10) antibodies. The same procedure was performed for Sir4-Nse3 except that lysates were clarified with one round of centrifugation at 13200 rpm before incubating with Myc antibody-coupled beads and immunoprecipitates were washed once with lysis buffer and twice with wash buffer (100 mM Tris (pH 8), 0.5% Nonidet P-40, 1 mM EDTA, and 250 mM LiCl, 1 mM PMSF and protease inhibitor pellet (Roche)). The co-IP between Sir4 and Smc6 was performed in stationary phase cultures without a chromatin spin and with a wash buffer containing 250 mM NaCl rather than 250 mM LiCl.
Measurement of telomere length was performed as described in [15]. Cells were grown for 48 hours to stationary phase in liquid YPAD at 34°C and harvested for Southern blotting. Genomic DNA from each strain were digested with XhoI and then separated by 1% agarose gel electrophoresis. Denatured DNA was transferred to Amersham Hybond-XL (GE Healthcare Life Sciences) membrane and hybridized with radiolabeled telomeric repeat probe (TG1-3/C1-3A). Rediprime II DNA Labeling System used to radiolabel telomeric probe (GE).
Exponentially growing WT and nse3-1 cells were incubated for 2 hours at 34°C prior to harvesting by centrifugation and snap freezing in liquid nitrogen. Cells were lysed and mRNA isolation was followed by reverse transcription Complementary DNA (cDNA) was amplified and quantified using the SYBR Green qPCR method. Primers are listed in S2 Table. Fold gene expression represents real time qPCR values relative to WT samples. Gene expression values were normalized to ACT1 expression as the internal control.
Total RNA was extracted as in [73]. 2 μg of RNA was treated with 4U of DNase I (Thermo-Fisher) for 4 hr at 37°C and then purified by phenol/chloroform extraction. 1μg of DNase-treated RNA was reverse transcribed by using RevertAid Reverse Transcriptase (Thermo Fisher) at 42°C for 1 hr. 0,5 μmol of a C-rich primer (CACCACACCCACACACCACACCCACA) and 0,5 μg of a poly(dT) primer was used for the reverse transcription reaction (RT). 20ng of cDNA was used for the qPCR, which was performed using the qPCR master mix SsoFAST EvaGreen Supermix from Bio-Rad. qPCR was carried out on a Roche LightCycler96. TERRA expression was normalized against ACT1 mRNA expression using the delta Ct method and than normalized against the WT yeast strain.
|
10.1371/journal.pbio.1000514 | Quiescent Fibroblasts Exhibit High Metabolic Activity | Many cells in mammals exist in the state of quiescence, which is characterized by reversible exit from the cell cycle. Quiescent cells are widely reported to exhibit reduced size, nucleotide synthesis, and metabolic activity. Much lower glycolytic rates have been reported in quiescent compared with proliferating lymphocytes. In contrast, we show here that primary human fibroblasts continue to exhibit high metabolic rates when induced into quiescence via contact inhibition. By monitoring isotope labeling through metabolic pathways and quantitatively identifying fluxes from the data, we show that contact-inhibited fibroblasts utilize glucose in all branches of central carbon metabolism at rates similar to those of proliferating cells, with greater overflow flux from the pentose phosphate pathway back to glycolysis. Inhibition of the pentose phosphate pathway resulted in apoptosis preferentially in quiescent fibroblasts. By feeding the cells labeled glutamine, we also detected a “backwards” flux in the tricarboxylic acid cycle from α-ketoglutarate to citrate that was enhanced in contact-inhibited fibroblasts; this flux likely contributes to shuttling of NADPH from the mitochondrion to cytosol for redox defense or fatty acid synthesis. The high metabolic activity of the fibroblasts was directed in part toward breakdown and resynthesis of protein and lipid, and in part toward excretion of extracellular matrix proteins. Thus, reduced metabolic activity is not a hallmark of the quiescent state. Quiescent fibroblasts, relieved of the biosynthetic requirements associated with generating progeny, direct their metabolic activity to preservation of self integrity and alternative functions beneficial to the organism as a whole.
| Many cells in the human body are in a reversible state of quiescence, where they have exited the cell cycle but retain the capacity to re-enter it and divide again. Previous experiments in lymphocytes had suggested that quiescent cells reduce their glucose uptake and metabolic rate. In our studies, we have investigated the metabolism of fibroblasts, cells found in connective tissue and skin. Using “metabolomics” to monitor flux through metabolic pathways, we discovered that fibroblasts remain highly metabolically active even though they are not dividing. They degrade and resynthesize protein and fatty acid, and secrete large amounts of protein into the extracellular environment. Despite our expectation that quiescent cells would not have a high demand for nucleotide biosynthesis, we found that they do divert glucose to the pentose phosphate pathway, presumably to generate NADPH. The NADPH created may help the quiescent fibroblasts to detoxify free radicals or to synthesize fatty acids. Experiments in which we inhibited the pentose phosphate pathway resulted in increased apoptosis in quiescent cells, suggesting a possible strategy for selectively killing nondividing cells.
| Proliferating and quiescent cells are expected to have vastly different metabolic needs. Proliferating cells must replicate the entirety of their cellular contents in order to divide. As a result, much of the metabolic energy in a proliferating cell is devoted to synthesizing DNA, proteins, and lipids. Quiescent cells are relieved of this massive metabolic requirement since they are not dividing and, in several well-studied model systems, they decrease their metabolic rates in accordance with their decreased proliferation rates. Yeast cultures, for instance, enter stationary phase when liquid cultures are grown to saturation in rich medium [1]. Within this population, the quiescent yeast cells fail to accumulate mass and volume [2], in part because quiescent yeast cells induce autophagy, or self-cannibalism [3]. In addition, the overall transcription rate is three to five times slower in stationary-phase than in logarithmic-phase cultures [4], and protein synthesis is reduced to approximately 0.3% of the rate in logarithmically growing cultures [5]. Therefore, the quiescent cells within a stationary-phase culture of yeast likely represent an example of a quiescent cell that has significantly reduced its metabolic activity.
Lymphocytes also undergo a major metabolic shift upon transitioning between proliferation and quiescence. Early studies showed that lectin stimulation of lymphocytes led to increased glucose uptake, and an increased rate of glycolysis and pentose phosphate pathway (PPP) activities [6],[7]. More recent experiments have focused on a murine pro-B cell lymphoid cell line, FL5.12, that proliferates in response to the cytokine interleukin IL-3 [8]. IL-3 stimulation results in an 8-fold increased glycolytic flux. IL-3 also induces the cells to consume less oxygen per glucose consumed, and to excrete much more lactate, indicating a shift away from oxidative toward glycolytic metabolism. For human peripheral blood T lymphocytes, stimulation resulted in a 30-fold increase in glycolysis [9]; for thymocytes, the increase was 50-fold [10]. These differences in quiescent and proliferating lymphocytes have played a pivotal role in our understanding of the quiescent state, and experiments with lymphocytes as a model system have been important contributors to the development of the idea that quiescence is characterized by decreased metabolic activity. Lymphocytes, however, are relatively unique among mammalian cells in having primarily a “watching and waiting” function when quiescent and performing much of their physiological role only after activation.
Our studies focus on newborn dermal fibroblasts as a model system of quiescence [11]–[13]. In vitro, primary fibroblasts isolated directly from newborn foreskin can be induced into reversible quiescence by serum withdrawal or contact inhibition. Unlike most primary cells, fibroblasts remain healthy in culture in a quiescent state for as long as 30 d with little apoptosis or senescence, and can then re-enter the cell cycle [13]. In vivo, quiescent fibroblasts are central to normal physiology as the primary synthesizers of extracellular matrix necessary for the formation of cellular tissues. In response to a wound, fibroblasts enter the cell cycle from quiescence, proliferate, and secrete a collagen-rich extracellular matrix [14], pro-angiogenesis factors that recruit new blood vessels [13], and other molecules that facilitate the wound healing response [15]. A better understanding of the transition between proliferation and quiescence in fibroblasts would have broad implications for physiology and medicine. Scarring and fibrosis result from excessive fibroblast proliferation and secretion of extracellular matrix during and after wound healing [16],[17]. Additionally, tumors may contain quiescent cells that contribute to cancer dormancy [18],[19]. Thus, a better understanding of the transition between proliferation and quiescence, including the metabolic changes that occur, could have implications for a wide range of medical conditions.
The emerging field of metabolomics promises to augment our understanding of mammalian cell physiology through the systems-level characterization of cell-wide metabolite concentrations and fluxes. Using liquid chromatography–triple quadrupole mass spectrometry, we have developed a methodology for monitoring the pool size and turnover of a large number of metabolites simultaneously [20]–[22]. Here we apply metabolomic technology, flux analysis, and biochemical assays to investigate metabolic changes after primary dermal fibroblasts enter quiescence. We discovered that contact-inhibited primary fibroblasts remain highly metabolically active while adjusting their metabolic emphasis to produce NADPH, steadily renew their proteins and lipids, and enhance secretion of specific extracellular matrix proteins.
We have developed a model system that allows us to monitor metabolic differences between proliferating and quiescent cells. Primary dermal fibroblasts were expanded and analyzed while actively proliferating, after 1 wk of growth to confluence (contact-inhibited for 7 d [CI7]), after 2 wk of confluence (contact-inhibited for 14 d [CI14]), or after 2 wk of confluence with serum concentrations decreased for the final week from 10% to 0.1% (contact-inhibited for 14 d and serum-starved for 7 d [CI14SS7]). Alternatively, fibroblasts were plated sparsely so that they did not touch each other and induced into quiescence by serum starvation and monitored after 4 d (serum-starved for 4 d [SS4]) or 7 d (serum-starved for 7 d [SS7]). In quiescent fibroblasts, the fraction of cells with 2N DNA content increased so that 80% or more of the cells were in the G0/G1 phase of the cell cycle (Figure 1A). The fraction of cells in S phase was significantly reduced, indicating that very few cells were actively dividing under these conditions. In both contact-inhibited and serum-starved fibroblasts, levels of the cyclin-dependent kinase inhibitor p27Kip1 were upregulated, as expected for cells that entered quiescence (Figure 1B) [23]. In addition, staining with pyronin Y for total RNA indicated that the fraction of cells with low pyronin Y, interpreted as cells in G0 [24], increased in fibroblasts induced into quiescence by all of these methods (Figure 1C). Pyronin Y labeling data indicate that in the contact-inhibited and serum-starved cell populations investigated as quiescence models, approximately 60%–75% of the cells are in G0 and most of the remainder are in G1.
Previous studies have reported that lymphocytes induced to exit the cell cycle in response to mitogen withdrawal exhibit decreased glycolytic activity [8]. We used several methods to assess metabolic rates in proliferating, CI7, CI14, and CI14SS7 cells. We monitored the rates at which glucose and glutamine were consumed from the medium, and lactate and glutamate were secreted into the medium. As shown in Figure 2A, the rate of glucose consumption was approximately 2-fold lower in the contact-inhibited than in the proliferating fibroblasts. Lactate secretion decreased less than 2-fold with contact inhibition alone, and roughly 2-fold with additional serum deprivation. Glucose consumption actually slightly increased in fibroblasts induced into quiescence by serum starvation (without contact inhibition) for 4 or 7 d (Figure S1). We also monitored metabolic rates in fibroblasts cultured in medium conditions containing physiological levels of glucose and glutamine (1 g/l glucose and 0.7 mM glutamine compared with 4.5 g/l glucose and 4 mM glutamine in Dulbecco's Modified Eagle Medium [DMEM]) [25],[26]. Metabolic rates were somewhat lower in proliferating fibroblasts in these low glucose/low glutamine conditions compared with proliferating fibroblasts in standard medium (Figure S1). Quiescent fibroblasts cultured in these conditions exhibited consumption and excretion rates approximately half that of proliferating fibroblasts. Our finding that glycolytic rates are similar within a factor of two in proliferating and quiescent fibroblasts is surprising given that changes in glycolytic rate have been shown to mirror changes in proliferative rate in multiple model systems [8]–[10]. Indeed, while there is a dramatic decrease in the fraction of cells in the proliferative cell cycle, even the CI14SS7 condition resulted in only a 2-fold change in glucose consumption, much less than reported in other systems. Thus, decreased metabolic activity is not a universal hallmark of quiescence.
To further assess glycolytic rates in proliferating and contact-inhibited fibroblasts, we monitored the steady state pool sizes of glycolytic intermediates using liquid chromatography coupled to tandem mass spectrometry [20]–[22]. In total, we monitored the levels of 172 metabolites, 62 of which gave signals above background in proliferating, CI7, and CI14 fibroblasts. Metabolite levels were normalized per microgram of protein in cells plated at the same density because quiescent fibroblasts are smaller and contain less protein per cell than proliferating fibroblasts (E. M. Haley, A. L.-M., and H. A. C., unpublished data). The ratio of metabolite levels in the contact-inhibited (CI7 and CI14) to proliferating fibroblasts was determined for each metabolite. Some metabolites were present at consistently higher levels in proliferating fibroblasts, while others were enriched in contact-inhibited fibroblasts, although the magnitude of these changes in metabolite levels was generally modest (Figure S2).
Levels of five glycolytic intermediates and pentose-5-phosphate (a combination of ribose-5-phosphate, ribulose-5-phosphate, and xylulose-5-phosphate, which could not be reliably differentiated in our liquid chromatography–tandem mass spectrometry [LC-MS/MS] method) are shown in Figure 2B. No statistically significant differences were observed in the levels of glycolytic intermediates between contact-inhibited (CI7 or CI14) and proliferating fibroblasts at a false discovery rate of 0.05. Some glycolytic metabolites were present at lower levels in contact-inhibited, serum-deprived (CI14SS7) fibroblasts. Thus, the transition between proliferation and quiescence induced by contact inhibition alone has little effect on the pool sizes of glycolytic metabolites in primary fibroblasts. While pool sizes are not a direct indication of changes in flux, the constant levels of glycolytic metabolites in proliferating, CI7, and CI14 fibroblasts are consistent with our finding that there is little change in the rate of glucose uptake or lactate secretion among fibroblasts in these different states.
To more directly assess the rate of flux through glycolytic pathways, we incubated fibroblasts with [U-13C]-glucose and determined how quickly the label was incorporated into glycolytic intermediates (Figure 2C). For hexose phosphate (a combination of glucose-1-phosphate, glucose-6-phosphate, and fructose-6-phosphate), fructose-1,6-bisphosphate (FBP), dihydroxyacetone phosphate (DHAP), and phosphoenolpyruvate, the unlabeled pools of intermediates were converted into fully 13C-labeled intermediates at a similar rate in proliferating, CI7, and CI14 fibroblasts.
We also developed a computational model based on ordinary differential equations (ODEs) of central carbon metabolism for the proliferating, CI7, CI14, and CI14SS7 fibroblasts. The ODEs in the model quantify the isotope labeling dynamics of the relevant metabolites after switching into 13C-labeled carbon sources (Figure S3). Model parameters (i.e., metabolic fluxes and some unmeasured pool sizes) were identified by fitting all of the available laboratory data (labeling dynamics, pseudo-steady-state labeling patterns, measured pool sizes, and uptake and excretion rates). This systems-level approach enabled quantitation of flux through different metabolic pathways in proliferating, CI7, CI14, and CI14SS7 fibroblasts (Figure S4 and Table S1). For glycolysis, the inferred fluxes from hexose phosphate to FBP, and from DHAP to 3-phosphoglycerate, were similar in proliferating, CI7, and CI14 conditions (Figures 3 and S4 and Table S1; see Materials and Methods for information regarding statistical significance). In CI14SS7 fibroblasts, hexose phosphate–to-FBP and DHAP-to–3-phosphoglycerate fluxes are approximately half those in the other conditions (Figure S4 and Table S1), consistent with an approximately 2-fold reduction in glucose consumption. We conclude that glucose consumption and lactate excretion proceed rapidly in fibroblasts induced into quiescence by contact inhibition.
The PPP produces ribose-5-phosphate, needed for the biosynthesis of nucleotides, and NADPH, which can be used as a cofactor for the biosynthesis of macromolecules including fatty acids. We anticipated that proliferating cells would have higher demands for both ribose-5-phosphate and NADPH than quiescent cells, and thus higher PPP flux. Surprisingly, the pentose phosphate pool incorporated 13C label very rapidly in proliferating, CI7, and CI14 fibroblasts when the cells were incubated with labeled [U-13C]-glucose (Figure 4A). Indeed, according to our computational model, hexose phosphate–to–pentose phosphate flux was actually slightly higher in contact-inhibited (both CI7 and CI14) fibroblasts than in proliferating fibroblasts (though the effect was not statistically significant). Additional serum deprivation only slightly decreased oxidative PPP flux, with the oxidative PPP flux–to–glycolytic flux ratio highest in CI14SS7 fibroblasts. Thus, the oxidative PPP is actively utilized in both proliferating and quiescent cells.
We anticipated that ribose generated from the PPP would be incorporated into nucleotide triphosphates more rapidly in proliferating than quiescent cells because of their increased need for nucleotide triphosphates for RNA and DNA synthesis. Indeed, in proliferating fibroblasts, ATP and UTP with labeled ribose rings accumulate more rapidly in proliferating fibroblasts (Figure 4A). The results confirm that biosynthesis of nucleotides is more rapid in the proliferating cells.
Given that quiescent fibroblasts do not commit ribose phosphate to nucleotide biosynthesis, we reasoned that quiescent cells might recycle ribose phosphate back to glycolytic intermediates through the non-oxidative branch of the PPP. To test this hypothesis, we monitored the ratio of 1×13C-lactate to 2×13C-lactate after incubating the cells with [1, 2-13C]-glucose. As previously described [27], 1×13C-lactate is formed when glucose is metabolized through the oxidative portion of the PPP to ribulose-5-phosphate. In this pathway, glucose molecules lose one 13C atom in the form of CO2, and are then returned to glycolysis through the non-oxidative branch of the PPP (Figure 4B). 2×13C-lactate is formed by the canonical glycolysis pathway from glucose to lactate. The ratio of 1×13C-lactate to 2×13C-lactate provides an indication of the extent to which the non-oxidative branch of the PPP is utilized. This ratio is significantly higher in CI7 than proliferating fibroblasts, and even higher in CI14 fibroblasts (Figure 4C).
As another indication of the rate of flux through the non-oxidative branch of the PPP, we monitored labeling of sedoheptulose-7-phosphate, a metabolic intermediate in the non-oxidative PPP. Sedoheptulose-7-phosphate is labeled rapidly in CI7 and CI14 but not proliferating fibroblasts fed [U-13C]-glucose (Figure 4A), indicating higher flux through the non-oxidative branch of the PPP in quiescent cells. Our systems-level flux analysis confirmed increased flux from ribose phosphate back to glycolysis in contact-inhibited compared with proliferating fibroblasts (Figures 3 and S4 and Table S1). Thus, ribose phosphate generated from the PPP is utilized for nucleotide biosynthesis in proliferating fibroblasts but is recycled back to glycolytic intermediates in quiescent fibroblasts.
To investigate the mechanistic basis for the high PPP flux in quiescence fibroblasts, we monitored protein levels of two key enzymes in the PPP, both of which generate NADPH: glucose-6-phosphate dehydrogenase (G6PD) and 6-phosphogluconate dehydrogenase (PGD). Protein levels of both G6PD and PGD were elevated in fibroblasts induced into quiescence by either contact inhibition or serum starvation in comparison to proliferating fibroblasts (Figure 5A). These results suggest that contact-inhibited and serum-starved fibroblasts may activate a program that results in increased levels of PPP enzymes.
Both proliferating and quiescent fibroblasts generate NADPH through the PPP. The NADPH may be used for biosynthesis or to regenerate the reduced forms of glutathione or thioredoxin. We monitored reduced and oxidized glutathione (GSH and GSSG, respectively) in proliferating, CI7, CI14, and CI14SS7 fibroblasts. As shown in Figure 5B and 5C, GSH was slightly increased, and the ratio of GSH to GSSG significantly enhanced, in quiescent (CI7, CI14, and CI14SS7) compared with proliferating fibroblasts. The results are consistent with a model in which quiescent fibroblasts upregulate NADPH production in part to ensure adequate GSH as protection against free radicals [28].
We then tested the functional importance of the PPP in quiescent and proliferating fibroblasts. We incubated proliferating or CI14 fibroblasts with dehydroepiandrosterone (DHEA), a small molecule inhibitor of the PPP [29],[30] for 4 d and monitored the fraction of cells that were dead with propidium iodide (PI) labeling followed by flow cytometry. We discovered that the contact-inhibited fibroblasts exhibited a statistically significant increase in cell death compared with the proliferating fibroblasts from DHEA treatment at 100 µM and 250 µM doses (p<0.01) (Figure 6A). This result is particularly impressive given that almost all known metabolic inhibitors and cytotoxins preferentially kill proliferating cells [18],[31],[32]. Assaying for caspase-3/7 activity revealed that the mechanism of DHEA-induced cell death in the quiescent fibroblasts is via apoptosis (Figure 6B). The apoptosis-inducing effect of DHEA was significantly stronger in fibroblasts that were confluent for 11 d than in proliferating fibroblasts, and yet stronger in fibroblasts serum-starved for 7 d in the absence of contact inhibition.
Previous studies concluded that proliferating lymphocytes actively utilize glycolytic pathways to generate ATP while quiescent lymphocytes generate energy via an influx of fatty acids and proteins that are metabolized through the tricarboxylic acid (TCA) cycle [8]. To investigate TCA cycle usage, we monitored metabolite labeling through the TCA cycle after addition of [U-13C]-glucose, [3-13C]-glucose or [U-13C]-glutamine in proliferating, CI7, and CI14 fibroblasts. As shown in Figure 7, proliferating and contact-inhibited fibroblasts incorporate two carbon units from glucose into citrate via acetyl-CoA at comparable rates. In CI7 and CI14 fibroblasts, the labeled carbons progress through the TCA cycle to form 2×13C-α-ketoglutarate, as expected. In proliferating fibroblasts, however, there is a substantial decrease in the transmission of labeled carbons from citrate to α-ketoglutarate, succinate, and malate. Experiments using [U-13C]-glutamine further support the truncation of the TCA cycle (Figure 8). While carbon from glutamine effectively transverses the left side of the TCA cycle in the standard clockwise direction to yield 4×13C-citrate in both proliferating and quiescent fibroblasts, subsequent formation of 3×13C-α-ketoglutarate by isocitrate dehydrogenase hardly occurs in proliferating fibroblasts. The decreased flux from citrate to α-ketoglutarate in proliferating fibroblasts was confirmed via our systems-level flux identification (Figures 3 and S4 and Table S1).
When carbon skeletons are removed from the TCA cycle for the synthesis of macromolecular precursors including amino acids, other long carbon skeletons are needed to replace them. This anaplerotic refilling should be especially important for proliferating fibroblasts since their TCA cycle activity is truncated at citrate. The major anaplerotic reaction from glycolysis involves the carboxylation of pyruvate to form oxaloacetate. This reaction can be monitored by feeding cells [3-13C]-glucose and monitoring the fraction of citrate or malate with label, since the 13C is retained only when the anaplerotic reaction via pyruvate carboxylase is utilized. Surprisingly, the ratios of 1×13C-citrate to unlabeled citrate and/or 1×13C-malate to unlabeled malate were significantly increased in CI7, CI14, and CI14SS7 fibroblasts compared with proliferating fibroblasts (Table S2). In addition, quantitative flux analysis revealed that anaplerotic flux from pyruvate to oxaloacetate is elevated in CI7, CI14, and CI14SS7 compared with proliferating fibroblasts (Figure S4 and Table S1), while the flux from pyruvate to acetyl-CoA is lower in CI14 and CI14SS7 fibroblasts than in proliferating fibroblasts. Thus, contact inhibition was associated with both an increase in canonical TCA cycle activity past citrate, and an increase in anaplerotic TCA cycle flux from pyruvate to oxaloacetate. Proliferating fibroblasts, in contrast, seem unlikely to have sufficient carbon skeletons from glucose for the production of proteogenic amino acids not present in the cell growth medium.
We hypothesized that proliferating fibroblasts rely on another source for carbon skeletons. Supplementation with glutamine has been shown to be necessary for cultured cells, especially actively proliferating cells [33]–[36]. Accordingly, we monitored the rate of glutamine consumption by proliferating, CI7, CI14, and CI14SS7 fibroblasts (Figures 2A and S1). CI7, CI14, and CI14SS7 fibroblasts consume approximately half as much glutamine per microgram of protein as proliferating fibroblasts. CI7 and CI14 fibroblasts secrete glutamate at a lower rate compared with proliferating fibroblasts, and CI14SS7 fibroblasts secrete glutamate at a lower rate than CI7 or CI14 fibroblasts. SS4 and SS7 fibroblasts, on the other hand, consume glutamine and secrete glutamate at a faster rate than proliferating fibroblasts (Figure S1). The relative rate of glutamine consumption in proliferating versus CI14 fibroblasts in low glucose/low glutamine conditions is similar to that in standard medium. As shown in Figure 8, incubation of proliferating, CI7, and CI14 fibroblasts with [U-13C]-glutamine results in rapid labeling of glutamate, α-ketoglutarate, succinate, malate, and citrate, indicating that glutamine is used by both proliferating and contact-inhibited fibroblasts for TCA cycle anaplerosis. Since very few glucose carbons are incorporated into the TCA cycle in proliferating fibroblasts, glutamine may serve as the major anaplerotic precursor in proliferating fibroblasts [36]–[39].
[U-13C]-glutamine is converted into 5×13C-glutamate and subsequently to 5×13C-α-ketoglutarate. 5×13C-α-ketoglutarate can proceed through the TCA cycle in the forward direction to generate 4×13C-succinate, or, alternatively, it can be reductively carboxylated to 5×13C-citrate using NADPH as the electron source [40],[41]. As shown in Figure 8A, introduction of [U-13C]-glutamine led to conversion of approximately 15% of the citrate to the 5×13C form in proliferating, CI7, and CI14 fibroblasts by 8 h, with more rapid labeling in contact-inhibited fibroblasts. These results support a model in which there is both forward and reverse flux between citrate and α-ketoglutarate, with greater flux in both directions in contact-inhibited than in proliferating fibroblasts (Figures 3 and S4 and Table S1). The forward and reverse flux likely occur in different compartments, with α-ketoglutarate reductively carboxylated by isocitrate dehydrogenase 2 (IDH2) in the mitochondrion, and the resulting citrate reconverted to α-ketoglutarate by IDH1 in the cytosol [42]. As both IDH1 and IDH2 use NADP(H) as their redox cofactor, the net effect is transfer of high energy electrons in the form of NADPH to the cytosol. Consistent with greater flux through this pathway in contact-inhibited fibroblasts, IDH1 protein is increased by contact inhibition at the transcript and protein levels (Figure 8C). Thus, two major pathways to cytosolic NADPH, the PPP and the IDH2/IDH1 shuttle, are upregulated at both the protein and flux level in contact-inhibited fibroblasts.
Quiescent cells do not dilute out older macromolecules, organelles, or membranes with cell division, and thus may be more dependent than proliferating cells on mechanisms to break down and resynthesize membrane components and macromolecules. Our data are consistent with increased fatty acid degradation in contact-inhibited fibroblasts. Carnitine, a metabolite involved in the transport of fatty acids from the cytoplasm to the mitochondria during fatty acid degradation, is present at higher levels in CI7 and CI14 fibroblasts than in proliferating fibroblasts (Figure S2). Also, quantitative flux identification revealed, based on long-term labeling patterns of citrate, increased fatty acid breakdown in CI7 and CI14 fibroblasts, but lower rates of fatty acid breakdown in CI14SS7 fibroblasts (Figures 3 and S4 and Table S1).
The enhanced rate of fatty acid degradation in contact-inhibited fibroblasts seems to be enabling fatty acid biosynthesis to occur at a similar rate in proliferating and contact-inhibited fibroblasts. During fatty acid synthesis, citrate is transported out of the mitochondria to the cytoplasm, where it is broken down by ATP citrate lyase into oxaloacetate and acetyl-CoA used in fatty acid biosynthesis. ATP citrate lyase activity can be monitored based on the conversion of 5×13C-citrate to 2×13C-acetyl-CoA and 3×13C-oxaloacetate (measured as 3×13C-malate). As shown in Figure 8A, 3×13C-malate is produced similarly in proliferating, CI7, and CI14 cells, consistent with fibroblasts in all of these states being actively engaged in fatty acid biosynthesis. To more directly assess fatty acid biosynthesis in proliferating and quiescent fibroblasts, we extracted lipids from proliferating, CI7, CI14, and CI14SS7 fibroblasts fed [U-14C]-glutamine. The contribution of carbons to fatty acids from glutamine was significantly higher in all of the quiescent fibroblasts compared with the proliferating fibroblasts (Figure 8B), consistent with higher “backwards” flux from α-ketoglutarate to citrate (Figures 3 and S4 and Table S1). The higher levels of fatty acid synthesis in contact-inhibited fibroblasts may contribute to the maintenance of membrane integrity, and may also provide a major sink for cytosolic NADPH.
Similarly, our results suggest that contact-inhibited fibroblasts may also be actively degrading existing protein, and thus resynthesizing protein to replace the degraded proteins. As shown in Figure 9, the fraction of glutamate that is labeled in fibroblasts under all conditions increases rapidly after switching cells into [U-13C]-glutamine and then drops off in CI7 and CI14 fibroblasts, but not in proliferating fibroblasts. This decline in the fraction of glutamate molecules with five labeled carbons corresponds to an increase in the fraction of unlabeled glutamate. One possible explanation for these data is a breakdown of unlabeled proteins and release of free amino acids into the glutamate pool. These results are in agreement with our quantitative flux analysis: protein synthesis rates are similar across all conditions [43],[44] (Figures 3 and S4 and Table S1). Protein synthesis rates in the best-fit model are 3.3 nmol/min/µg protein for proliferating fibroblasts, 4.3 nmol/min/µg protein for CI7 fibroblasts, 4.1 nmol/min/µg protein for CI14 fibroblasts, and 2.9 nmol/min/µg protein for CI14SS7 fibroblasts. Thus, one reason for the active metabolism observed in contact-inhibited fibroblasts may be to rebuild and thus refresh their lipid and protein contents.
The high metabolic activity of quiescent fibroblasts might also be partially explained by their synthesis and secretion of extracellular matrix molecules needed for the structural integrity of tissue. While proliferating fibroblasts would be expected to secrete molecules important for wound healing [15], quiescent fibroblasts might be expected to secrete extracellular matrix molecules required at the end of a wound healing process or for maintenance of quiescent tissue [45]. We monitored the levels of secreted protein in conditioned medium collected from plates containing proliferating or CI14 fibroblasts [13]. Because serum interferes with immunoblotting for specific proteins, these experiments were performed in no serum and 0.1% serum conditions. As shown in Figure 9, the levels of fibronectin, collagen 21A1, and laminin alpha 2 in conditioned medium from CI14 fibroblasts was higher than the levels in conditioned medium from proliferating fibroblasts, thus demonstrating a biosynthetic commitment for contact-inhibited fibroblasts that may contribute to their high metabolic rate.
The metabolic profiles of proliferating and CI14 fibroblasts are summarized in Figure 3. Fibroblasts in both proliferating and contact-inhibited states utilize glycolysis extensively. Proliferating fibroblasts rely on the PPP to generate ribose for nucleotide biosynthesis and NADPH for biosynthetic purposes. Contact-inhibited fibroblasts employ the oxidative PPP to generate NADPH, and the carbon skeletons are largely returned to glycolysis as glyceraldehyde-3-phosphate and fructose-6-phosphate. Fibroblasts in both proliferating and contact-inhibited states contribute some glucose carbons to the TCA cycle. In contact-inhibited fibroblasts, carbons contributed by glucose are transmitted through the TCA cycle; in proliferating fibroblasts, there is little forward flux between citrate and α-ketoglutarate. Contact-inhibited fibroblasts rely more heavily on anaplerotic flux from pyruvate to oxaloacetate via pyruvate carboxylase; proliferating fibroblasts rely more heavily on glutamine, perhaps because of their higher demand for nitrogen. Glutamine drives the forward flux through the TCA cycle and also reverse flux from α-ketoglutarate to citrate, especially in the contact-inhibited fibroblasts. This reverse flux provides a mechanism for shuttling NADPH from mitochondria to the cytosol.
We discovered that fibroblasts induced into quiescence by contact inhibition maintain a high metabolic rate. In contact-inhibited fibroblasts, nucleotide biosynthesis is reduced, yet the rate of glycolytic, PPP, and TCA flux is almost completely maintained. Even fibroblasts that have been contact-inhibited for 2 wk and starved of serum for the final week show only a 2-fold reduction in glycolytic flux. Contact-inhibited fibroblasts also presumably generate substantial energy through the TCA cycle, where we observed flux of both glucose- and glutamine-derived carbons through more than a complete cycle. Consistent with these multiple routes of energy generation, the ATP/AMP ratio is high in contact-inhibited fibroblasts (Figure S2).
What then do the quiescent fibroblasts do with all of their energy? Our data suggest three avenues for energy utilization. First, contact-inhibited fibroblasts may continuously degrade and resynthesize their macromolecules and membrane components via increased autophagy [43],[44] (E. M. Haley, A. L.-M., and H. A. C., unpublished observation), a strategy that would help to ensure that old and potentially damaged macromolecules and membranes do not accumulate. Our data suggest that contact-inhibited fibroblasts may degrade protein and fatty acids at an enhanced rate compared with proliferating fibroblasts. The conclusion most consistent with our data is that the proliferating and contact-inhibited fibroblasts synthesize amino acids and fatty acids at rates that are comparable, with the new biomass contributing to new cells in proliferating fibroblasts and the new biomass replacing degraded molecules in the contact-inhibited fibroblasts.
Second, contact-inhibited and serum-starved fibroblasts induce pathways that generate NADPH. We discovered that three NADPH-generating enzymes, G6PD, PGD, and IDH1, are expressed at higher levels in quiescent than in proliferating fibroblasts. The results suggest that quiescent fibroblasts activate an NADPH-generating program of enzyme induction. One role of the NADPH may be to ensure the availability of GSH and thioredoxin for the detoxification of free radicals. Indeed, levels of total free radicals are lower in the contact-inhibited than in proliferating fibroblasts (E. M. Haley and H. A. C., unpublished data). Another role for the NADPH generated may be to support resynthesis of fatty acids, as fatty acid degradation yields NADH while synthesis requires NADPH.
Third, quiescent fibroblasts may acquire new cell-type-specific functions. In contrast to lymphocytes, which, with the exception of plasma cells, lack a major biosynthetic function in their quiescent state, fibroblasts secrete proteins and other molecules needed for the extracellular matrix even when they are quiescent. Contact-inhibited fibroblasts direct some of their metabolic activity toward this biosynthetic purpose, as we observed elevated levels of specific extracellular matrix proteins in contact-inhibited compared with proliferating fibroblasts (Figure 10). Thus, quiescent fibroblasts, relieved of the biosynthetic requirements associated with creating progeny, can turn their protein synthesis machinery toward the synthesis of proteins that are beneficial for the organism as a whole.
Our findings shed light on some larger questions about quiescence: What are the fundamental attributes of a quiescent state? Is there a single quiescent state or are there multiple quiescent states? Our results suggest that quiescence is not necessarily associated with a shutdown of glycolysis, as reported for lymphocytes and thymocytes [6]–[10]. Quiescent cells can actually be highly metabolically active. In this respect, quiescent fibroblasts resemble terminally differentiated cells like cardiomyocytes, neurons, and renal tubular epithelial cells, which are among the highest energy consumers in mammals. These terminally differentiated cells are well-known to employ nutrients to achieve their contractile, signaling, and transport functions. Whether their metabolic activity, like that of contact-inhibited fibroblasts, is also directed to continuously refreshing their protein and lipid composition merits further study.
In addition to differing from quiescent lymphocytes, different types of quiescent fibroblasts can vary. While CI7, CI14, and CI14SS7 fibroblasts are indistinguishable morphologically or by traditional cell cycle analysis, they differ with regard to their metabolic profiles. Compared with fibroblasts induced into quiescence by contact inhibition, fibroblasts also deprived of serum exhibited a decrease in lactate excretion rates, smaller pool sizes of glycolytic intermediates, and decreased flux from pyruvate to acetyl-CoA. Our findings suggest that cells of different types may actually be in distinct quiescent states, and may have discovered distinct solutions to the metabolic challenges associated with quiescence.
Finally, our findings suggest that contact-inhibited and serum-starved fibroblasts are particularly susceptible to apoptosis induced by treatment with DHEA, a pentose phosphate pathway inhibitor. The ability to selectively kill quiescent cells could have therapeutic potential [46],[47]. For instance, tumor stem cells may exist in a quiescent state for years, while retaining the capacity to emerge from dormancy, proliferate, and initiate a tumor recurrence. Small molecules that target the pathways invoked by these cells to facilitate their survival during dormancy could be useful additions to our therapeutic arsenal. We discovered that contact-inhibited and serum-starved fibroblasts rely on the PPP and possibly other NADPH-generating reactions for viability. Small molecule inhibitors like DHEA might ultimately prove valuable for targeting quiescent tumor cells.
Primary human fibroblasts were isolated from foreskin as previously described (see Supplemental Data in [48]). Fibroblasts were maintained in DMEM (Hyclone, Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (Hyclone) and 100 µg/ml penicillin and streptomycin (Invitrogen). Cells were collected while proliferating, after 1 wk of confluent maintenance (CI7), after 2 wk of confluent maintenance (CI14), after 2 wk of maintenance with the last 7 d in 0.1% serum (CI14SS7), and after serum starvation in 0.1% serum for 3 d, 4 d (SS4), or 7 d (SS7). Cells made quiescent by serum starvation alone were plated sufficiently sparsely so that they did not contact surrounding cells. Medium was changed every 2 d. Proliferating cells were sampled the day after seeding. In order to better simulate conditions in vivo, we also used low glucose/low glutamine conditions in which the glucose level was 1 g/l and the glutamine level was 0.7 mM, compared with a glucose level of 4.5 g/l and a glutamine level of 4 mM in standard DMEM. While cells were confluent, the medium was changed regularly. For analysis, cells were transferred to DMEM (Invitrogen) with 7.5% dialyzed fetal bovine serum (Atlanta Biologicals or Hyclone) the day before the experiment. Fibroblasts were photographed through a Nikon Eclipse TS100 microscope using a Scion 8-bit color firewire 1394 digital camera. Images were captured with Scion VisiCapture software (Scion).
Cells were trypsinized and collected into phosphate-buffered saline (PBS) containing 5% bovine growth serum (Hyclone). Cells were pelleted, resuspended in 67% ethanol in PBS, and stored at 4°C. For flow cytometry, cells were pelleted, washed with PBS, and resuspended in PBS with PI (40 µg/ml) (VWR) and RNAse A (200 µg/ml) (Thermo Fisher Scientific). Samples were incubated in the dark for 1 h at room temperature, and analyzed using a FACSort flow cytometer (BD Biosciences). The PI was excited at 488 nm, and emitted fluorescence was collected on detector FL2 with a bandpass filter of 585/42 nm. At least 20,000 cells were collected and analyzed with CellQuest software (BD Biosciences). Cell cycle distributions were calculated with ModFit LT software using the Watson Pragmatics algorithm.
To differentiate cells in G0 versus G1, fibroblasts representing each quiescence condition were trypsinized and suspended in cold Hank's buffered saline solution at a concentration of 2×106 cells/ml, then added to a fixative of ice-cold 70% ethanol. Cells were fixed for at least 2 h, washed, and resuspended at 4×106 cells/ml. A solution of 4 µg/ml pyronin Y and 2 µg/ml Hoechst 33342 was added to the cell suspension and incubated on ice for 20 min before measuring cell cycle status by flow cytometry. To determine RNA content, pyronin Y was excited at 488 nm and emission was measured at 562–588 nm. DNA content was determined by Hoechst 33342. Excitation was measured at 355 nm and emission was measured at 425–475 nm. Cells in G0 were identified as the population with 2N DNA content and an RNA content lower than the level in S phase [49].
Cells were made quiescent by contact inhibition, serum starvation, or a combination as indicated in the text or figure, and collected at the indicated times. The cells were lysed in RIPA buffer (50 mM Tris-Cl [pH 7.4], 150 mM NaCl, 1% Triton X-100, 1% sodium deoxycholate, and 0.1% SDS) containing protease and phosphatase inhibitors (10 mM NaPO4 [pH 7.2], 0.3 M NaCl, 0.1% SDS, 1% NP40, 1% Na deooxycholate, 2 mM EDTA, protease inhibitor cocktail [Roche, Basel, Switzerland] and Halt Phosphatase inhibitors [Thermo Fisher Scientific]). Lysates were sonicated with five pulses for 15 s each at 60 J/W. Lysates were then incubated for 30 min on ice with periodic vortexing and cleared by centrifugation for 2–5 min at 4°C at 10,000 rpm. Total protein amount was assessed by the Lowry method using the Bio-Rad DC Protein Assay Kit II (Bio-Rad) as described by the manufacturer. Spectrophotometer readings taken at 650 nm were compared against a standard curve to determine lysate concentration. Total protein content was determined as the product of lysate concentration and lysate volume. Equal amounts of total cellular proteins were resolved on 12% SDS-PAGE and electro-transferred onto a PVDF membrane. Membranes were blocked for 1 h at room temperature in Tris-buffered saline containing 0.1% Tween-20 (TBS-T) (10 mM Tris [pH 7.6], 15 mM NaCl, and 0.1% Tween-20) or phosphate-buffered saline containing 0.1% Tween-20 (PBS-T) containing 5% nonfat dried milk. Membranes were incubated with antibodies to p27 (1∶500 diluted in TBS-T/5% milk) (Santa Cruz Biotechnology), IDH1 (1 µg/ml diluted in PBS-T/1% milk) (Lifespan Biosciences), G6PD (1∶1,500 diluted in PBS-T/1% milk) (Novus Biologicals), or PGD (1∶1,000 diluted in PBS-T/1% milk) (GeneTex) overnight. Following incubation, the membranes were washed three times in TBS-T or PBS-T and incubated for 1 h with horseradish peroxidase–conjugated anti-rabbit secondary antibody (1∶3,000 diluted into TBS-T/5% milk for p27 or 1∶10,000 diluted in PBS-T/1% milk for IDH1 and G6PD) (GE Healthcare). The membranes were washed three times with TBS-T or PBS-T, and immunoreactive bands were detected with an enhanced chemiluminescence kit (Pierce, Thermo Scientific). The membranes were stripped using Restore Western Blot Stripping Buffer (Thermo Scientific) according to the manufacturer's instructions and immunoblotted with GAPDH (Abcam) (1∶5,000 dilution) in PBS-T/1% milk or TBS-T/5% milk as a loading control.
Highly parallel measurement of intracellular metabolites was performed as previously described [21]. Metabolites were extracted from proliferating, CI7, CI14, or CI14SS7 cells by aspirating the medium from the plate and flash-quenching metabolic activity with 80% methanol maintained at −80°C. Cells were incubated in methanol for 15 min, scraped on dry ice, and pelleted with centrifugation at 4,400 rpm for 5 min. Samples were re-extracted twice with 80% methanol on dry ice. The three extractions were pooled and dried under nitrogen gas, dissolved in 300 µl of 50% methanol, and spun at 13,000× g for 5 min. Methanol supernatant was then passed through an aminopropyl column [50]. Eluate from the column was analyzed with positive ion mass spectrometry via a Finnigan TXQ Quantum Ultra triple-quadrupole mass spectrometer equipped with an electrospray ionization source (Thermo Fisher Scientific) [22]. A TSQ Quantum Discovery MAX mass spectrometer, also equipped with an electrospray ionization source, was used to collect data on negative mode ions after separation on a 25-cm C18 column coupled with a tributylamine ion pairing agent to aid in the retention of polar compounds [51],[52].
To quantify metabolites, peak heights were initially assigned using XCalibur software (Thermo Fisher Scientific) and then evaluated manually. Metabolites enriched at least 5-fold in a sample compared with a control plate containing only medium were retained in the analysis. Of the 172 metabolites monitored, 62 met these criteria. Signals that were below the limit of detection were assigned 100. Metabolite levels were normalized by the amount of protein present.
To monitor the flux through metabolic pathways, samples were incubated with medium containing isotope-labeled nutrient for different amounts of time. Dulbecco's medium lacking glucose and glutamine was isotope-labeled by adding back glucose or glutamine ([U-13C]-glucose, [1, 2-13C]-glucose, [3-13C]-glucose, or [U-13C]-glutamine; Cambridge Isotope Laboratories) to a final concentration of 4.5 g/l glucose or 0.584 g/l glutamine. Samples were taken at the indicated time points after medium change and processed as described above. Levels of 12C and 13C forms of metabolic intermediates were monitored with LC-MS/MS [53].
Medium was sampled from cells under a variety of conditions: proliferating, CI7, CI14, CI14SS7, SS4, SS7, low glucose/low glutamine proliferating, and low glucose/low glutamine CI14. Conditioned medium was sampled over a time course from 0 to 96 h for fibroblasts, depending upon the experiment. The levels of glucose, lactate, glutamine, and glutamate were measured using a YSI 7100 Select Biochemistry Analyzer (YSI Incorporated). The rate of glucose consumption, lactate excretion, glutamine consumption, and glutamate excretion was determined as the rate that these metabolites appeared or disappeared from the medium divided by the time integral of the protein mass of cells on the plate during that time period.
The total GSH and GSSG content of proliferating, CI7, CI14, and CI14SS7 fibroblasts were determined using Cayman Chemical's Glutathione Assay Kit according to the manufacturer's instructions (Cayman Chemical). Cayman's GSH assay kit employs a carefully optimized enzymatic recycling method, using glutathione reductase for the quantification of GSH. Briefly, cells were harvested using a cell lifter in 1.5 ml of cold buffer (i.e., 50 mM MES or phosphate buffer [pH 6–7] containing 1 mM EDTA) and were centrifuged at 10,000× g for 15 min at 4°C, followed by metaphosphoric acid deproteinization and addition of triethanolamine solution. Half of the samples were then treated with 2-vinylpyridine to allow quantification of the GSSG pool exclusively. Assay Cocktail (a mixture of 2-(N-morpholino) ethanesulfonic acid Buffer [11.25 ml], reconstituted Cofactor Mixture [0.45 ml], reconstituted Enzyme Mixture [2.1 ml], water [2.3 ml], and reconstituted 5,5′-dithiobis-(2-nitrobenzoic acid) [0.45 ml]) was added, and total GSH and GSSG in the deproteinated samples were measured at 405 nm in a spectrophotometer. GSH concentration of the samples was determined by the endpoint method and expressed in micromolar concentrations.
Proliferating and CI14 fibroblasts were treated with DHEA dissolved in ethanol or dimethylsufoxide (0.1% vol/vol) for 4 d. On the fourth day of treatment with the inhibitor, cells were trypsinized and collected into conditioned medium. Cells were then centrifuged for 5 min at 1,000 rpm. The supernatant was aspirated and cells were taken up in PBS with 1 µg/ml PI (VWR). Cells were kept on ice and immediately analyzed by flow cytometry using a BD LSRII multi-laser analyzer (BD Biosciences). PI was excited at 488 nm, and emitted fluorescence was collected through a 610/20 bandpass filter. At least 40,000 cells were collected and analyzed with FACSDiVa software (BD Biosciences). PI-negative cells were counted as live cells, and PI-positive cells were counted as dead cells.
Apoptosis was measured based on the levels of caspase-3/7 released into the medium using the ApoTox-Glo Triplex Assay according to the manufacturer's instructions (Promega). Cells were plated in triplicate at 10,000 cells per well in white-walled, clear-bottom 96-well plates (Costar, Corning Life Sciences). For contact inhibition, cells were plated 7 d prior to the start of treatment; for serum starvation, cells were plated 4 d prior to treatment and switched to 0.1% serum medium for the remaining 3 d; proliferating cells were plated the day prior to the start of treatment. Increasing concentrations of DHEA or ethanol vehicle alone were added to the medium in each well, and treatment proceeded for 4 d. Cells in serum starvation conditions were incubated in 0.1% serum during treatment as well. The apoptosis reagent was added at 100 µl per well and incubated for 1 h prior to reading. Luminescence was read from the top using a Synergy-2 plate reader (Biotek). Luminescence data were normalized to the vehicle only condition.
Lipid synthesis from glutamine was measured using a modified version of a previously published protocol [53]. Briefly, proliferating, CI7, CI14, and CI14SS7 fibroblasts were incubated in medium containing 5 µCi/ml [U-14C]-glutamine at 4 mM (0.4% labeled). After incubation for 24 h, the culture medium was aspirated, cells were washed with PBS, and phospholipids were extracted by addition of 500 µl of 3∶2 hexane∶isopropanol. The culture dishes were then washed with an additional 500 µl of the hexane∶isopropanol mixture. The resulting total extract was dried using a speed-vac, resuspended in 500 µl of 1 N KOH in 90∶10 methanol∶water, and incubated at 70°C for 60 min to saponify lipids. Sulfuric acid (100 µl, 2.5 M) was then added, followed by hexane (700 µl) to extract the saponified fatty acids. The organic and aqueous phases were separated by centrifugation and scintillation-counted.
To monitor gene expression levels, proliferating, CI7, or CI14 fibroblasts were trypsinized, removed from the plate, pelleted, and stored at −80°C. Total RNA was isolated using the mirVana miRNA Isolation kit (Ambion) according to the manufacturer's instructions. RNA quality was verified using a Bioanalyzer 2100 (Agilent Technology), and the amount was determined with a NanoDrop spectrophotometer (NanoDrop Technologies). Total RNA (325 ng) was amplified using the Low RNA Input Fluorescent Labeling Kit (Agilent Technologies) according to the manufacturer's protocol. Cy-3 (PerkinElmer) was directly incorporated into the cRNA from proliferating cells during in vitro transcription. Cy-5 was incorporated into complementary RNA from CI7 or CI14 fibroblasts. Mixtures of Cy-3-labeled and Cy-5-labeled cRNA were co-hybridized to Whole Human Genome Oligo Microarray slides (Agilent Technologies) at 60°C for 17 h and subsequently washed according to the Agilent Technologies standard hybridization protocol. Slides were scanned with a dual laser scanner (Agilent Technologies). Images were monitored for quality control. The Agilent Technologies feature extraction software, in conjunction with the Princeton University MicroArray database (http://puma.princeton.edu/), was used to compute the log ratio of the two samples for each gene after background subtraction and dye normalization. The entire experiment was performed twice.
For the analysis of extracellular matrix proteins in conditioned medium, we could not perform the experiments in the presence of high amounts of serum because serum inhibited protein transfer after immunoblotting. As previously described [13], proliferating fibroblasts were conditioned at low cell density in the presence of platelet-derived growth factor with either no serum or 0.1% serum. Quiescent fibroblasts were cultured at high density in the absence of platelet-derived growth factor with either no serum or 0.1% serum. Medium was conditioned over 4 d and during that time, protein lysates were collected over a time course. The protein content of the cell lysates was plotted against the time of lysate collection. A curve that fit the data was generated and the area under the curve, the integrated protein–hour quantity, was divided by the volume of medium collected from the proliferating or quiescent plate. The total protein–hour/volume for each sample was used to adjust the volume of conditioned medium, which was then mixed with 25% volume of trichloroacetic acid (Sigma-Aldrich) containing 0.1% sodium deoxycholate (Sigma-Aldrich), and incubated for 30 min on ice. Following centrifugation, samples were washed 3–4 times with −20°C acetone, resuspended in sodium dodecyl sulfate-polyacrylamide gel electrophoresis sample buffer and separated under reducing conditions on 5% (for fibronectin and COL21A1) or 12% (for LAMA2) sodium dodecyl sulfate-polyacrylamide gels. Proteins were transferred for 1 h at 100 V to Westran polyvinylidene fluoride membranes (PerkinElmer). Membranes were blocked for 1 h at room temperature in 5% nonfat dried milk in PBS-T. Membranes were then incubated overnight at 4°C with a mouse monoclonal anti-fibronectin clone HFN7.1 (1∶2,000 dilution, generous gift of Jean Schwarzbauer, Princeton University), mouse polyclonal antibody against COL21A1 (1∶750 dilution, Abcam), or mouse monoclonal antibody against LAMA2 (3 µg/ml, Abnova) diluted in PBS-T/1% milk. Following overnight incubation in the primary antibody, membranes were washed three times in PBS-T, incubated for 1 h in a 1∶10,000 dilution of horseradish peroxidase–conjugated sheep anti-mouse secondary antibody (GE Healthcare) in PBS-T/1% milk. Membranes were exposed to X-ray film, and film was scanned with a Hewlett-Packard Scanjet 4890 using Hewlett-Packard software. The intensity of individual bands was determined with ImageJ analysis software.
Fluxes were determined by integration of all available forms of experimental data within a quantitative flux-balanced framework using the same strategy as described in Munger et al. [53]. An ODE model (Figure S3) of central carbon metabolism was constructed. The model assumes steady-state, mass-balanced flux and simulates the resulting labeling dynamics after switching cells from unlabeled medium to uniformly 13C-labeled glucose or glutamine. The model consists of 55 ODEs, describing the rate of loss of unlabeled metabolites and the rate of accumulation of labeled metabolites. It builds upon the previously described model [53] with a few changes. An exchange flux (F12) was introduced in glycolysis between DHAP and FBP. Backward flux (F11) from α-ketoglutarate to citrate, together with a latent citrate pool that is never labeled (determined by the lowest unlabeled citrate pool size observed in all experiments), was introduced in the TCA cycle. The latent citrate pool was added because for citrate, but not other metabolites, a substantial fraction of the pool (approximately 40% for the proliferating cells) did not label over the course of the experiment. Beyond labeling dynamics, additional input data included metabolite levels, rates of metabolite consumption and excretion, and the glycolysis–PPP flux convergence ratio determined after feeding [1, 2-13C]-glucose for 2 h. Model parameters (fluxes, as well as pool sizes of a small number of metabolites that could not be directly experimentally measured) were identified by a genetic algorithm that minimizes a cost function defined as the sum of weighted differences between the experimental data and computational results (Table S3) [54]. As a global search algorithm, the genetic algorithm computationally probes for alternative flux solutions consistent with the experimental results. For each cell type, the algorithm was run until 1,000 consistent solutions (i.e., parameter sets that produced the lowest cost values when the algorithm reached convergence) were obtained. The distribution of the 1,000 values was then used to quantitatively represent each identified parameter. Since the distributions are not Gaussian, a flux is considered quantitatively different between proliferating and quiescent cells only when the distributions from the proliferating and quiescent fibroblasts do not overlap. This measure minimizes the false positives that may occur when only one or a few solutions are identified [54]. Although qualitatively supportive of the model-inferred enhancement of anapleurotic flux from glucose in quiescent fibroblasts, labeling data for [3-13C]-glucose, which was taken at 8 h, were quantitatively inconsistent with the other labeling data, which covered the first 2 h of incubation only. The [3-13C]-glucose data were accordingly excluded from the computational analysis. The computer code is available upon request.
The Entrez Gene (http://www.ncbi.nlm.nih.gov/gene) accession numbers for the proteins discussed in this paper are G6PD, 2539; IDH1, 3417; IDH2, 3418; and PGD, 5226.
|
10.1371/journal.pgen.1006015 | Recombination-Independent Recognition of DNA Homology for Repeat-Induced Point Mutation (RIP) Is Modulated by the Underlying Nucleotide Sequence | Haploid germline nuclei of many filamentous fungi have the capacity to detect homologous nucleotide sequences present on the same or different chromosomes. Once recognized, such sequences can undergo cytosine methylation or cytosine-to-thymine mutation specifically over the extent of shared homology. In Neurospora crassa this process is known as Repeat-Induced Point mutation (RIP). Previously, we showed that RIP did not require MEI-3, the only RecA homolog in Neurospora, and that it could detect homologous trinucleotides interspersed with a matching periodicity of 11 or 12 base-pairs along participating chromosomal segments. This pattern was consistent with a mechanism of homology recognition that involved direct interactions between co-aligned double-stranded (ds) DNA molecules, where sequence-specific dsDNA/dsDNA contacts could be established using no more than one triplet per turn. In the present study we have further explored the DNA sequence requirements for RIP. In our previous work, interspersed homologies were always examined in the context of a relatively long adjoining region of perfect homology. Using a new repeat system lacking this strong interaction, we now show that interspersed homologies with overall sequence identity of only 36% can be efficiently detected by RIP in the absence of any perfect homology. Furthermore, in this new system, where the total amount of homology is near the critical threshold required for RIP, the nucleotide composition of participating DNA molecules is identified as an important factor. Our results specifically pinpoint the triplet 5'-GAC-3' as a particularly efficient unit of homology recognition. Finally, we present experimental evidence that the process of homology sensing can be uncoupled from the downstream mutation. Taken together, our results advance the notion that sequence information can be compared directly between double-stranded DNA molecules during RIP and, potentially, in other processes where homologous pairing of intact DNA molecules is observed.
| Recombination-independent pairing of homologous double-stranded DNA molecules is associated with many important biological processes including the alignment of homologous chromosomes in early meiosis, monoallelic gene expression in mammals, and somatic pairing of homologous chromosomes in Drosophila. The molecular mechanism(s) by which homologous sequences are recognized in all of these cases remain(s) largely unknown. The phenomenon of Repeat-Induced Point mutation (RIP) in the filamentous fungus Neurospora crassa provides an especially advantageous model system for elucidating the general mechanism of recombination-independent recognition of DNA homology. Here we show that imperfect (interspersed) homologies with overall sequence identity of only 36% can be efficiently detected by RIP in the absence of nearby perfect homology. Under these particularly stringent conditions, specific DNA sequence motifs are found to play a critical role. Our analysis of one such situation identifies the triplet 5'-GAC-3' as an especially favorable unit of homology. We also present experimental evidence that the process of homology recognition for RIP integrates sequence information over hundreds of base-pairs of chromosomal DNA, and that this process can be uncoupled from the downstream mutation. Taken together, our results advance the notion that sequence information can be compared directly between double-stranded DNA molecules during RIP and, potentially, in other processes where homologous pairing of intact DNA molecules is observed.
| Many biological systems exhibit specific co-localization (pairing) of homologous DNA molecules independent of recombination-based mechanisms. A paradigmatic example is provided by the persistent association of homologous chromosomes in somatic nuclei of the Diptera insects [1]. Somatic pairing of homologous loci has been documented in yeast (references in [2]) and in mammals (references in [3,4]). In early meiosis, significant pairing of homologous chromosomes occurs prior to or in the absence of recombination in mice, flies, worms, fission yeast and filamentous fungi (references in [2]). The molecular mechanisms by which homologous sequences are recognized in all of these cases remain largely unknown [5]. Roles of direct interactions between DNA molecules or the indirect readout of the base-pair sequence by RNA or proteins have been proposed [6,7].
In addition to the physical co-localization, recombination-independent interactions between homologous DNA molecules can trigger their chemical modification. In filamentous fungi, two closely-related processes can detect duplicated DNA sequences irrespectively of their origin and location in the genome [8]. Once detected, DNA duplications can undergo cytosine methylation (the phenomenon of Methylation Induced Premeiotically, MIP, discovered in Ascobolus immersus [9]) or cytosine-to-thymine mutation (the phenomenon of Repeat-Induced Point mutation, RIP, discovered in Neurospora crassa [10]). Both RIP and MIP are restricted to the sexual phase of the fungal life-cycle and occur in haploid germline nuclei that prepare to enter meiosis. Both processes are mediated by specialized cytosine methyltransferases of the RID/Masc1 family [11,12]. Although RIP coincides with a period of increased intrachromosomal recombination [13], it does not require MEI-3, the only RecA homolog in Neurospora [14]. No DNA sequence, with the exception of ribosomal DNA in the nucleolar organizer region, can escape RIP [15]. The general and efficient nature of RIP makes it an especially attractive experimental system for elucidating the general mechanism of recombination-independent recognition of DNA homology.
Previously, we developed a sensitive genetic assay for RIP based on the quantitative analysis of individual mutations induced by a pair of closely-positioned DNA repeats [14]. Using this experimental system, we demonstrated that as few as 155 base-pairs of perfect homology could trigger RIP, and that 220 base-pairs of perfect homology could promote substantial RIP. We further found that the number of RIP mutations accurately reflected the length of homology for repeats ranging between 220 and 520 base-pairs. We then explored the ability of RIP to recognize different patterns of imperfect homology by extending the 220-bp region of perfect homology with additional 200 base-pairs of synthetic DNA that provided partial homology in the adjacent region. Here we discovered that RIP could detect weak partial homologies organized as arrays of base-pair triplets interspersed with a matching periodicity of 11 or 12 base-pairs along participating chromosomal segments. These and other observations led us to propose that sequence homology could be sensed by direct interactions between intact, slightly underwound double-stranded DNA molecules [14].
Our previous studies examined recognition of interspersed homologies in the repeat system that always included the 220-bp region of adjacent perfect homology, which was expected to provide a strong, persistent point of interaction. Those studies did not give any indication that RIP might be sensitive to the actual base-pair sequence of homologous units. In the present study, we have focused on the analysis of interspersed homologies spanning 500 base-pairs. We demonstrate that interspersed homologies with overall sequence identity of only 36% can be efficiently detected by RIP in the absence of perfect homology; yet short regions of adjacent perfect homology can have a strong activating effect. We have dissected the sequence requirements for RIP within one such short region of adjacent homology, under conditions where the overall amount of homology in the repeat system remains near the minimum threshold required for RIP. Analysis under these particularly stringent circumstances reveals that the base-pair composition of participating sequences can play a critical role in RIP and identifies one candidate trinucleotide, 5'-GAC-3', as an especially efficient unit of homology. We also present experimental evidence suggesting that RIP integrates sequence information over hundreds of base-pairs, and that the actual process of homology sensing can be distinguished from ensuing mutation. These and other results further illuminate the properties of DNA homology recognition during Neurospora RIP, paving the way for more mechanistic studies in the future.
We previously developed a sensitive quantitative assay for RIP based on the analysis of individual mutations induced by a pair of closely-positioned DNA repeats [14]. The sequence of the “left” repeat (so-called "Reference") was always held constant, while the sequence of the “right” repeat (so-called "Test") was varied as desired (Fig 1A). To examine the capacity of different interspersed homologies to trigger RIP, we designed test sequences that contained short units of homology of length X separated by units of non-homology of length Y. The periodicity with which homologous units occurred along a particular pair of reference/test sequences is defined by the sum of these two distances (X+Y). Our previous research showed that only certain periodicities (X+Y = 11 and X+Y = 12) could promote RIP in conjunction with an adjacent region of perfect homology [14]. Because an array of homologous units can be positioned in a particular relationship with respect to the reference sequence, an additional parameter Z is now introduced to define the sequence position of the first homologous unit (Fig 1B). Patterns of interspersed homology investigated in this study are thus represented as XH-YN_Z (Fig 1B). Because a particular reference region is selected arbitrarily from the Neurospora genome, the actual nucleotide sequences of homologous units are expected to be different for every combination of X, Y, and Z. However, by keeping X and Y constant and only changing the sequence position parameter Z, the sequences of homologous units can be altered without disrupting basic homology-block structure (Fig 1C). The present study shows that such sequence differences can be important for the efficiency of homology recognition for RIP.
In our previous work, we analyzed mutation of the 500-bp interspersed homology 4H-7N_1 created by replacing the cyclosporin-resistant-1 gene (csr-1) with synthetic DNA that was designed to be partially homologous to an arbitrarily selected reference region in a nearby gene (Fig 2A). In that study, we found that 4H-7N_1 alone exhibited barely detectable RIP but that, when a 337-bp region of perfect homology was present at an adjacent position, strong mutation of 4H-7N_1 was observed [14]. The present study was initiated to further define the role of perfect homology in promoting mutation of the interspersed homology 4H-7N_1. As in our previous analysis, the mean number of mutations (per spore) in the entire sequenced region was used as a quantitative measure of RIP. Empirical distributions of mutation counts were compared by the Kolmogorov-Smirnov test.
Systematic analysis shows that combining 4H-7N_1 with at least 50 base-pairs of perfect homology significantly increases RIP, and that 100 base-pairs of perfect homology promote very efficient RIP (Fig 2B–2D). A similar effect is observed regardless of whether the fully-homologous region is added to the "left" or the "right" end of the partially homologous region, implying that two different 100-bp sequences, at two different positions in the repeat construct, are equivalently effective (Fig 2D: compare v and vi). We further find that the insertion of 11 base-pairs of random DNA sequence at the junction between 4H-7N_1 and the fully-homologous region leads in a partial reduction in RIP, but that substantial mutation still occurs despite the apparent physical impediment to overall alignment (Fig 2D: compare v and vii). A similar effect is conferred by replacing 22 base-pairs of perfect homology by 22 base-pairs of non-homology in the middle of the 100-bp region, thereby interrupting the 100-bp perfect homology into two segments of 38 and 40 base pairs (Fig 2D: compare vii and viii).
We similarly examined the effect of adjacent perfect homology on mutation of another interspersed homology, 5H-6N_1 (Fig 3). This new homology pattern relates to the same reference sequence, and has the same periodicity (X+Y) and sequence position (Z) as 4H-7N_1, but contains homologous units of five base-pairs (X = 5) instead of four. We find that 5H-6N_1 alone can trigger substantial RIP, demonstrating a significant improvement over 4H-7N_1 (Fig 3). This observation provides a new conclusion: given an appropriate pattern of interspersed homology, perfect homology is dispensable for RIP.
We further find that, despite the substantial intrinsic activity of 5H-6N_1, the addition of only 15 base-pairs of perfect homology (representing 3% of the total homology length) increases RIP nearly 2.5-fold (from 3.6±0.9 to 9.0±1.5 mutations per spore, Fig 3A). This result has two implications. First, the presence of short perfect homology can still be highly advantageous even against the substantial background activity of 5H-6N_1. Second, such a strong response to 15 base-pairs of perfect homology contrasts with a negligible effect of combining the same 15-bp sequence with the interspersed homology 4H-7N_1 (Fig 3A and 3B). This difference might be explained by a synergistic interaction of the fully- and partially-homologous regions, such that more activity in the partially-homologous region leads to a greater effect of the fully-homologous region and vice versa. However, even from the more robust starting point of 5H-6N_1, and similarly to 4H-7N_1, a steep increase in mutation is observed when the region of perfect homology is extended from 75 base-pairs to 100 base-pairs, pointing to a critical threshold at about this length (Fig 3A and 3B). We also note that for all examined combinations of perfect and interspersed homologies, mutations tend to be distributed symmetrically near or around the center of each compound repeat, well inside the partially-homologous region. Taken together, the above findings further imply that the mechanism of homology recognition for RIP integrates sequence information from interspersed and perfect homologies collectively over several hundreds of base pairs.
The above results show that recognition of long interspersed homologies can be promoted by relatively short segments of adjacent perfect homology. To obtain more insight into the interplay between these two homology types, we investigated the effects of substituting interspersed homology for perfect homology in a 100-bp region adjacent to 4H-7N_1 (Fig 2D, construct v). We chose to use 4H-7N_1 because it displayed almost no activity in the absence and strong activity in the presence of the adjacent 100-bp region (Fig 2C). We designed new patterns of interspersed homology in this 100-bp region by fixing the homologous unit length to 6 base-pairs (X = 6) and varying the other two parameters: periodicity (X+Y) and sequence position (Z) (Fig 4A). We have chosen X = 6 because our previous studies [14] suggested that homologous units of this particular length would provide a strong signal which, while not involving continuous homology, might have a significant effect on RIP in combination with 4H-7N_1. Systematic analysis shows that different patterns of adjacent interspersed homology can elicit different effects.
We first varied the periodicity parameter from 8 to 15 while keeping constant sequence position Z = 1. Corresponding interspersed homologies are shown in Fig 4A (from 6H-2N_1 to 6H-9N_1). A periodicity of 10 base-pairs confers the strongest effect, reaching nearly 50% the level observed with perfect homology; periodicities of 8 and 9 base-pairs induce lower but significant levels of RIP; and periodicities of 11–15 base-pairs produce little or no detectable effect (Fig 4B, left). These results were quite unexpected, because our previous analyses implied that periodicities of 11 or 12 should be favored [14], but are explained by further investigation (below).
We next explored the effects of varying the sequence-position parameter Z. In the above analysis, the 100-bp interspersed homology 6H-4N with sequence position Z = 1 (pattern 6H-4N_1) conferred the strongest positive effect on RIP. Thus, we have varied the sequence-position parameter of 6H-4N in this same context from Z = 1 to Z = 12 (Fig 4A). We find that sequence positions of 1 and 2 correspond to equally strong levels of RIP, with a progressive decrease for sequence positions 7–10, followed by an increase again for sequence positions 11 and 12 (Fig 4B, right). These results are intriguing: since the sequence-position parameter Z varies the nucleotide composition of the homologous units without altering basic homology-block structure, the possibility is raised that DNA sequence per se can have an important effect on RIP. In the context of this possibility, particular combinations of base-pairs in the homologous units of 6H-4N_1/11 and 6H-4N_2/12 might be optimal in synergizing with 4H-7N_1. However, alternatively, the differences in the ability to synergize, observed in this particular situation, could potentially arise from the variation in the distance of separation between homologous units at the junction between the 500-bp and 100-bp regions.
To further explore the issues raised in the previous section, we compared mutation of two 500-bp homologies that differed with respect to the sequence-position parameter: 4H-7N_1 (discussed above) and the newly designed interspersed homology 4H-7N_7 (Fig 5A, compare i and ii). We first examined these interspersed homologies alone, without any accompanying 100-bp region. Here we have observed a very dramatic difference with respect to their capacities to promote RIP (Fig 5B and 5C, compare i and ii): while RIP is virtually undetectable in the case of 4H-7N_1 (0.08±0.08 mutations per spore), it is quite strong in the case of 4H-7N_7 (10±1.2 mutations per spore). This comparison provides a clean demonstration that, in a situation where the overall level of homology is weak, differences in DNA sequence at positions of homology can have a substantial effect. In addition, these results show that periodically-spaced homologous units of four base-pairs, corresponding to overall sequence identity of only 36%, are sufficient to drive homology recognition.
We have then further examined the effects of the adjacent 100-bp regions 6H-4N_1 and 6H-4N_7 in combination with the 500-bp regions specifying 4H-7N_1 or 4H-7N_7 (Fig 5A). We find that each 100-bp region confers its characteristic effect, which appears superimposed on the effects conferred by the 500-bp regions: 6H-4N_1 increases RIP in both contexts, whereas 6H-4N_7 has no detectable effect in either context (Fig 5B and 5C). These results show that the relative sequence positions of homologous base-pairs within the 500-bp and 100-bp regions are not relevant—only the absolute sequence positions matter. These findings provide strong additional support for the notion that the sequence composition of homologous units is important for RIP.
If DNA sequence plays a critical role in RIP, what is the basis for this effect? Our previous work has suggested that the triplet of base-pairs represents the basic unit of homology recognition for RIP [14]. Therefore, we wondered whether, among other sequence determinants, “RIP-proficient” homologies might have a higher abundance of certain triplets as compared to “RIP-deficient” homologies. We have shown that the 500-bp homologies 4H-7N_1 and 4H-7N_7, which differ in the positions of the homologous units and, therefore, have the homologous units comprising different base-pair combinations, exhibit a 100-fold difference in their ability to promote RIP. Thus, we have compared nucleotide sequences of 4H-7N_1 and 4H-7N_7 with respect to the natures of triplets included in their homologous units (Fig 6A). We find that most triplets are present in comparable numbers, but with two exceptions: six GAC and eight TGA triplets are present in 4H-7N_7 (which promotes substantial RIP), while neither of these triplets appears in 4H-7N_1 (which promotes very little RIP).
To investigate whether either or both of these differences were significant for RIP, we first explored the role of GAC triplets. We have designed two variants of 4H-7N_7 that lacked either two or all six GAC triplets (Fig 6B, patterns “Δ2GAC” and “Δ6GAC”). The single-nucleotide substitutions that specifically eliminated GAC triplets were chosen to accommodate the fact that each 4-bp unit of homology in 4H-7N_7 contains two overlapping triplets (Fig 1C), making it possible to alter the GAC triplet without affecting the other overlapping triplet. This is accomplished by mutating G to C if the GAC triplet occupies positions 1–3 of the 4-bp unit, or by mutating C to G if the GAC triplet occupies positions 2–4. Our results show that removing the two GAC triplets has reduced RIP by about 25%, while removing all six GAC triplets nearly eliminated all RIP activity (Fig 6C and 6D).
We next analogously examined the possible role of TGA triplets (Fig 6B, pattern “Δ7TGA”). In contrast to the results obtained for GAC triplets, deleting seven TGA triplets using the above approach had no significant effect on RIP (Fig 6C and 6D). Taken together, these results strongly suggest that GAC triplets, but not TGA triplets, may play a privileged role in RIP.
As an additional confirmation of this conclusion, we have examined the role of GAC triplets in another context. Our results have shown that the 100-bp region of perfect homology produces a much stronger increase in mutation compared to the 75-bp region (Fig 3A). We have noticed that the 100-bp region, but not the 75-bp region, contains two GAC triplets (Fig 6E). These GAC triplets are also present in the 100-bp interspersed homology 6H-4N_11 that promotes substantial RIP in combination with 4H-7N_1 (Fig 4B, right). Using the same approach as described above, we mutated the two GAC triplets in 6H-4N_11 (Fig 6E: “ΔGAC”). As a control, we introduced two C/G and G/C substitutions in unrelated triplets (Fig 6E: “Mock”). Mutation analysis shows that the two single-nucleotide changes that deleted the GAC triplets have effectively eliminated the ability of 6H-4N_11 to activate RIP, while the two changes that removed non-GAC triplets have had no significant effect (Fig 6F and 6G). These findings further support the idea that GAC triplets may indeed have a special role in RIP.
Previous research had shown that repeat units were mutated specifically over the extent of shared homology, with only a few mutations occurring outside the duplicated regions [16]. Such high accuracy implies tight coupling between the processes of homology recognition and ensuing mutation. However, our previous work showed that single-copy regions located between pairs of closely-positioned repeats could be strongly mutated, suggesting that homology recognition and mutation can, in fact, be uncoupled regionally [14]. Here we explored the possibility of uncoupling at the base-pair level. Specifically, we have asked whether the local occurrence of mutations is affected by the positions of homologous units. For this purpose, we have compared mutation patterns produced by interspersed homologies 4H-7N_1 and 4H-7N_7 in combination with appropriate 100-bp regions: (i) 4H-7N_1 with 100 base-pairs of perfect homology and (ii) 4H-7N_7 with 100 base-pairs of the interspersed homology 6H-4N_1 (Fig 7A). These repeat constructs share the same reference sequence, but, because of the difference in the sequence-position parameter, their 4-bp homologous units do not overlap (Fig 7B). While the overall RIP profiles appear somewhat different between the two repeat constructs, with the apparent paucity of mutations in the first ~130 base-pairs of 4H-7N_1 (Fig 7A, marked with “*”), similar levels of RIP can be observed within the 200-bp portion of the reference sequence (Fig 7A, outlined). Focusing our analysis on this 200-bp region, we find that cytosines at identical positions are mutated similarly and irrespectively of their spatial relationship to the underlying homologous units (Fig 7B and 7C). This result suggests that, at the level of individual base-pairs, homology recognition and mutation are separable events, either functionally and/or temporally.
We previously examined recognition of interspersed homologies using repeat constructs that included a 200-bp region of tested interspersed homology adjoining a 220-bp region of perfect homology [14], see also (S1 Fig). In this context, the effective patterns of interspersed homology collectively implied that the sequence information was sensed in units of three base-pairs spaced at intervals of 11 or 12 base-pairs, over the total length of several hundred base-pairs. We interpreted this result as evidence that two co-aligned double-stranded DNA molecules were compared by direct contacts. The current study extends these findings in several respects.
Taken together, previous and current findings make it clear that a number of different sequence features contribute to DNA homology recognition for RIP. Our previous study [14] identified triplet homology units and 11/12 base-pair periodicities of those units as important features for RIP. The present study uncovers another important factor: the underlying DNA sequence in general and GAC triplets in particular. It is important to appreciate that different features emerged in the two studies because of the built-in differences in the repeat configurations being used. In the previous work, we examined variations in the basic pattern of interspersed homology within a 200-bp region that was positioned adjacent to a 220-bp region of perfect homology which, by itself, triggered substantial RIP. Thus, the effectiveness of different homology patterns was being evaluated in a situation where the basal level of homology (provided by the 220-bp region) was already above the critical threshold required for RIP. In this experimental system, a preference for homology in triplets of base-pairs spaced with the 11- or 12-bp periodicity was revealed. The present study, in contrast, examined situations in which the basal level of homology was close to or below the critical threshold for RIP. Starting with the intrinsically weak interspersed homology 4H-7N_1 (Fig 2), we found that 100 base-pairs of adjacent perfect homology promoted strong RIP, whereas 75 base-pairs of perfect homology were much less effective, implying that the 100 base-pairs were just barely effective. Starting from this suboptimal situation, examining requirements for interspersed homology in the 100-bp region revealed a role of the underlying DNA sequence. The actual patterns of interspersed homology that permitted the strongest effect did not obviously involve 11/12-bp periodicities. This is likely because the 100-bp region is predicted to contain only ~9 duplex/duplex contact points rather than ~18 as in the previous study, thus giving the DNA sequence a more prominent role. Overall, these observations highlight the fact that homology recognition for RIP can collectively integrate and evaluate diverse underlying features over substantial distances.
Evidence that RIP involves direct dsDNA/dsDNA homology recognition is still indirect. In principle, homology recognition might involve dsRNA interacting with dsDNA or, even, some other type of sequence-specific interactions not involving two double-helical nucleic acids. We note, however, that RIP can recognize 4-bp units of homology embedded in a completely non-homologous sequence, and that four base-pairs are significantly below the threshold length of 6–7 nucleotides required for stable pairing of single-stranded nucleic acids in the context of Argonaute proteins [17], arguing against the role of these proteins in homology sensing for RIP.
The presented observations provide new grounds to justify consideration of existing models of sequence-specific pairing between intact DNA molecules. Two long-standing models invoke the intriguing principle of self-complementarity of Watson-Crick base-pairs, by which identical base-pairs can form planar quadruplexes via major-groove [18] or minor-groove [19] interactions. The possibility of such interactions was confirmed experimentally by NMR [20,21]. Further, in response to our previous work, the principle of base-pair self-complementarity was suggested to be capable of mediating pairing between long intact DNA molecules by the formation of short interspersed dsDNA/dsDNA quadruplexes [22]. This model also predicts that such quadruplex interactions may be sensitive to the underlying DNA sequence [22]. However, pairing of long double-stranded DNAs by this mechanism could involve two duplexes that are either plectonemically intertwined over the "pairing region" (which would permit many contacts along the region) or paranemically related (which would limit direct contacts to one per ~55 base-pairs) [22]. Neither of these conditions is impossible but also neither is necessarily attractive a priori.
Another proposed model includes the intertwining of two negatively supercoiled double-stranded DNA molecules in a form of so-called “PX-DNA”, where homology recognition is mediated by standard Watson-Crick hydrogen bonds [23]. Recent data, however, suggest that the actual structure of the paired complex in this case remains elusive [24]. In a third type of model, non-Watson-Crick hydrogen bonds are proposed to mediate association of specialized sequences, such as G-quartets [25] or triplex structures involving long polypurine/polypyrimidine runs [26]. However, the RIP phenomenon in general, and the newly discovered patterns of sequence recognition in particular, are not compatible with such specialized mechanisms.
Finally, interactions between homologous DNA molecules were also proposed to occur in the absence of direct contacts, by a very different type of a mechanism called "electrostatic zipper" [27]. In this model, two double-stranded DNA molecules with identical sequences can align by making multiple electrostatic contacts between negatively charged backbone phosphates on one duplex and positive charges in the grooves of the other duplex. In this model, DNA homology is read out indirectly: pairing occurs specifically between homologous sequences because only such sequences can produce complimentary patterns of negative and positive charges [27]. As pointed out by the authors, this model cannot account energetically for pairing between DNA molecules that are significantly different as is observed for the interspersed homologies that trigger RIP [28].
Consideration of these models, together with our constraining results, suggests that direct DNA/DNA homology recognition, which also occurs in vitro [29] may involve mechanism(s) that still remain(s) to be described. Furthermore, regardless of the local basis for homology recognition, there also remains the fundamental unsolved issue of how a pair of relatively short chromosomal regions with similar nucleotide sequences can accurately identify one another within the vast space and genomic complexity of the nucleus on a time scale that makes the “genome-by-genome” homology search for RIP feasible.
In Neurospora, RIP evolved as a genome defense mechanism to control the expansion of mobile DNA [15]. In most eukaryotic organisms, including mammals, genomes contain vast amounts of repetitive DNA normally silenced in the form of heterochromatin [30]. While the role of RNA intermediates and sequence-specific DNA binding proteins were clearly implicated in many cases involving epigenetic silencing of DNA repeats [31,32], in the other cases there is no clear understanding how the heterochromatic state can be induced over repetitive sequences [33,34,35]. It is possible that homology-dependent interactions analogous to those that underlie RIP could play a not yet considered role in other phenomena that promote the assembly of heterochromatin on repetitive DNA. We also note that recombination-independent pairing is a prominent feature of inter-chromosomal interactions in both somatic and meiotic systems [2], and thus the rules that underlie homology recognition for RIP may well underlie a wide variety of homology-dependent phenomena in other biological systems.
Methods for creating interspersed homologies, constructing plasmids and strains, setting up crosses, recovering RIP products and analyzing mutations were previously described [14]. Briefly, interspersed homologies were designed in silico by substituting each designated base with one of the three remaining alternatives chosen with equal probabilities and independently from neighboring bases. The exact algorithm for creating interspersed homologies (written in Perl) is provided in S4 File. Synthetic DNA was ordered as “gBlocks” from Integrated DNA Technologies. Repeat cassettes were integrated as a replacement of the csr-1 gene in the mus-52Δ strain FGSC#9720, which is deficient in the non-homologous end joining pathway and can only be transformed by homologous recombination [36]. 1–2 homokaryotic transformants were typically selected for further analysis. All integration events were validated by sequencing. Plasmids and strains created in this study are listed in S1 Table. Individual plasmid maps (in GenBank format) are provided in S1 File (as tar/gz archive).
The standard wildtype strain FGSC#4200 was used as a female parent for all the crosses. 1–3 replica crosses were analyzed for each repeat construct; at least 30 random “late” spores were randomly sampled from each cross. The number of replica crosses and the total number of analyzed spores are provided in S1 Table. PCR-amplified repeat cassettes were sequenced directly by the Sanger method. Sequencing reactions were read with an ABI3730xl DNA analyzer at the DNA Resource Core of Dana-Farber/Harvard Cancer Center (funded in part by NCI Cancer Center support grant 2P30CA006516-48). Individual chromatograms were assembled into contigs with Phred/Phrap. Assembled contigs were inspected manually in Consed. Sequences of all contigs analyzed in this study are provided in S3 File.
|
10.1371/journal.pntd.0005378 | Improving our forecasts for trachoma elimination: What else do we need to know? | The World Health Organization (WHO) has targeted trachoma for elimination as a public health concern by 2020. Mathematical modelling is used for a range of infectious diseases to assess the impact of different intervention strategies on the prevalence of infection or disease. Here we evaluate the performance of four different mechanistic mathematical models that could all realistically represent trachoma transmission. We fit the four different mechanistic models of trachoma transmission to cross-sectional age-specific Polymerase Chain Reaction (PCR) and Trachomatous inflammation, follicular (TF) prevalence data. We estimate 4 or 3 parameters within each model, including the duration of an individual’s infection and disease episode using Markov Chain Monte Carlo. We assess the performance of each models fit to the data by calculating the deviance information criterion. We then model the implementation of different interventions for each model structure to assess the feasibility of elimination of trachoma with different model structures. A model structure which allowed some re-infection in the disease state (Model 2) was statistically the most well performing model. All models struggled to fit to the very high prevalence of active disease in the youngest age group. Our simulations suggested that for Model 3, with annual antibiotic treatment and transmission reduction, the chance of reducing active disease prevalence to < 5% within 5 years was very low, while Model 2 and 4 could ensure that active disease prevalence was reduced within 5 years. Model 2 here fitted to the data best of the models evaluated. The appropriate level of susceptibility to re-infection was, however, challenging to identify given the amount and kind of data available. We demonstrate that the model structure assumed can lead to different end points following the implementation of the same interventions. Our findings are likely to extend beyond trachoma and should be considered when modelling other neglected tropical diseases.
| Trachoma is the world’s leading infectious cause of blindness. Mathematical models are used by researchers to examine the spread of infectious diseases and understand how they can be controlled. Such models are developed based on the natural history of infection. For trachoma we identify four different model structures which could all represent the natural history of trachoma infection. We fit each of the models to infection and disease prevalence data for 3 different age groups. We find that one of the models is able to fit the data better than others, however some factors about the model are difficult to identify due to limited data. The ease of eliminating disease within a community assuming the same interventions varied depending on the model structure assumed. Our results highlight that some models of trachoma fit to infection and disease data better than others, but that more data is needed to identify more specific aspects of the model structure. In addition we show that different model structures may give different results in terms of the effort required to control trachoma transmission.
| Trachoma remains the world’s leading infectious cause of blindness. 200 million people are reported to be at risk of infection, across 42 endemic countries [1]. The causative agent of infection is the bacterial pathogen Chlamydia trachomatis [2]. The World Health Organization through the Alliance for the Global Elimination of Trachoma by 2020 (GET2020) is aiming to eliminate trachoma as a public health problem by 2020. Two goals have been developed to assist endemic countries striving to achieve the elimination of trachoma as a public health problem. The first goal aims to reduce the prevalence of Trachomatous inflammation, follicular (TF) in children aged 1–9 years, to less than 5% by 2020. Mathematical modelling has been successful in helping to formulate guidelines for the ongoing surveillance and control of a range of infectious diseases including malaria [3], onchocerciasis [4], lymphatic filariasis [5] and soil-transmitted helminths [6]. Furthermore, mathematical models can be used to provide guidance on suggested timelines to elimination or control, for a given set of initial conditions and available interventions. However, to generate informative and accurate predictions, models need to be informed by high quality epidemiological data, particularly in terms of the duration of infection and disease, as it is these states which are detectable through diagnostic tests.
For trachoma, the control guidelines are based on the disease which occurs as a consequence of infection with C. trachomatis bacteria. Therefore, guidelines are based not on monitoring the causative agent of infection directly, but the longer-term disease associated with it [7]. It is understood that individuals can remain TF positive with detectable disease for far longer than they are Polymerase Chain Reaction (PCR) positive. Despite this, estimates of the duration of PCR detectable infection and the duration of disease are not commonly available and are rarely estimated [8]. Nonetheless, a good understanding of the time spent in these states is vital if accurate model projections on time to elimination of trachoma are to be developed. For example, in the absence of on-going sustained transmission in an endemic region, if one assumed that the duration of a disease episode was less than 1 month (as estimated for individuals 15 years or older in the community [8]) the expected time to reach the < 5% elimination threshold would be much shorter than if the assumed duration of disease was 2 years; thus, the assumed rate of recovery from disease is likely to have a large impact on the expected time to reach elimination targets.
Mathematical models are developed and informed by the natural history of infection. For trachoma, the Susceptible, Infected, Susceptible (SIS) model structure has most commonly been used [9–15] where individuals in the S state are susceptible to infection and those in the I state are infected and infectious and, thus, are PCR detectable. Such models are fitted to PCR data collected during clinical trials of trachoma treatment [9, 10, 12]. However, current control guidelines are based on disease, not PCR detectable infection. Therefore, if models are to be informative in terms of whether the guidelines on TF prevalence will be achieved, it may be desirable to capture the dynamics of both PCR and TF positivity, although this has rarely been done [16, 17]. The lack of modelling work in this field is likely to have been exacerbated by the limited longitudinal population level data that measure both PCR and TF positivity across multiple age groups, particularly following the implementation of interventions.
In addition to the SIS compartmental model structure, several other variant structures may also be considered appropriate given the natural history of trachoma infection [18–21]. It has been reported for other infectious diseases that the structure of the model assumed can impact the estimated effort required to control and eliminate that disease [22–24]. Therefore, when modelling the transmission of trachoma to make projections on the feasibility of elimination, it is important to select not only the most parsimonious and statistically appropriate model, but also to understand how the assumed model structure may impact elimination projections.
Here we compare four different model structures which could realistically all represent the natural history of trachoma infection as understood by epidemiologists through the interpretation of experimental data. We statistically fit each model to cross-sectional data on bacterial load, PCR and TF prevalence for three different age groups. We then evaluate the performance of each model structure using the Deviance Information Criterion (DIC). With the parameter estimates obtained from each of the best performing models we assess the feasibility of and time to elimination, to understand if and how they differ.
We used cross-sectional data on PCR and TF prevalence in individuals aged 1–4, 5–14 and 15 years or older, collected from a hyperendemic community in Tanzania at one point in time, prior to the roll out of trachoma interventions [25]. Data on the mean bacterial load by age group were available from [25]. Data on age-specific bacterial load, PCR and TF prevalence were used to fit each of the models evaluated.
We evaluate 4 different plausible natural histories of infection [9, 11, 13, 26] and disease [7, 16, 18, 19] that may occur following exposure to trachoma. The model structures we evaluate highlight the clinical and epidemiological observations made in the field and laboratory [19, 21, 27, 28]. The first model, Model 1, follows the structure represented in Fig 1a. Here susceptible individuals (S) become infected at a rate λ, they incubate infection in the (I) state, and progress at a rate σ to the infected infectious state(ID) where they test PCR positive and TF positive. Individuals leave ID at a rate ω and progress to the disease only state (D), where they are only TF positive, and recover from the disease only state at a rate ρ and return to the susceptible state (S).
For Model 2 we assume the same structure as Model 1 (Fig 1b), however we do not assume individuals in the D state are 100% immune to re-infection [19]. Instead we explore 3 levels of susceptibility to re-infection (Γ): 20%, 50% and 80%. All other transitions are the same as described in Model 1.
In Model 3 (Fig 1c) we evaluate the structure previously postulated by Shattock et al [18]. Here, the first 3 states are identical to those described in Model 1 and 2. However, for Model 3 we split the duration of time spent in the D state across two compartments. In the D state, individuals are immune to re-infection. They then progress to the PD state at a rate γ [19]. In this state individuals are susceptible to re-infection with the same susceptibility levels described in Model 2.
Model 4 (Fig 1d), we introduce an additional infected state, the IO state, which comes after the incubating state, where individuals are not infectious (I), but prior to the ID state. In the IO state individuals have a PCR detectable infection, but are not yet TF positive. From here individuals progress to IA at a rate η where they are PCR and TF positive, individuals recover from their infection and progress to the D state, where they are only TF positive, but as with Models 2 and 3 individuals could experience re-infection in the (D) state with the same susceptibility levels described in Models 2 and 3 [19].
All models follow the ‘ladder of infection’ structure [11, 16, 26], whereby each subsequent infection leads to improved immunity following re-infection. In all 4 model structures we reflect improving immunity as an increase in the rate of recovery from infection and disease episodes, in addition to a reduction in infectivity with each successive infection. We assume that the infectivity of an individual is proportional to their bacterial load. In the model we reflect declines in bacterial load with repeated infection as reductions in an individual’s infectivity to others. This represents a trend in agreement with the data from trachoma endemic communities in which the bacterial load decreases with age [25, 29, 30]. For each model structure (A-D, Fig 1) we have two sub-variants, the 4-parameter and the 3-parameter versions. These models pertain to two alternative sets of assumptions about how bacterial load and infectivity decline with consecutive infections. We assume for the 4 parameter model that infectivity is proportional to bacterial load and therefore declines exponentially with the number of prior infections. For the 3 parameter model we assume that infectivity declines linearly with the log of the bacterial load i.e. linearly with the number of prior infections. We chose exponential functions as fairly flexible low-parameter functions that, for the rates of recovery, would accomplish the goal of a) rising from an initial value–for no and low numbers of infections–to b) saturating at a high value for high numbers of infections. We note that we also tested the use of a log-logistic function instead of an exponential, however it was no better performing than the exponential function.
Additional detail on the model parameters and state variables are presented in Table 1. Details on the immunity functions and mathematical equations for each model are presented in S1 File.
All parameter values and definitions are provided in Table 1. We assume that the mean minimum duration of an infection episode was 10 weeks and the duration of a disease only episode was 1 week (Table 1), the same as those estimated for the oldest age group in Grassly et al [8]. We take estimates from the oldest age group to parameterise the minimum duration of an infection and disease episode. Immunity to trachoma is thought to develop through repeated infections. Therefore as those in the highest age group are most likely to have experienced the highest number of infections, we assumed that they would have the highest levels of immunity. It is inherently challenging to estimate immunity functions [32] and, given only 3 data points were available, the true values of any immunity parameters were likely to be unidentifiable. As such, exponential increases in the rate of recovery from infection and disease, with the number of prior infections experienced by an individual, were informed by Grassly et al [8]. Age-specific estimates of the duration of infection and disease, and were fixed for the purposes of model fitting.
Each of the four model structures evaluated were fitted as 3 and 4 parameter models to the data. An additional factor: the relative susceptibility of the diseased, non-infected state for new infections was varied. One value was used for Model 1, and 3 different values for each model structure 2, 3 and 4. Therefore a total of 10 different models were fitted for each parameterisation of the model structures. For the 4 parameter models we estimated the transmission rate parameter β per day−1 for the data, the duration of an individual’s first infection and disease episode in the ID and D states, and the rate at which infectivity changed with each successive infection for the bacterial load function (Table 1). For the 3 parameter model we estimated the first three parameters listed in the 4 parameter model, but assumed a constant linear decline in the log load of an individual’s bacterial load with each successive infection, thus, in the 3 parameter model we did not estimate the rate of increase in improved immunity with re-infection.
Parameter estimation was performed using Markov Chain Monte Carlo (MCMC). The chains were run for 10,000 iterations. The Robbins-Munro algorithm was implemented as part of the adaptive stage of the MCMC-Metropolis Hastings algorithm, to ensure the proposal distributions were adaptively tuned ensuring efficient exploration of the posterior [33]. Selection of the most parsimonious model and fit of each model to the data was assessed using the DIC [34], therefore we assumed our posterior distribution was approximately multivariate normally distributed. Fits to the data for each model are presented in Table S2 in S1 File. Estimates from the 4 parameter model are provided in Table S3 in S1 File and estimates from the 3 parameter model are provided in Table S4 in S1 File. Uninformative uniform priors were specified for all parameters. MCMC diagnostics are presented in Table S5 in S1 File. We calculate the Gelman-Rubin statistic for 2 MCMC chains to assess convergence [35] and the Effective Sample Size (ESS) for each model fit.
For each model structure (described above) we performed simulations to assess the potential impact of Facial cleanliness and Environmental improvements (F and E) within the community, along with the implementation of mass drug administration (MDA). All simulations were started from endemic equilibrium. For communities with greater than 20% TF, 5 annual rounds of MDA were performed, and for those with TF 20% or less we performed 3 rounds of annual MDA. We also assessed the possible impact of F and E to reduce transmission. The true impact of F and E remains poorly quantified [36], therefore we consider a range of reductions in transmission that may be possible (between 0–50%). β was assumed to decline exponentially over the intervention period, to model an increasing uptake of transmission reduction interventions in the community over time. We model changes in β as an instantaneous drop when each annual round of MDA is performed as we assumed intensified health promotion activity would be conducted when MDA was distributed. Reductions in transmission which were only considered to occur through the implementation of F&E and were assumed to be maintained following the cessation of treatment.
For each model structure we assessed the time taken and the feasibility of reducing TF prevalence to less than 5% within the community in children under 10 years old. A constant level of treatment coverage (80%) between each round and across model comparisons was assumed, along with a fixed treatment efficacy of 85% [12, 31]. A schematic of the movement of individuals between compartments following treatment is illustrated in Fig 1. We assess the sensitivity of findings on the feasibility of elimination for the different model structures to variation in 6 different fixed parameters, these are: treatment efficacy, duration of first infection and disease episodes, maximum rate of recovery from infection, maximum rate of recovery from disease and the degree of age mixing in the population. These were assessed across both transmission settings and all levels of transmission reduction due to F and E.
None of the model structures evaluated captured the very high reported prevalence of TF in children aged 1–4 years (Fig 2) but they were able to capture TF prevalence in the older two age groups well (Fig 2). Equally, all models fit the age-specific PCR data well. Across all structures explored it was not possible to capture the high average bacterial loads reported in young children, assuming the exponential function for the development of immunity with bacteria load, but this was possible when assuming a linear decline (Table S4 and Figure S1 in S1 File). Although, assuming the linear model typically resulted in higher estimates in the bacterial load in older age groups, not seen in the data (Table S4 and Figure S1 in S1 File).
In general across the 3 and 4 parameter models, predicted age-specific prevalence of infection and disease were lower for Model 3 in comparison to Model 4 (Fig 2). Typically with Model 4 the estimated PCR prevalence was too high for all age groups in comparison to the data. The higher level of PCR prevalence obtained by Model 4 reflects the overall longer duration of PCR detectable infection for Model 4 in comparison to Model 3. Thus, we would expect that these models’ equilibrium age-specific prevalence levels would differ from one another. Model 2 and 3 had similar predicted values of age-specific TF prevalence but predicted PCR prevalence was lower for Model 3 in comparison to Model 2 (Fig 2). This may be because the duration of the disease only episode in Model 3 was longer than Model 2, therefore it takes longer for individuals to get re-infected and test PCR positive under Model 3 in comparison to Model 2. The estimated age-specific prevalence for infection and disease was lower for Model 1 in comparison to Model 2, resulting in the fits from Model 2 more closely aligning with the data than the fits from Model 1. This difference may reflect the importance of allowing some susceptibility to re-infection on individuals in the TF only state.
The mean estimated duration of infection and disease periods for an individual’s first infection were roughly comparable across all models when 3 or 4 parameters were estimated (Table S2 and Table S3 in S1 File). However, the estimated value of β (day−1) across the different structures evaluated varied substantially, this was true for both the 3 and 4 parameter models. Estimates from Model 3 provided the largest estimated values of β. This is likely to be because this model structure includes two states that do not contribute to the overall force of infection, thus in order to fit to the high level of infection and disease a very high value of β was needed relative to the other models evaluated. Within a given model structure estimates of β were less affected by the assumed level of susceptibility to re-infection in the TF only state for Model 4. This may be due to the longer duration of infection reflected under this structure in comparison to the others. This longer duration of infection also meant that the estimated value of β for Model 4 was the lowest of all structures evaluated.
The 4 parameter version of Model 2 with varying levels of susceptibility to re-infection provided the most parsimonious fit to the data, according to the DIC (Fig 2, Table S2 in S1 File). For Model 4, assuming high levels of susceptibility to re-infection in the disease state was comparable in performance to Model 3. DIC scores for Models 1, 2, 3 and 4 were: 1942.44, 1928.04, 1949.79 and 1950.60 for each model respectively. For the three parameter model (Table S3 in S1 File) assuming a linear decline in bacterial load, Model 2 also had the lowest DIC score out of the four structures assessed. These findings suggest that while there is statistical evidence to suggest that re-infection in the disease only state is important, with the current data available, it is not possible to identify which level of susceptibility is most appropriate.
Across both transmission settings infection was more likely to re-emerge if the infectivity was assumed to decline exponentially (Figs 3, 4a–4c). However, in general, this functional form led to an overall better fit to the cross-sectional data (Table 1). This somewhat counterintuitive effect with exponentially declining infectivity (explored in [37]) results in the reproduction number associated with the full model increasing with each subsequent treatment round of MDA. This is as a result of the concentration of infectivity increasing with multiple rounds of MDA, as a higher number of individuals in the population have experienced fewer infections, resulting in individuals aggregating at higher infectivities as their progress along the ‘ladder of infection’ is slowed or halted due to MDA. When assuming an exponential decline in infectivity it was not possible to eliminate disease from the community with 40% TF prevalence with Models 2 or 3, even with annual MDA for 5 years, this was only possible when some reduction in the transmission rate was also included (Fig 3a–3c). It was only possible to eliminate disease when a linear decline in bacterial load was assumed under Models 2 and 4 with at least a 10% reduction in transmission and annual MDA for 5 years (Fig 3d and 3f); in all other structures and transmission reduction scenarios, infection re-bounded. It was not possible to reach the elimination threshold guideline at all with Model 3 (Fig 3b and 3e).
When TF baseline prevalence was 40%, under Models 2 and 4, it was possible to eliminate infection within 5 years with annual MDA and an overall transmission reduction when assuming a linear decline in bacterial load (Fig 3e and 3f), provided transmission reduction was greater than 10%. However, if an exponential decline in bacterial load was assumed (Fig 3a–3c), it was only possible to eliminate infection in Model 2 with 5 annual rounds of MDA and 50% reduction in transmission, but this was not sufficient for Model 4. Under no intervention conditions evaluated here was it possible to eliminate infection with Model 3 (Fig 3b and 3e).
Assuming TF prevalence was 20% we implemented 3 annual rounds of MDA and transmission reduction (between 0–50%) (Fig 4). Considering a linear decline in bacterial load it was possible with at least 10% transmission reduction to reduce disease prevalence below the target threshold in Model 2 and 4. However, if there was no transmission reduction disease appeared to re-emerge (Fig 4d and 4f). As with the previous prevalence levels it was not possible to achieve the target level of disease prevalence with Model 3 (Fig 4e). When assuming an exponential decline in bacterial load, with all four model structures, it was possible to reduce the prevalence of disease to the target level of less than 5%. However, without subsequent rounds of MDA, this was not maintained in Model 3 and disease re-emerged (Fig 4b). However, for Models 2 and 4, 10% reduction in transmission respectively was sufficient to ensure that disease did not re-emerge in the community (Fig 4a and 4c) and suppression below the target level was maintained.
Observing the rates of re-bound under different model structures we found that for Model 2 with modest levels of transmission reduction rapid re- emergence was observed for the 4 parameter model. This is in-part likely to be because for a given prevalence level we need a higher force of infection if immunity was high, resulting in faster rates of rebound until β was reduced sufficiently. For Model 3 under both parameter versions and all scenarios rapid rebound of disease was seen. This is likely to be because this structure includes 2 compartments which do not contribute to the overall force of infection. Therefore in order to obtain a fixed level of prevalence the value of β must be increased substantially in comparison to other model structures, thus making persistence of disease more likely under this structure. Limited evidence of re-bound was seen when evaluating Model 4 in comparison to other models this model had an overall longer duration of PCR detectable infection, therefore a lower value of β was needed to attain any given level of prevalence. This meant that when an intervention was applied, the lower overall force of infection resulted in a slower rate of rebound. The on average higher infectivity of individuals in the 3 parameter model in comparison to the 4 parameter model (Table S4 in S1 File) is likely to explain the slightly higher re-bound rates in the 3 parameter version of Model 4, in comparison to the 4 parameter version.
We conducted one-way univariate sensitivity analyses with 6 of the fixed parameters in the model to assess their impact on the models assessment of the feasibility of elimination (Table S6 and Table S7 in S1 File). For the 4 parameter version of Model 1 and Model 2 when TF was 40% or 20%, variation in treatment efficacy and the minimum duration of infection had the largest impact on final TF prevalence. Here, higher treatment efficacy resulted in faster infection rebound, leading to a higher final TF prevalence. For Model 2 final TF prevalence was 70% compared to 61% when treatment efficacy was increased from 85% to 100% (Table S6 in S1 File). In contrast, a 50% reduction in the minimum duration of infection resulted in fast infection re-bound resulting in a high final TF prevalence above the baseline results, until a 50% reduction in transmission was implemented (Table S6 in S1 File). Decreasing the minimum duration of disease resulted in a final higher TF prevalence when transmission reduction < 50% was implemented. For the 3 parameter version of Model 1 and Model 2 final TF prevalence decreased with increasing treatment efficacy. (Table S6 in S1 File). While reduction in the minimum duration of infection and disease episodes resulted in higher final TF prevalence levels when little or no transmission reduction was implemented. Final TF prevalence for Model 2 when transmission was reduced by 10%, was 60% when the minimum duration of infection was 5 weeks, but 8% for the 10 week baseline value (Table S6 in S1 File).
Across the 3 and 4 parameter versions of Model 3, little to no variation in the final TF prevalence level was seen when sensitivity to the fixed parameters was conducted (Table S6 and Table S7 in S1 File), and the inability to even come close to reducing or eliminating disease was seen across all parameter sets tested. Insensitivity of Model 3 to perturbations in the six different parameter sets, is likely to be a consequence of the high force of infection needed to obtain a fixed level of prevalence with this model structure (as described in the model fitting results), which increases the persistence of infection and disease. Equally, individuals spend a long time in the TF +ve only state under this structure, therefore changes in infection rate parameters are less likely to have a profound impact on this model.
For Model 4, lower treatment efficacy resulted in a higher final TF prevalence than when the baseline value was used. Prevalence was 11% instead of 2% when treatment efficacy was 65% in the 4 parameter model when no transmission reduction was modelled (Table S6 in S1 File). As with Models 1 and 2, reductions in the maximum duration of infection and disease episodes at low levels of transmission reduction resulted in higher final TF prevalence for the 4 parameter model. If the maximum rate of recovery from infection was changed to 0.008 from 0.006 final TF prevalence was 40%, in comparison to 2% (Table S6 in S1 File). However, when endemic prevalence of TF was 20% for the 4 parameter model, results were consistent across all variation in the parameter sets. For the 3 parameter versions of Model 4 results were consistent across all parameters sets tested when endemic prevalence of TF was 40%. However, when endemic TF prevalence was 20% for the 3 parameter model, an increase in final TF prevalence from the baseline was observed when the minimum duration of infection and disease episodes was reduction, particularly when no transmission reduction was implemented. Final TF prevalence was 14% instead of 4% when the minimum duration of infection was decreased from 10 weeks to 5 weeks (Table S7 in S1 File).
In this study we present the first attempt to fit a mechanistic epidemiological model of trachoma transmission to bacterial load data, PCR and TF prevalence data across 3 different age groups. We demonstrate that it is possible to fit to the age-structured PCR data well but the very high level of TF in the youngest age group analysed in this hyper-endemic setting proved challenging to capture with all model structures tested. In addition, we highlight that predictions about the future prevalence of TF within a community can depend on the model structure assumed.
While a range of different model structures can describe the natural history of trachoma infection well, Model 2, with re-infection in the D state (TF positive, PCR negative), was statistically the best performing model under all conditions. Model 2 represented the most parsimonious model structure when assessed by the DIC score obtained through fitting to the dataset used here. The appropriate level of susceptibility to re-infection was, however, challenging to identify given the amount and kind of data available. We can only therefore confidently say that our model selection study suggests that individuals with active disease but no current infection remain susceptible to infection, but it does not suggest what their susceptibility to infection is, relative to those with neither active disease nor infection.
We demonstrate that overall a better fit to the data, i.e. ensuring infection and disease age-specific prevalence were captured, was provided by an exponential reduction in bacterial load in comparison to a linear decline in load. However, the use of an exponential rather than linear bacterial load decline was also shown to suggest that more effort may be required to reduce the prevalence of TF in the long term, due to the persistently high levels of load associated with those who have experienced very few prior infections i.e. those likely to be the few remaining infected individuals when elimination is close.
Estimates of the effort required and feasibility of elimination were markedly different under different model structures. Model 3 showed that the prospect of reaching TF less than 5% was very low, while with Model 2 annual MDA and transmission reduction together, in prevalence settings below 40% TF, ensured that the goal was reached within 5 years. To gain further understanding into the long-term transmission dynamics of trachoma and generate accurate elimination timelines, further insight into the duration of infection and disease episodes is required, ideally through at least one further study designed to measure these durations. Furthermore, our results highlight the importance of identifying and understanding the most appropriate and parsimonious structure to model trachoma transmission, this is essential if we wish to use mathematical models to help understand the transmission dynamics of trachoma and to model current and alternative intervention strategies.
Our sensitivity analysis highlights that projections on the feasibility of elimination under different model structures were sensitive to a number of key parameters, particularly for Models 2 and 4. Final TF prevalence after 7 years was most impacted by the assumed duration of an individual’s first infection and disease episodes, in addition to the efficacy of treatment. Suggesting that a more thorough understanding of these parameters would be valuable for future model forecasting. A small amount of variation in the final TF prevalence was observed when the maximum rates of recovery from infection and disease were perturbed, however the impact was not as profound as the outcomes from the aforementioned parameters. In general, we observed that as the modelled reduction in the transmission rate increased sensitivity of the model prediction of the final TF prevalence level decreased. However, the final TF prevalence outcomes from Model 3 appeared insensitive to perturbations across all parameter evaluated.
We were consistently unable to capture the high prevalence of TF in the youngest age group, but were able to capture PCR prevalence for this age group. This suggests that the models evaluated may be missing or mis-specifying a key component of the epidemiological system. For example, it could be that the functional form used to describe the development of immunity has been mis-specified here (as an exponential function) or that age-specific prevalence ratios of PCR vs TF vary according to the transmission setting. However, particularly for Model 2, prevalence of disease and infection were matched well for the two older age groups. However, it has been suggested that at the population level the relationship between TF and PCR positivity is approximately linear [38], which can also be seen in our model projections. Since, we have only fitted to cross-sectional data from a single time-point from a single region and time point, we cannot disregard the possibility that there may be an anomaly in the data, and extrapolating our findings to a wider context should only be done with caution. Equally, the prevalence of infection within the adult age group may be considered high. However, in the absence of cross-sectional data collected across a wide range of age groups from different study sites, it is difficult to assess whether or not this observed infection prevalence is abnormally high or not.
The models evaluated here have only been fitted to one high prevalence site, however trachoma transmission can be highly heterogeneous between neighbouring communities. Therefore it is possible that if we had used data from a different community we may have achieved different results. However our hope from fitting to data from this study site is that while certainly the baseline level of transmission within other neighbourhoods may not be as high, we would hope that some of our findings would be generalisable to other settings. Nevertheless the use of data from a single high prevalence study site is a clear limitation of the study.
Statistical models have been shown to forecast prevalence of infection and disease well and can predict changes in prevalence over time [7, 9, 12], however there is less flexibility within a statistical framework to explore the impact of novel or future alternative intervention strategies. Therefore, selecting an appropriate mechanistic model structure is important if we wish to more accurately model trachoma transmission and assess the possible impact of different intervention strategies in the lead up to 2020. Furthermore, we demonstrate that our understanding on the feasibility of trachoma elimination varies under different model structures.
In this study certainty about the appropriate model structure and susceptibility level to re-infection was hampered by a limited amount of data relating to the durations spent in different infection and disease states, in addition to longitudinal post-intervention follow-up data on infection and disease from a range of different communities and transmission settings. For example, if we knew the average duration an individual spends as PCR-positive but TF-negative we could parameterise our models with more certainty, a point that is even more important for the duration in which individuals remain only TF positive. However, in all of our models, PCR and TF positivity are inherently linked. We suggest that further validation of appropriate model structures can be provided through fitting different structures and different model types to longitudinal data from a range of different transmission settings, coupled with more large scale model and data comparisons, as we seek to develop models which help provide guidelines on time to elimination. Our findings may also be applicable to other NTDs where certain key parameters are not well known, where limited data exists and limited investigation has been done to validate the model structure being used to model transmission.
|
10.1371/journal.pntd.0003324 | Patients' Perceptions on the Performance of a Local Health System to Eliminate Leprosy, Paraná State, Brazil | In Brazil, leprosy has been listed among the health priorities since 2006, in a plan known as the “Pact for life” (Pacto pela Vida). It is the sole country on the American continent that has not reached the global goal of disease elimination. Local health systems face many challenges to achieve this global goal. The study aimed to investigate how patients perceive the local health system's performance to eliminate leprosy and whether these perceptions differ in terms of the patients' income.
A cross-sectional study was conducted in Londrina, State of Paraná, Brazil. Interviews were performed with the leprosy patients. The local health system was assessed through a structured and adapted tool, considering the domains judged as good quality of health care. The authors used univariate, bivariate and multivariate analyses. One hundred and nineteen patients were recruited for the study, 50.4% (60) of them were male, 54.0% (64) were between 42 and 65 years old and 66.3% (79) had finished elementary school. The results showed that patients used the Primary Health Care service near their place of residence but did not receive the leprosy diagnosis there. Important advances of this health system were verified for the elimination of leprosy, verifying protocols for good care delivery to the leprosy patients, but these services did not develop collective health actions and did not engage the patients' family members and community.
The patients' difficulty was observed to have access to the diagnosis and treatment at health services near their homes. Leprosy care is provided at the specialized level, where the patients strongly bond with the teams. The care process is individual, with limited perspectives of integration among the health services for the purpose of case management and social mobilization of the community to the leprosy problem.
| Brazil still has not achieved the goal of leprosy elimination established by World Health Organization. The diagnosis and treatment of leprosy is easy and the country is striving to fully integrate leprosy services into existing general health services. Access to information, diagnosis and treatment with multidrug therapy (MDT) remain key elements in the strategy to eliminate the disease as a public health problem, defined as reaching a prevalence of less than one leprosy case per 10,000 population. Thus, this study aimed to investigate the performance of a local health system to eliminate leprosy through a cross-sectional study with leprosy patients. One hundred and nineteen patients were recruited for the study, 50.4% of them (60) were male, 54.0% (64) were between 42 and 65 years old and 66.3% (79) had finished elementary school. The health teams have invested only in the individual perspective and they do not coordinate the patients' care and, also, their practices are fragmented without perspectives of integration between different health services and social mobilization, an essential condition for the development of an effective and responsive system.
| Iniquity in access to health services has been discussed among health authorities in developing countries, especially regarding poverty and neglected diseases, such as leprosy [1]. In recent years, in Brazil, the prevalence of this disease has progressively declined; nevertheless, this is the only country on the American continent that has not reached the global goal of disease elimination, with a detection rate of approximately 17.2 per 100,000 inhabitants [2].
Therefore, leprosy has been listed among the health priorities since 2006, in a plan known as the “Pact for life” (Pacto pela Vida) [3]. Local health systems face many challenges to achieve the global goal of leprosy elimination though. Most leprosy cases have been diagnosed late and this entails serious consequences for the individuals, due to the possibility of physical disabilities, negatively affecting self-care, work capacity and social relationships. [4], [5], [6], [7], [8].
It is also highlighted that only 1/3 of the patients are reported in the country, so that many continue transmitting the disease in the communities without being reached by local health systems [9]. Among the patients who get treated, it is estimated that a large majority do so irregularly or drop out, leading to drug-resistant bacilli, which aggravates the context of the disease in Brazil [10].
Although the country registers a drop in prevalence rates in recent years, it is far from achieving the elimination target of leprosy (<1 case per 10,000 population), unless the local health systems perform their functions as good as possible, changing their organizational and management logic [11].
The importance of a system focused on chronic conditions has been described in the literature, which strengthens prevention and health promotion actions [12], important measures to break the transmission chain of leprosy. Nevertheless, few studies intend to investigate the performance of local health systems from this more comprehensive perspective.
Despite conceptual disagreement among the authors, the term performance has been employed to express the extent to which the local health systems reach their objectives [13]. These objectives are in line with the values representations and needs of each population, so that each system will have a peculiar form of defining and producing its health actions [13]
Therefore, it is important to analyze how these local health systems have operated, within what logic and the extent of their evolution towards the elimination of leprosy. Studies on local health system performance can play a strategic role in the achievement of the elimination target, in terms of rethinking processes and triggering changes. Thus, the aim in this study was to investigate how patients perceive the local health system's performance to eliminate leprosy and whether these perceptions differ in terms of these patients' income.
Ethical approval for the study was obtained from the University of São Paulo at Ribeirão Preto College of Nursing (Protocol number 08811212.0.0000.5393). All study participants signed a written consent form.
A cross-sectional study [14] was carried out in Londrina, State of Paraná, Brazil. Londrina is located in the State of Paraná, Brazil, a hyperendemic area for leprosy. In 2012, its detection rate corresponded to 8.72 per 100,000 inhabitants and nearly one case per 100,000 inhabitants was diagnosed with grade 2 disability, which refers to the presence of a visible deformity or apparent physical damage and when sight is severely compromised [15]
The city is located in the South of Brazil, at a distance of 1,103.1 kilometers from Brasilia. The population amounted to 515,707 inhabitants, with a life expectancy of 71.37 years old in 2008 and a human development index of 0.824 [16]. The Family Health Program, an important program launched in Brazil to strengthen Primary Health Care (PHC), covers nearly 58% of the population [17]. Since 2009, the local health authorities have initiated a decentralization of leprosy control actions to the PHC.
The study population consisted of 165 leprosy patients, identified through the Notifiable Diseases Information System (SINAN). For the study, the authors considered patients diagnosed from 01 January 2009 to 31 December 2012. The inclusion criteria were: patient with an address in the urban area of Londrina and who were 18 years old or older. The authors considered as exclusion criteria: patients who lived in rural areas, corresponding to the regions of the city beyond the urban perimeter [18] or who were not found at home after three visits by the researchers.
Figure 1 shows the numbers of individuals in each study stage and the eligible and analyzed participants.
The data was obtained through the application of a questionnaire adapted from the PHC Assessment Tool (PCATool) [19] and validated by Villa and Ruffino-Neto (2009) [20] to study the performance of health systems in the control of chronic transmissible and neglected diseases, such as tuberculosis, HIV and leprosy. The authors also collected sociodemographic data and patient characteristics and the type of health services sought at the onset of signs and symptoms. Participants answered a tool using a five-item Likert response scale: never true (1), somewhat true (2), true half of the times (3), mostly true (4), always true (5).
The tool was structured in two parts. The first investigated the participants' sociodemographic characteristics and the second was structured in nine attributes, being First contact, Access to the diagnosis, Access to treatment, Comprehensiveness of services, Longitudinality-relational, Coordination and Collaborative health actions, Family centeredness, Community orientation and Interpersonal communication. These domains have been proposed as assessment criteria to judge the quality of health care [13], [14]. The key definitions of the domains were based on the study developed by Haggerty et al. [21]. Table 1 presents the key definitions of each domain investigated and the number of items each domain contains.
Data collection was conducted between June and September 2013. First, authors contacted the Epidemiological Health Surveillance of the Londrina City Health Secretariat to identify patients diagnosed with leprosy and their addresses during the study period.
The authors visited all health care services (HCS) where the patients were being followed to introduce the study to the coordinators and health professionals. On that occasion, all addresses and phones of patients were updated. Subsequently, the patients were invited by telephone and, when they did not possess a phone number or had a telephone but the authors were not able to contact them, the researchers visited their home to invite them to the study. If they consented, a convenient time and place were set for the interviews.
Two persons typed the data independently, after which both files were confronted to check for inconsistencies, using the software Statistica version 12.0.
To assess the performance of the local health system to eliminate leprosy, the steps defined in other studies for performance assessment were followed [19].
Initially, univariate analysis was carried out with description of position (mean and median) and dispersion measures (standard deviation) of the study variables. Then, the mean score of each indicator in the tool was obtained, considering the sum of scores for each item divided by the number of participants. It also was computed a 95% confidence interval for the mean. The attributes were constructed based on the mean item scores. The following criteria were adopted: indicator below 3 unsatisfactory; between 3 and 4 regular; and 4 or more satisfactory.
Bivariate analyses were conducted, comparing leprosy patient groups with different level of incomes in relation to the attributes investigated. To stratify the groups regarding the income, the researchers considered the quartiles low, medium and high. Subsequently, the authors used one-way ANOVA and Kruskal-Wallis, the latter when the criteria of normality and homoscedasticity were not confirmed [22]. A two-sided p-value of ≤0.05 was defined as statistically significant.
One hundred and nineteen subjects participated in the study. None of the subjects contacted refused to participate. Twenty two percent of the patients (26) were classified as paucibacillary and 78.0% (93) as multibacillary. Table 2 shows the participants' sociodemographic characteristics. There is a balance among the sexes in terms of patients affected by leprosy. The participants' ages varied between 42 and 65 years old and 66.4% (79) had finished primary school. In addition, 50.4% (69) patients were married and most of them gained between 1.2 and 3.3 minimum wages (MW) per month.
Considering the employment status, 47.1% (56) were employed, most of them self-employed; 30.2% (36) were retired and 9.2% (11) disability retired because of the disease. Regarding the patients' housing conditions, most of them lived in their own house, made of concrete.
Table 3 displays the items related to the domains investigated to assess the performance of the local health system. Regarding the first contact domain, it was observed that leprosy patients eventually seek the PHC for preventive actions and also to solve their health problems. It can also be identified that the patients do not access the emergency services when needed.
As to the access to diagnosis (Table 3), it was verified that the patients had to visit the health services thrice before the leprosy diagnosis was reached. The patients had easy access to schedule their medical appointments by telephone; agility in getting the consultations was not verified though, showing unsatisfactory results.
The results also revealed that patients have transportation costs before the diagnosis and during the treatment because the HCS are distant from their homes and the treatment is not supported by the PHC. It was observed that health care professionals have oriented the patients about the leprosy, its signs and symptoms and mode of transmission, among others, but that these professionals do not visit the patients' home and, in case of problems, such as side effects of the medication, the patients do not get a medical appointment within 24 hours.
Table 4 shows the items related to the longitudinality-relational and interpersonal relationship domains. In relation to the longitudinality-relational domain, it was observed that health care professionals register all issues of the patients in the charts and understand their doubts and concerns. The HCP clearly respond to these issues, but do not dialogue about other subjects beyond the disease. Most item scores are satisfactory in this domain, except for the orientation about the types of medication used in the patients' treatment and social support offered by the HCP.
The results also show that patients positively assessed the help offered by HCP and felt welcomed. The patients also refer that the relationship between the HCP and community was positive and well established. It was observed that the HCP were not always available to the leprosy patients on workdays.
Table 5 presents the domains associated to the comprehensiveness of services provided by the HCS. It was observed that all patients received their medication, without any shortages. These aspects were considered satisfactory. There were regular medical appointments for the leprosy patients but the BCG vaccine was not available to all patients' contacts. Although most of the participants were multibacillary, the microscopy was produced insufficiently. It was also observed that HCP did not visit the patients' homes and that care is provided in the HCS only. Also, the HCP did not develop health promotion actions with the patients.
In relation to the coordination of care, Table 5 shows that HCP used the chart to register the issues of the patients and help them to obtain a medical appointment with the specialist professionals, in case of need. Nevertheless, there was no communication flow between these services and the health unit where the patients were being followed to treat leprosy. It was verified in this table that HCS did not share opinions with the specialist professionals about the patients' health of the patients and did not inquire of the patients about the quality of care these providers offered.
Table 6 shows that the health care professionals did not provide care to the patients' family members, nor did they offer the BCG vaccine or investigated the living conditions of all patients' family members. There is no surveillance control of the patients' contacts and the HCP did not engage their families in the care. The family members were not taught about the leprosy disease, its signals and symptoms, transmission mode, therapy, among others.
In relation to collective actions, the results show that HCP did not engage the community in discussions about the problem of leprosy; there was no social mobilization to raise the community's awareness regarding the problem. Active case finding was another issue observed in this study.
In Table 7, when the income groups were compared, statistically significant differences were observed in relation to the first contact; the high-income group evaluated this domain more unsatisfactory than the low and medium income groups (p = 0.002). With regard to access to diagnosis, the low-income group assessed this domain as more unsatisfactory than the medium and income groups (p = 0.03). Concerning the comprehensiveness of services, it was also observed that the high-income group related more unsatisfactory values when compared with the low and medium income groups (p = 0.03).
The study aimed to investigate how the patients perceive the system's performance to eliminate leprosy. As observed, globally, this performance has not been satisfactory, as leprosy is diagnosed in a late stage and not close to the patient's place of residence. It is highlighted that, although the patients turned to the Primary Health Care level when the symptoms started, that is not where the leprosy was diagnosed, which has entailed costs for the patients. At the specialized services where the patients were monitored, they satisfactorily assessed the relation with the health professionals at those services, but these are not articulated with the other care management and coordination services.
Another relevant result is the lack of social support for the patients, without the supply of basic food packages, food tickets, transportation, among other benefits, which are common in other public social programs in Brazil [11]. This system has not prioritized the active search for suspected cases in the community either, nor social mobilization activities with regard to the leprosy problem. Another objective of the study was to identify if the patients' perceptions varied with their income, showing that, with regard to the attributes entry door, access to the diagnosis and list of services, the patients with the highest income assessed these dimensions more negatively.
As regards the profile of the study participants, their sociodemographic characteristics are similar to the results found in the study by Hacker et al [23] regarding the absence of gender differences, with a slight predominance of male patients. The research findings also confirmed the patient profile described in the literature concerning advanced age, low education, own home and informal economic activity [24]–[28];
With respect to the access to the leprosy diagnosis, the patients turn to the units located near their place of residence, but that they did not get their diagnosis at these services, in line with data by Arantes et al [29]. Concerning the Longitudinality-relational dimension, satisfactory results were found for the referral centers, which indicate a good relation between the subjects and health professionals. These findings raise an important discussion about the quality of leprosy care at specialized services which, perhaps due to the establishment of protocols or clinical guidelines, are able to trigger a favorable ambience for bonding, free from discrimination, prejudice and stigma [30].
When considering the attribute Comprehensiveness of services, it was verified that the supply of some technologies for diagnosis and case monitoring is restricted, such as the supply of the BCG vaccine, Mitsuda test and dermato-neurologic review.
According to CDC recommendations [31], surveillance actions should mainly be focused on the patient's intra-domestic contacts, as they people are at a greater risk of being infected than the general population. These cases should be submitted to the dermato-neurologic review and receive the BCG vaccine, which increases the organism's resistance, mainly to the multibacillary forms of the disease. Nevertheless, the study results show weaknesses in the development of these actions, due to the lack of material, and also because the health services do not consider it a priority action.
As regards the collective actions, the results revealed that the health systems have not accomplished interventions in which the community and family are considered in the discussions about leprosy, whether through groups or other work methods, although health education and physical damage prevention and reduction are recommended in the National Plan for the Elimination of Leprosy (PNEH) as fundamental actions for disease control [32]. According to the authors [33], educative activities should become a part of the service routine, in order to disseminate appropriate information and sensitize the community about the disease.
The results also indicated weaknesses in the care coordination, especially in terms of counter-referral and articulation with other specialties for the purpose of case management. Coordination is an important attribute for the organization of the health production process, as this is a structuring principle in the transformation process of Primary Health Care from the perspective of service systems [34]. Without effective coordination, longitudinality loses its potential, comprehensiveness is hampered and the first contact turns into a merely administrative act [35]. In this conjuncture, it is considered that there are few services in Brazil that have been able to advance in terms of integrated health systems or Health Care Networks (HCN) coordinated by Primary Health Care (PHC) [36].
Another attribute assessed refers to the families' participation in care production for the patients, in which weaknesses were identified with regard to the surveillance of domestic contacts, orientations to the families about the disease, its treatment, clarification of other health problems and the application of the BCG vaccine to domestic contacts. The family's role is extremely important, serving as a support network that contributes to the patient's physical and mental balance [37]. When the family does not feel sufficiently informed about the disease, however, negative reactions may be triggered in the care process, besides contributing to the abandonment of treatment [38].
The findings regarding the orientations to the communities affected by the disease permit highlighting the importance of moving beyond the formal teaching spaces, such as knowledge transmission, to a space that accepts pedagogical practices that encourage critical reflections in the community and permit the attribution of a new meaning to the reality [39]. Therefore, orientations to the community need to be provided in a dialogic, emancipating, participatory and creative perspective, so as to contribute to the subjects' autonomy regarding their condition as subjects with rights and authors of their health and disease trajectory [40]. The results indicated a strong relation in the user's perception of the health professional, which may be related to the fact that the care process goes beyond technical competency, encompassing the interpersonal and humanistic aspects of the professional-patient relation. In that context, the health professionals should pay attention to the complaints and propose solutions in combination with the patient, establishing a relation based on welcoming and humanization [41].
When the health systems' performance attributes were verified in relation to patient groups with different income levels, it was observed that high-income groups assessed the entry door as less satisfactory than the low-income groups; the same was true for the comprehensiveness of services attribute.
These results raise the question about the social representation of the Unified Health System (SUS) for segment data about socially disadvantaged populations. In the study by Sorj and Martuccelli [42], the challenge is raised to establish a social protection model in Latin America that rescues the population in its citizenship and state of right. In that sense, the health production model needs to be considered, as well as the extent to which it is contributing to the construction of a universal and fair system.
Also according to the results, the decentralization has not culminated in the expanded access to Primary Health Care services in the local health system investigated. The different hypotheses to explain this result include the difficulty to conform to a new care model that is legitimized by the population as well as the health managers. Brazil has been a stage for disputes among different care models, especially the physician-centered model, which remains very strong [11]. Hence, the study results contain characteristics of this model centered on acute conditions and the disease. Nevertheless, the literature has shown more solid models that can achieve better health results [12].
As potentials, the originality of the research is emphasized, as no other study was found that assessed leprosy care in view of the PHC attributes. In addition, the study can serve as an important tool for managers to define local health policies for the elimination of leprosy and the reorganization of care services under the coordination of primary care professionals in the city under study. In line with some authors, a performance study should neither be an end in itself nor be forwarded as a strictly academic exercise, but should be focused on driving the development of health policies, strategies and programs, which was the direction taken in this study.
Based on the results, it can be concluded that it is important to advance in the reorganization of the health services, to establish communication protocols among the different professionals working in these systems, and to invest in and value PHC, as investments at this care level come with a lower cost and, when well structured and with good problem-solving ability, they can promote a balance between the improvement of the population's health and equity in the distribution of resources [34]. Using PHC as a partner can be an interesting measure to reduce the patients' expenses and involve the community in the treatment of leprosy.
The study limitations refer to the memory bias, in which many patients may not have remembered facts or occurrences, due to the time passed since the end of treatment. In addition, the use of an adapted instrument for leprosy should be highlighted, which may cause an information bias, and the number of losses, as many patients were not located due to a changed or non-existent address.
|
10.1371/journal.pbio.0060149 | Crystal Structure of the FeS Cluster–Containing Nucleotide Excision Repair Helicase XPD | DNA damage recognition by the nucleotide excision repair pathway requires an initial step identifying helical distortions in the DNA and a proofreading step verifying the presence of a lesion. This proofreading step is accomplished in eukaryotes by the TFIIH complex. The critical damage recognition component of TFIIH is the XPD protein, a DNA helicase that unwinds DNA and identifies the damage. Here, we describe the crystal structure of an archaeal XPD protein with high sequence identity to the human XPD protein that reveals how the structural helicase framework is combined with additional elements for strand separation and DNA scanning. Two RecA-like helicase domains are complemented by a 4Fe4S cluster domain, which has been implicated in damage recognition, and an α-helical domain. The first helicase domain together with the helical and 4Fe4S-cluster–containing domains form a central hole with a diameter sufficient in size to allow passage of a single stranded DNA. Based on our results, we suggest a model of how DNA is bound to the XPD protein, and can rationalize several of the mutations in the human XPD gene that lead to one of three severe diseases, xeroderma pigmentosum, Cockayne syndrome, and trichothiodystrophy.
| Preserving the structural integrity of DNA, and hence the genetic information stored in this molecule, is essential for cellular survival. It is estimated that the DNA in each human cell acquires about 104 lesions per day. Consequently, efficient DNA repair mechanisms have evolved to protect the genome. One of these DNA repair mechanisms, nucleotide excision repair (NER), is present in all organisms and is unique in its ability to repair a broad range of damage. In humans, NER is the major repair mechanism protecting DNA from damage induced by ultraviolet light. Defects in the genes and proteins responsible for NER can lead to one of three severe diseases: xeroderma pigmentosum, Cockayne syndrome, and trichothiodystrophy. The XPD protein is one of the key components of a ten-protein complex and is essential to initiate NER. In particular, the XPD protein verifies the presence of damage to the DNA and thereby allows DNA repair to proceed. We have solved the 3-dimensional structure of the XPD protein, and show how XPD has assembled several domains to form a donut-shaped molecule, which is able to separate two DNA strands and scan the DNA for damage. The structure also helps to explain why some of the mutations that have been identified in humans are associated with disease.
| Nucleotide excision repair (NER) is the most versatile DNA repair pathway. [1–5]. NER is well known for its ability to remove bulky DNA lesions and is unique in its ability to repair structurally and chemically different substrates, including benzo[a]pyrene-guanine adducts caused by smoking, as well as guanine-cisplatin adducts formed during chemotherapy [6]. NER is the only repair mechanism in humans that is able to remove photoproducts induced by ultraviolet light. The phenotypic consequences of defective genes involved in NER are apparent in three severe diseases: xeroderma pigmentosum, Cockayne syndrome, and trichothiodystrophy [1,7–10]. The mechanism of the human NER system, while analogous to the well-characterized bacterial system, is less well understood. Over 30 proteins have been identified in humans that are critical for mediating the individual steps leading from damage recognition to incision and repair. However, due to the paucity of specific structural intermediates, the precise role for each protein has not been fully delineated.
NER has been proposed to proceed through either a “bipartite substrate discrimination” or a “multi-partite damage recognition” model [11,12]. It is generally believed that NER is initiated by the combined action of XPC and RAD23B, which recognize a general disruption of Watson-Crick base-pairing created in the vicinity of the damaged nucleotide. Both proteins are required to recruit the ten-subunit transcription factor TFIIH to this site. The XPD and XPB proteins are two helicases that are present in TFIIH, and which open the DNA around the lesion in an ATP-dependent fashion. This is the first catalytic step in this reaction pathway, leading to a conformational change that allows the recruitment of additional NER factors [5,13,14]. A second, more important function of the two helicases is damage verification. Recent data suggest very different roles for XPB and XPD [15]. The helicase activity of the XPB protein seems to be dispensable; however, its ATPase activity is essential for NER. This has been interpreted to suggest a wrapping of the DNA around XPB, which leads to an opening of the double-stranded DNA (dsDNA) close to the lesion. This opening allows the correct binding of XPD, which then utilizes its helicase activity to verify the damage and ensures that the backbone distortion is not the result of an unusual DNA sequence. This process was termed “enzymatic proofreading” and supports the bipartite damage recognition model in which the function of XPC-RAD23B is limited to the observation of a backbone distortion, and XPD is required to verify the damage through its helicase activity [16,17].
Very recently, it has been shown that the XPD protein contains an FeS cluster, which is essential for its function [18]. However, it is not clear whether the cluster has a structural role or is actively involved in the damage recognition process [19]. We solved the crystal structure of the XPD protein from Thermoplasma acidophilum, which shares high sequence identity to its eukaryotic homologs, and show that it contains two RecA-like helicase domains. The XPD protein displays high structural similarity to the bacterial UvrB protein, which is also required for enzymatic proofreading in NER. Two additional domains emerge from the first helicase domain and form a hole that is sufficient to allow passage of ssDNA. Furthermore, the structure delineates how different mutations in the protein lead to the human genetic disorders xeroderma pigmentosum, Cockayne syndrome, and trichothiodystrophy.
Two different XPD-related protein sequences from T. acidophilum have been deposited in the National Center for Biotechnology Information (NCBI) and the Swiss-Prot databases, respectively. They differ only with respect to their N-terminus, with one of them containing 19 additional amino acids. We cloned both constructs and obtained crystals of the shorter protein, which was also active with respect to both its helicase and its ATPase activity (Figure S1). The protein crystallized in space group P65 and the asymmetric unit contains one XPD molecule, indicating no higher oligomeric states, which is consistent with size-exclusion chromatography results and an analysis of the model using the PISA server [20]. The structure was solved by multiwavelength anomalous diffraction (MAD) using the anomalous Fe signal of the endogenous FeS cluster in the protein and was refined at 2.9 Å resolution to an R-factor of 0.209 and Rfree of 0.287 (Table 1). The current model contains residues 23–507 and 515–615 (586 out of 602 residues) of the XPD construct with residues 20 to 22, 508 to 514, and 616 to 620 presumably being disordered.
The structure of the protein can be divided into four distinct domains. Domain 1 is formed by residues 23–87, 178–225, and 366–407, domain 2 by residues 88–177, domain 3 by residues 226–365, and domain 4 by residues 408–615 (Figure 1A and 1B). The first three domains together with α-helix 22 from domain 4 form a donut-shaped structure containing a hole with a diameter of approximately 13 Å (Figure 1A). The remainder of domain 4 is positioned in front of the ring without obstructing the hole of the donut. The overall dimensions of the protein can therefore be divided into the donut with a width and height of 65 Å and 75 Å and a thickness of 29 Å. At the location of domain 4, the width of the ring is increased to 45 Å (Figure 1A and 1B). Domains 1 and 4 represent the “classical” RecA-like fold that is present in all helicases of superfamilies 1 and 2 (SF1 and SF2) [21]. Both domains share approximately 9% sequence identity and can be superimposed with a root mean square (rms) deviation of 2.4 Å using 101 Cα-atoms out of 153 from domain 1, and 201 from domain 4, respectively. Both domains display a similar α/β/α sandwich architecture with a central parallel seven-stranded β-sheet surrounded by seven α-helices in domain 1 and a six-stranded β-sheet surrounded by seven α-helices and two 310 helices in domain 4. The interface between domains 1 and 4 forms the composite ATP binding site. Domain 1 contains helicase motifs I, Ia, II, and III, whereas domain 4 harbors helicase motifs IV, V, and VI [22] (Figure 2). In the context of the overall XPD structure, domain 1 can be viewed as the core domain surrounded by the other three domains. Domains 2 and 3 are insertions, which emerge from domain 1. Domain 2 is inserted between β-strands β3 and β4, while domain 3 is inserted between α-helices α11 and α17. Domain 4 is situated adjacent to domain 1 within the linear protein sequence (Figures 1 and 2).
Notably, the closest related homolog of the full-length XPD structure as revealed by similarity searches [23] was UvrB [24], which has been proposed to be the prokaryotic equivalent to XPD and utilizes its helicase activity for damage verification. XPD and UvrB can be superimposed with an rms deviation of 2.6 Å using 254 aligned Cα atoms out of 588 and 505 residues, respectively. The match is mostly mediated via the two helicase domains, whereas the other domains have no significant structural similarity to each other (Figure S2). In addition, we compared XPD to Hel308 and NS3, two SF2 helicases (Figure S3). The superposition shows that structural similarities are again mainly confined to the RecA domains, whereas the auxiliary domains are highly variable. Hel308 and NS3 have been structurally characterized with DNA substrates, and both represent a closed state of the helicase framework [25,26]. No adenosine nucleotide is bound in these structures, but they are presumed to be in a preprocessive state that only requires ATP binding to reach the processive state [25]. Using the first RecA domain (domain 1) as a reference point for superposition with either Hel308 or NS3, XPD assumes a more open state that is mainly mediated via a rotation of the second RecA domain (domain 4) of about 30° or 16°, respectively, relative to domain 1 (Figure S3C). The composite ATP binding site is located near the hinge region when compared to the closed state of the other two helicases. Our structure may therefore reflect a ground state of XPD prior to nucleotide and/or DNA binding that underlines the conformational flexibility necessary to translate chemical energy into motion.
The first insertion into helicase domain 1 is of particular interest since it contains an FeS cluster, a unique feature among the XPD-like SF2 helicases [18]. Domain 2 displays an exclusively α-helical architecture consisting of six α-helices and one 310 helix that surround the central 4Fe4S cluster (Figure 1A, 1C, and 1D). The FeS cluster is coordinated by four cysteines, consistent with the coordination typically observed in 4Fe4S clusters, and all four cysteines display continuous connectivity in the electron density maps (Figure 1C). A comparison of the B-factors between the 4Fe4S cluster and the surrounding protein residues reveals similar values, indicating full occupancy of the cluster. Three of the coordinating cysteines (Cys92, Cys128, and Cys164) are located in loops, whereas the fourth cysteine, Cys113, is located in a central position within α-helix 5 (Figures 1 and 2). Surprisingly, it was shown that the helicase activity is not affected when Cys102 or Cys105 in Sulfolobus acidocaldarius or Ferroplasma acidarmanus XPD, respectively, were mutated to serine [18,19]. These two residues correspond to Cys113 in our structure. Pugh et al. [19] suggested that the aerobically purified protein most likely contained a degraded 3Fe4S cluster, which is still functional, but presumably a 4Fe4S cluster is present in vivo. When any of the remaining cysteines is mutated to serine, however, the helicase activity of the enzyme is abrogated [18,19].
The cluster is further stabilized predominantly by hydrophobic interactions. Residues Arg88 and Tyr166, which shield the cluster from solvent exposure, are strictly conserved and face towards a pronounced solvent-exposed groove that is formed by α-helices 5 and 8 from domain 2 and α-helix 10 from domain 1 at the back of the protein (Figures 1D and 3). The closest structural homolog for this domain identified by a secondary structure matching search [23] revealed c-myb, a transcription factor that does not contain an FeS cluster [27]. Although c-myb superimposes with a relatively low Q-score of 0.15 (Figure S2B), it is notable that the structural similarity is restricted to the DNA binding interface of c-myb. c-Myb superimposes well with α-helices 5, 6, 7, and 8 of domain 2, of which helices 5 and 8 coincide with the DNA binding interface of c-myb (Figure S2B). In the XPD structure, these helices form part of the groove mentioned above, thus indicating a possible DNA binding site. This is further emphasized by the basic nature of this groove (Figure 3), which is composed of several highly conserved, positively charged residues. However, no significant sequence conservation can be identified between c-myb and XPD in the structurally homologous regions.
Domain 3 consists mostly of extended α-helices (α-helices 12, 13, 14, 15, and 16) and four additional antiparallel β-strands (β6, β7, β8, β9) building a “β-bridge” to domain 1. The β-bridge is further stabilized by α22, an α-helical extension located between β15 and α23 of domain 4. The helices can be grouped into two α-helical hairpins that stack with each other, with one hairpin containing α12 and α13, and the second containing α15 and α16, which is slightly distorted by the insertion of a loop. The two helical hairpins intersect at an angle of approximately 60° and create an extensive hydrophobic core between them. Helix α14 is situated in the V-shaped opening that is formed by the tilt between the two α-helical bundles (Figure 1A and 1B). Similarity searches revealed no significant hit, indicating that this fold has not been encountered previously.
The ring of the donut is closed at its thinnest side via an interface between domains 2 and 3 that has a buried surface area of approximately 620 Å2. The interface is formed by 17 residues from each domain, which display little sequence conservation apart from Phe326, which is always an aromatic residue (Figure 2). Most of the interactions are hydrophobic in character, additionally four salt bridges can be observed between Lys323/Asp99, Arg335/Glu103, Arg235/Glu103, and Glu315/Lys111.
Since the presence of the FeS cluster is essential for helicase activity on dsDNA [18,19], it prompted us to investigate the only other structurally characterized DNA-binding proteins with such a feature, the base excision repair proteins, MutY and Endo III [28,29], with a focus on the first because a structure of a MutY-DNA complex has been described [28]. For MutY, it was shown that its FeS cluster is required for enzymatic activity and DNA binding [30]. The XPD protein contains a loop motif in the FeS cluster domain with a high density of positively charged residues similar to the FeS cluster loop motif (FCL) in MutY [31]. The superposition of the XPD and MutY FeS cluster domains (Figure 4) reveals a similar orientation of two conserved arginines (Arg88 in XPD and Arg153 in MutY). In MutY, it was shown that a neighboring conserved arginine, Arg149, is perfectly positioned for an interaction with the DNA backbone, and bridges the distance between Arg153 and the DNA [32]. Based on the similarity to MutY, Arg88 in XPD may fulfill a similar function. Furthermore, the position of Arg88 at the surface of a pocket where DNA recognition could take place supports the idea proposed by Lukianova et al. that the FeS cluster plays an important role in arranging the residues of the FCL motif for DNA binding [31].
For MutY, it was shown that the redox properties of the [4Fe-4S]2+ cluster are modulated by the presence of DNA [33]. DNA-binding activates the cluster and facilitates oxidation [34]. Boal et al. proposed a model for DNA-mediated charge transfer (CT) in DNA repair in which one electron is transferred from the cluster to the DNA. In this model, the CT acts as an initial sorting mechanism, enabling a rapid scanning of undamaged regions by several glycosylase molecules, so that they are able to relocate themselves onto sites near the damage [34]. In NER, an analogous scanning mechanism seems unlikely, but a change in oxidation state of the 4Fe4S cluster upon DNA binding and as part of the damage verification step may be required, thus suggesting a functional role for the 4Fe4S cluster and not just a structural role. This hypothesis is further supported by site-directed mutagenesis studies that demonstrate that single mutations of three of the four 4Fe4S cluster coordinating cysteines to serine lead to a loss of the 4Fe4S cluster, and abrogate helicase activity, but retain a correctly folded protein that is still able to translate along ssDNA [18,19].
The XPD protein is a member of the SF2 helicases. To obtain insight into the DNA binding mode of XPD, we calculated the electrostatic surface potential of the protein and searched for conserved solvent-exposed amino acids (Figures 2, 3, 5, and 6). The surface potential indicates a positively charged path for dsDNA along domain 4, leading towards a highly conserved groove along domain 4 and domain 1, which provides sufficient space for ssDNA and directs the DNA towards the hole formed by domains 1–3. The dsDNA requires separation into ssDNA prior to entering the groove. Recently, the structure of the SF2 helicase Hel308 was determined in complex with DNA, and a prominent β-hairpin in the second helicase domain was identified that is responsible for initial strand separation [25]. It was proposed that this β-hairpin could be a general feature of SF2 helicases. In XPD however, this “wedge” is formed more likely by an α-helical extension in domain 4 (Figure 5). Despite the difference in secondary structure, it is located between helicase motifs V and VI as demonstrated for Hel308 and proposed for NS3 [25] (Figure S3). Two α-helices in XPD, α22 and α23, form two walls of the wedge and extend farther out towards the solvent compared to other helicases such as UvrD and PcrA [35,36].
We propose that the tip of the wedge composed of residues in the loop between α22 and α23 separates the two DNA strands. The last two turns of α22 and the first two helical turns of α23 contain several aromatic amino acids, which could stabilize the separated DNA strands in a fashion similar to that observed for Hel308. On one side of this wedge, the highly conserved residues Tyr540 and Tyr545 are oriented with their side chains pointing towards the solvent where they could easily form stacking interactions with the exposed bases of ssDNA. These stacking interactions can then be continued by additional solvent-exposed aromatic residues, such as Tyr23, leading the ssDNA along the back of the protein to a position where the two strands meet again to reform dsDNA. Although exposed aromatic residues are also present on the other side of the wedge, their degree of conservation is relatively low. In our structure, Phe538 and Tyr425 could both stack against the bases in ssDNA. However, only Tyr425 is conserved, whereas Phe538 is replaced by a leucine in eukaryotic XPDs. This substitution appears to be compensated by the occurrence of Phe651 in human XPD, which substitutes for Ser552 in T. acidophilum XPD; and due to the close spatial proximity of the two side chains, they would assume similar positions (Figure 5). Consequently, there is one phenylalanine available that would represent the required stacking partner. In addition, several highly conserved, positively charged residues, such as Lys583 and Lys424, apparently define the path for the second strand leading into the groove described above and from there continues through the central hole (Figures 5 and 6).
Despite the fact that we crystallized the protein in the absence of DNA and phosphate buffer, we identified significant peaks with heights of more than 2.5 times the rms deviation in difference electron density maps (Figure 6) that are spaced by approximately 6.5 Å, as well as slightly longer distances and cannot be explained by the protein model. Since the distance between phosphates in ssDNA is approximately 6.4 Å, it is therefore very tempting to speculate that some DNA remains bound to the protein during purification and gives rise to these residual electron density features. Further support for this hypothesis is provided by the superposition of our structure with NS3 helicase in complex with ssDNA [26] (Figure S3). Based on this superposition, we have built a model for a ssDNA binding mode (Figure 6) in which the extension of the ssDNA towards the hole positions three of the phosphates into the residual electron density peaks.
The postulated DNA route passes by another highly conserved surface feature in XPD, a narrow pocket that is formed by the strictly conserved Arg88 and Tyr166 on one side and Tyr185 on the other side, and is located in the wall of the central hole, directly adjacent to the 4Fe4S cluster (Figure 6). The dimensions and shape of this pocket are ideally suited to accommodate a nonmodified purine or pyrimidine base, which would be held in place through van der Waals interactions with the residues mentioned above. Due to its location, this surface feature would allow a direct coupling between the FeS cluster and a readout of the DNA. This pocket is reminiscent of the pocket for the flipped-out base that was observed in the UvrB-DNA structure [24] .
Initial DNA distortion recognition in eukaryotes is achieved through the combined action of XPC and RAD23B [37]. It was shown that with the recruitment of TFIIH to the site of damage, the helicase XPD is required for proofreading, whereas XPB fulfills a structural role [15]. In the absence of an XPD-DNA complex containing a lesion, the process of proofreading remains highly speculative. The structure of XPD clearly reveals structural homology to its prokaryotic homolog UvrB. In UvrB, it was shown that a β-hairpin, which emerges from the first helicase domain, is critical for damage recognition [38–40]. However, despite the structural similarity between the two proteins, XPD does not contain a corresponding feature. In our model of the XPD-DNA complex (Figure 6), we propose that one of the DNA strands passes through the central hole, which is formed by domains 1–3. According to studies by Naegeli et al. [41], this would be the translocating strand, which contains the lesion, and leads to a stalled protein-DNA complex. The dimension of this hole, with a diameter of 13 Å, however, is most likely too big to provide a trap for damaged DNA. One possible candidate for the “analysis” of each base with respect to their correct structure would be the narrow pocket in the wall of the central hole described above. The size of this pocket suggests that only nondamaged bases could be accommodated, whereas a bulky DNA substrate would be excluded. This pocket is also an attractive candidate for the damage recognition process due to its close proximity to the 4Fe4S cluster and the involvement of Arg112 of human XPD (Arg88 in our structure), which has been shown to cause trichothiodystrophy when mutated to histidine.
TFIIH in humans is not only required for DNA repair, but is also essential for transcription [42]. XPD represents one of the ten protein subunits of TFIIH and interacts tightly with the N-terminal 236 amino acids of p44. This interaction results in a 10-fold increase in its helicase activity [43]. It has been shown that the helicase function of XPD is not required for transcription, but is essential for NER [44]. On the other hand, XPD is required to stabilize the interaction between the core TFIIH complex, which contains seven subunits, and the cdk-activating kinase (CAK) subcomplex, consisting of the remaining three subunits [45,46]. Mutations in XPD (Figure 2 and 7) can therefore lead to three different effects. The first class of mutations affects the activity of the protein directly, whereas the second group can lead to impaired interactions with p44, thus affecting its own activity in an indirect way. The third group of mutations may lead to a destabilization of TFIIH, thereby reducing overall transcriptional activity. Based on our structure the effects of several point mutations leading to xeroderma pigmentosum, Cockayne syndrome, or trichothiodystrophy can be explained (Figure 7). Point mutations associated with xeroderma pigmentosum, such as G47R, D234N, and R666W, are located in helicase motifs I, II, and VI, respectively, and impair the ability to bind and hydrolyze ATP, thus inactivating the enzyme; however, point mutations within other regions have quite distinct effects.
Arg112 (Arg88 in T. acidophilum) is located in the FeS cluster domain and is in direct contact with the cluster. A mutation of this residue to histidine has been identified in several TTD patients [47]. Analysis of the equivalent residue in S. acidocaldarius XPD abolished its helicase activity [18]. Arg88 is located in close vicinity to Cys113 one of the Fe-ligands, and shields the cluster, with its long side chain, from solvent. It is the first residue in a short α-helix, α 3, which together with the opposite side of the helix forms one wall of the hole where ssDNA most likely passes through (Figure 6). The proposed role for Arg88 in analogy to MutY as described above may be accomplished by Arg112 in the human XPD protein and a mutation to histidine, as observed in trichothiodystrophy patients, could prevent this interaction, thus reducing the affinity of the protein to the DNA. However, the exact role of the 4Fe4S cluster, whether it is involved directly in the recognition process or the translocation along the DNA, remains speculative at this point. It is interesting to note that Egly and coworkers have shown this variant in human XPD to be completely devoid of helicase activity [48].
The effects of the C259Y variant can also be readily explained. This cysteine is replaced in T. acidophilum by another small hydrophobic residue, Ala236, in α-helix 12, which points into the hydrophobic core within domain 3. This core stabilizes the relative position of the four α-helices within this domain as outlined above. Replacing this small hydrophobic residue with a tyrosine leads to severe steric clashes within this core and thereby destabilizes the entire domain.
The two mutants Y542C and G602D are very close to each other in the structure. Tyr458 (Tyr542 in human XPD) is located at the beginning of α-helix 20 in domain 4 and forms hydrophobic interactions with another strictly conserved residue, Val501 (Val599 in human XPD), in a neighboring β-strand. Replacing the tyrosine with a cysteine would weaken the interactions between this helix-strand pair. Gly504 (Gly602 in the human enzyme) is positioned between β-strands 14 and 15 in domain 4. If this residue were to be replaced by a larger residue, it would point towards Tyr458 (Tyr542 in human XPD) and would thereby interfere with this side chain.
The remaining four mutations D673G, G675R, D681N, and R683W/Q, although causing different diseases, are all clustered closely together towards the C-terminal end of the human XPD protein and correspond to residues Asp574, Gly576, Asp582, and Arg584 in T. acidophilum XPD, respectively. It has been speculated that residues at the C-terminal end of human XPD interfere with p44 binding, thus leading to an inability to stimulate the helicase activity of XPD [43]. Of these four mutations, only G675R was analyzed with respect to its ability to interact with p44, and it was shown that the interaction was severely diminished [43]. All other analyzed disease mutants are located further towards the C-terminal end of human XPD where our archaeal XPD contains no corresponding residues, which is not unexpected since T. acidophilum does not contain a p44 homolog. T. acidophilum Asp574, Gly576, Asp582, and Arg584 are located in domain 4 and fulfill important structural roles. Asp574 forms interactions with the strictly conserved Arg570 (Arg669 in the human enzyme), which is located at the end of helix 24, and thereby stabilizes the transition from the helix to the following β-strand 16. Gly576 is positioned in this β-strand and points towards two hydrophobic residues, Leu568 and Ile569 (Ala667 and Ile668 in human XPD) in α24. A mutation of Gly675 to an arginine would push the entire helix away from the β-strand and thereby destabilize the integrity of the domain. Asp582 is located directly behind β16 and forms tight interactions with the strictly conserved Arg584 (Arg683 in human XPD), and the latter forms additional interactions with Asp426 and Phe527 (Glu509 and Tyr625 in human XPD), two highly conserved residues. The point mutations at the C-terminal end of XPD thus clearly play important structural roles, and any of the four mutations would interfere with the fold of domain 4, which could also diminish the interactions with p44. According to our protein–DNA model, however, T. acidophilum Arg584 (Arg683 in human XPD) also plays an important role in DNA binding and is one of the residues that may bind to the DNA close to the double-strand/single-strand junction. Replacing this positively charged residue with either a glutamine or tryptophan may severely interfere with DNA binding and thereby lead to the disease phenotype.
The crystal structure of XPD from T. acidophilum revealed that the protein contains two RecA-like helicase domains and two additional domains that emerge from the first helicase domain. Surprisingly, the first three domains form a donut-shaped structure and a protein–DNA model is proposed in which one of the ssDNA strands passes through this central hole in close spatial proximity to the 4Fe4S cluster in the second domain. The high sequence homology to eukaryotic XPDs allowed the analysis of mutations leading to one of the three severe diseases xeroderma pigmentosum, Cockayne syndrome, or trichothiodystrophy and provides the basis for a more detailed analysis to understand the combined action of the helicase and the 4Fe4S cluster to achieve damage verification within the NER repair cascade.
The genes encoding two XPDs from T. acidophilum with variable N-termini (residues 1–622 and 23–622) were cloned into the pET16b vector (Novagen) using the NdeI and XhoI restriction sites. XPD was expressed as an N-terminally His-tagged protein in Escherichia coli BL21-CodonPlus (DE3)-RIL cells (Stratagene) by induction with 0.1 mM isopropyl-β-thiogalactoside at 14 °C for 18 h. The protein was purified by metal affinity chromatography (Ni-NTA; Invitrogen) followed by size-exclusion chromatography (HiLoad 26/60 Superdex 200 prep grade; GE Healthcare) in 20 mM Tris (pH 8) and 500 mM NaCl. The protein was concentrated to 5 mg/ml based on a molar absorption coefficient of 65,140 M−1 cm−1.
For construction of the 5′ overhang DNA substrate, a 25-mer oligonucleotide (MDJ1, 5′-GACTACGTACTGTTACGGCTCCATC-3′) was 5' end labeled and annealed to the 3' end of a 50-mer oligonucleotide (NDB, GCAGATCTGGCCTGATTGCGGTAGAGATGGAGCCGTAACAGTACGTAGTC). The helicase assay was carried out as described by [18] with slight modifications. Briefly, the reactions (10 μl) were incubated at room temperature in 20 mM MES (pH 6.5), 1 mM DTT, 0.1 mg/ml BSA, 5 mM MgCl2, 10 nM 32P-labeled DNA substrate, and 500 nM XPD for 10 min. The reactions were started by the addition of 3 mM ATP and transferred to a 45 °C water bath. After the specified time, 20 μl of stop solution (10 mM Tris-HCl [pH8]. 5 mM EDTA, 5 μM cold competitor [MDJ1], 0.5% SDS, and 1 mg/ml proteinase K) was added and incubated for 15 min at 37 °C to allow proteinase K digestion. Samples were separated on a native 10% acrylamide:bis TBE gel for 1 h at 100 V.
XPD crystals were grown by vapor diffusion in hanging drops containing equal volumes of protein in 20 mM Tris/HCl (pH 8.0) and 500 mM NaCl at a concentration of 5 mg/ml, and a reservoir solution consisting of 200 mM MgCl2, 100 mM Hepes (pH 8), and 5%–10% PEG 400 equilibrated against the reservoir solution. Crystals grew within 7 d at 20 °C to a maximum size of 100 × 50 × 50 μm3. Prior to data collection, the crystals were cryocooled by sequential transfer into mother liquor containing increasing amounts of glycerol in 5% steps to a final concentration of 30%.
The crystals were flash cooled in liquid nitrogen, and data collection was performed at 100 K. Data sets were collected at beamline BM14 (European Synchrotron Radiation Facility [ESRF]) at wavelengths of 1.0 Å, 1.7 Å, 1.7367 Å, and 1.7419 Å. All data were indexed and processed using Moslfm and Scala [49,50]. The crystals belong to space group P65 with unit cell dimensions of a = b = 78.9 Å, c = 174.0 Å. Structure solution was achieved utilizing the anomalous signal of the endogenous Fe belonging to the 4Fe4S cluster by MAD data collection at the Fe edge. The peak and inflection datasets were obtained from one crystal and were merged with a highly isomorphous dataset collected at the remote wavelength. The Fe sites were located using ShelxD [51], and phase improvement was achieved with Sharp [52]. Substructure solution and refinement was carried out at 4 Å resolution, and the 4Fe4S cluster was treated as a “super” atom for phasing. The initial maps were subjected to solvent flattening and phase extension to 3.6 Å using the programs Solomon [53] and Pirate [54]. The solvent-flattened maps were autotraced using the low-resolution quick-build option in ARP/WARP [55] and further extended manually using the programs O and Coot [56,57]. After assigning the maximum number of residues and side chains possible, the model was subjected to phase-restrained simulated annealing and maximum likelihood refinement using the program phenix.refine [58]. Refinement was carried out against the highest resolution dataset up to 2.9 Å. The model was further improved by alternating rounds of refinement and manual model building. When the model was sufficiently complete, refinement continued with TLS and restraint maximum likelihood refinement using Refmac5 [54]. The final model contains 586 out of 602 amino acid residues, the 4Fe-4S cluster, one calcium ion, and one water molecule.
Coordinates and structure factors for the XPD structure have been deposited in the Protein Data Bank (http://www.rcsb.org/pdb/home/home.do) using the Autodep tool from the European Bioinformatics Institute (http://www.ebi.ac.uk/) with the entry code 2VSF.
The Protein Data Bank accession numbers for the proteins discussed in the paper are as follow: c-myb (1h89), UvrB (2fdc), Hel308 (2p6r), and NS3 (1a1v).
|
10.1371/journal.pntd.0000716 | New Assembly, Reannotation and Analysis of the Entamoeba histolytica Genome Reveal New Genomic Features and Protein Content Information | In order to maintain genome information accurately and relevantly, original genome annotations need to be updated and evaluated regularly. Manual reannotation of genomes is important as it can significantly reduce the propagation of errors and consequently diminishes the time spent on mistaken research. For this reason, after five years from the initial submission of the Entamoeba histolytica draft genome publication, we have re-examined the original 23 Mb assembly and the annotation of the predicted genes.
The evaluation of the genomic sequence led to the identification of more than one hundred artifactual tandem duplications that were eliminated by re-assembling the genome. The reannotation was done using a combination of manual and automated genome analysis. The new 20 Mb assembly contains 1,496 scaffolds and 8,201 predicted genes, of which 60% are identical to the initial annotation and the remaining 40% underwent structural changes. Functional classification of 60% of the genes was modified based on recent sequence comparisons and new experimental data. We have assigned putative function to 3,788 proteins (46% of the predicted proteome) based on the annotation of predicted gene families, and have identified 58 protein families of five or more members that share no homology with known proteins and thus could be entamoeba specific. Genome analysis also revealed new features such as the presence of segmental duplications of up to 16 kb flanked by inverted repeats, and the tight association of some gene families with transposable elements.
This new genome annotation and analysis represents a more refined and accurate blueprint of the pathogen genome, and provides an upgraded tool as reference for the study of many important aspects of E. histolytica biology, such as genome evolution and pathogenesis.
| Entamoeba histolytica is an anaerobic parasitic protozoan that causes amoebic dysentery. The parasites colonize the large intestine, but under some circumstances may invade the intestinal mucosa, enter the bloodstream and lead to the formation of abscesses such amoebic liver abscesses. The draft genome of E. histolytica, published in 2005, provided the scientific community with the first comprehensive view of the gene set for this parasite and important tools for elucidating the genetic basis of Entamoeba pathogenicity. Because complete genetic knowledge is critical for drug discovery and potential vaccine development for amoebiases, we have re-examined the original draft genome for E. histolytica. We have corrected the sequence assembly, improved the gene predictions and refreshed the functional gene assignments. As a result, this effort has led to a more accurate gene annotation, and the discovery of novel features, such as the presence of genome segmental duplications and the close association of some gene families with transposable elements. We believe that continuing efforts to improve genomic data will undoubtedly help to identify and characterize potential targets for amoebiasis control, as well as to contribute to a better understanding of genome evolution and pathogenesis for this parasite.
| Although many infectious diseases receive little attention in today's world, the pathogenic intestinal parasite E. histolytica occupies a major place in the list of ignored illnesses. The parasite is the causative agent of amoebiasis, causes a significant level of morbidity and mortality in developing countries, and affects at least 50 million people every year, causing over 100,000 deaths [1]. Yet, a lot is there to be learned about this important protozoan. Genome information allows for better understanding of pathogenic processes and consequently helps improve the prevention, diagnosis, and treatment of the disease. Therefore, accurate and up to date data is fundamental to generate a reliable tool for both research and medical use. The E. histolytica genome was automatically annotated and published in 2005 [2]. This draft genome provided the scientific community with the first blueprint of this pathogen, its gene organization and content. However, genome annotation was performed in an automated way, leading to a very raw dataset to work with. Here, in an effort to improve the structural and functional annotation for this organism, we have reviewed, re-assembled and re-annotated the E. histolytica genome. The ultimate goal was to generate a high-quality annotation dataset to be used as gold standard by the scientific community and to carry on comparative analysis with the closely related species Entamoeba dispar and Entamoeba invadens. Using a combination of manual and automated methods we significantly improved the E. histolytica assembly. In addition, we generated a new training set of genes for training gene finders, created new gene models and reevaluated and refined previous gene structures based on up to date information, reassessed gene functions, and mapped transposable elements to remove overlapping predicted genes. Here we present an overview of the methods employed for this task and protocols followed, summarizing the contents of the latest data release, with special emphasis on our final assembly and annotation release.
Reads were obtained directly from the Sanger Institute and JCVI databases. Reads were filtered based on similarity to an E. histolytica plasmid sequence [3] or to tRNA models [4]. Reads were assembled with UMD Overlapper [5] and Celera Assembler [6]. See Text S1 for assembly details. The re-assembled sequence was deposited at the National Center for Biotechnology Information (NCBI) with the accession number AAFB02000000.
A set of 20,192 ESTs and 71 full-length cDNAs were downloaded form GenBank. ESTs were assembled and aligned to the newly assembled genome using PASA [7]. A training set consisting of 300 genes supported by 60 full length cDNAs and 240 assembled ESTs was created to train the following gene finders: Genezilla [8], and GlimmerHMM [8]. EVidenceModeler (EVM) [9] was used to generate the new gene dataset, as a weighted consensus of all available evidence, including proteins and conserved protein-domains alignments, cDNAs/ESTs and gene finder output predictions. The new datas[6]et was manually inspected in areas covered by transposable elements (see below). Coding regions shorter than 300 bp supported by no evidence other than Gene Finders were eliminated from the gene dataset. To generate more accurate gene structures in our new dataset, we focused on structural reannotation by improving the accuracy of existing gene models, validating intron/exon boundaries, incorporation of UTRs when available (using PASA), identifying pseudogenes and eliminating spurious genes.
First, we created a comprehensive custom database containing all reported E. histolytica repetitive elements: LINEs, SINEs, EhERE1 and EhERE2 [10]. Then, we ran RepeatMasker (http://www.repeatmasker.org/) on the current assembly to map and quantify the elements. Regions of the genome that match the repeats were masked to avoid gene prediction on these regions. Any gene predicted on masked regions was removed from the annotation.
Predicted gene models from the previous assembly were mapped to the new assembly using a combination of methods (Fig. 1). First we identified the correspondence between the scaffolds in the first assembly and the new assembly. Once this correspondence was identified, gene models from the old annotation were mapped onto the new assembly in a multistep fashion. During the first mapping iteration performed with an in-house tool, annotation_transfer, based on the software Mummer [11], not all models were transferred as expected due to small sequence variation resulting from a new, independent assembly. In a second mapping round, unmapped genes were aligned to the new assembly using GeneWise [12] an algorithm that combines protein alignment and gene prediction into a single statistical model as a paired Hidden Markov Model (HMM) and provides a gene prediction based on protein homology. Then, genes that failed to map by the previous methods were positioned on the new assembly by tblastn, using a coverage of at least 80% identity, 80% coverage, and an e-value <1×10−20. Finally, structural changes between OGA and NGA predictions was assessed using GSAC (Gene Structure Annotation Comparison, unpublished), a JCVI in-house tool that evaluates coordinate differences between two gff3 (generic feature format version 3) files (http://www.sequenceontology.org/gff3.shtml).
To evaluate the structural improvement of gene models in the new annotation we selected a dataset of 1024 pairs of genes. Each pair was composed of an OGA and a NGA gene that only map to each other (i.e. they represent the same gene in each annotation) but are structurally different. This dataset was used to perform two types of analyses. First we ran HMM-searches on each pair against the Pfam HMM database and then, we evaluated NGA HMM-searches statistic (e-value, score or number of a particular Pfam domain) compared to their OGA counterparts.
In addition, we performed local blastp searches against our internal non-redundant protein database, PANDA.db (ftp.jcvi.org/pub/data/panda) and identified pairs that shared the same top-hit to run stretcher, a global pairwise alignment tool (bioweb2.pasteur.fr/docs/EMBOSS/stretcher.html), between each gene and its corresponding top-hit. Pairs having hits with percent identity below 30% were removed from the results to eliminate false positive hits and results for each pair were analyzed according to their alignment statistics (score, percent identity, percent similarity and percent of gaps) to determine the level of improvement between the annotations. For measuring functional annotation improvement, we estimated the number of genes in the NGA that acquired a descriptive name or an improved name with respect to the OGA only for those genes that did not undergo structural changes to discard functional improvements associated with drastic structural changes, such as incorporation of new exons and changes in coding frame.
Gene level searches were performed against protein, domain and profile databases including JCVI in-house non-redundant protein database Panda-AllGroup.niaa, Pfam [13] and TIGRfam [14] HMMs, Prosite [15], and InterPro [16]. In addition, programs to predict membrane localization such as SignalP [17], TMHMM and TargetP [17] were run. After the working gene set had been assigned function, predicted proteins were organized into protein families as previously described [7] with the purpose of refining the annotation in the context of related genes in the genome. Predicted genes were assigned informative names and classified using Gene Ontology (GO) [18]. GO assignments were attributed automatically, based on other assignments from closely related organisms using Pfam2GO, a tool that allows automatic mapping of Pfam hits to GO assignments as well as manually by expert annotators. All assignments were reviewed manually for consistency, on a family based approach, via Manatee, a web-based genome annotation tool that can view, modify, and store annotation for prokaryotic and eukaryotic genomes. Names between OGA and NGA were compared by simple query for corresponding genes to determine the level of change and improvement. Annotation of transporter proteins was performed using TransportDB (http://www.membranetransport.org/) [19].
Segmental genome duplications along the E. histolytica genome were identified with DAGchainner [20], a program that looks for chains of syntenic genes within complete genome sequences, using default parameters. Briefly, we performed all-vs-all blastp searches using the E. histolytica proteome. The blastp output was then filtered out to remove repetitive matches that could potentially contribute noise to the data. Finally, all segmental genome duplications containing five or more duplicated set of genes were further analyzed.
Close examination of the initial assembly of E. histolytica strain HM-1:IMSS revealed multiple problems. Sequence analysis using intra-scaffold dot-plots exposed 161 artifactual tandem duplications (Figure S1, panel A) located at the boundaries between neighboring contigs (a contiguous assembled sequence ordered together to form a scaffold). Tandem duplications spanned 364,707 bp of genomic sequence with a median length of 892 bp. In the previous assembly, genes predicted on these regions and on unmasked repetitive regions caused an over-estimation of genes by approximately 18%. Indeed, of the 399 genes located in those regions, 61 hit transposable elements (TEs) or were likely pseudogenes, while most of the remaining 338 coding sequences were artifactually duplicated and so collapsed into 206 individual genes in the new annotation (Figure S1, panel B). A comparative description of the features of the original and the new E. histolytica assemblies is summarized in Table 1A.
The new genome assembly consists of ∼20 Mb of sequence organized into 1,496 scaffolds. To generate a “core” assembly for functional annotation, scaffolds lacking predicted genes were not considered. The resulting core assembly consisted of 818 non-redundant scaffolds with a total of 19,220,345 bp. All scaffolds that were excluded from the core assembly as well as degenerate contigs and singleton reads, although not annotated, were considered to survey the presence or absence of genes when necessary, and all sequences were deposited in GenBank (see Methods).
The results of the new assembly show higher fragmentation and a reduction in genome size with respect of the published assembly. However, our comparative analysis between the two annotations shows that there is no loss of coding information from one assembly to the other.
The new assembly contains 8,201 de novo predicted protein coding genes, 1,784 fewer than previously reported for this genome (Table 1) [2]. To determine the origin of these differences and to evaluate changes in gene structure between the original (OGA) and new (NGA) annotation, genes from OGA were mapped onto the new assembly and structural differences were estimated using GSAC (see Methods and Fig. 1A). Mapping results indicated that the main reason for gene number reduction is the elimination of genes within repetitive regions and artifactual tandem duplications, and the removal of genes smaller than 300 bp without any supporting evidence (Fig. 1B). Noteworthy, less than 0.2% of the genes from the original annotation do not map onto the new assembly, despite the fact that the assembly is 2,562,911 bp smaller than the published one. These missing OGA genes contained no supporting evidence and were originally annotated as hypothetical protein coding genes. This analysis also showed that 51% of the OGA genes keep the same structure in the new annotation (same isoform in Fig. 1B), while 36% underwent structural change (different isoform in Fig. 1B).
As part of the curation process, the structure of 740 genes was manually reviewed and curated based on supporting evidence such as ESTs. An important hallmark of this work is the concerted effort from scientists of the Entamoeba community that contributed to the curation of the genome by direct communication with the authors as well as participation via specific workshops held at JCVI.
To evaluate whether structural changes in the new annotation reflect an overall improvement of gene structures we selected a group of 1,024 OGA-NGA pairs of genes that map to each other but are structurally different. Then, we ran HMM-searches and global pairwise alignments on each pair of proteins against Pfam HMMs and our PANDA database (see Methods). Finally, we compared the resulting statistics between OGA and NGA peptides from each pair (Fig. 2). These analyses showed that translated products from NGA genes consistently give better hits against Pfam and PANDA databases when compared to OGA genes, demonstrating an overall improvement in gene structures for the new annotation. In those cases where NGA genes gave worse hits compared to their OGA counterparts, we manually inspected and corrected gene structures in the new annotation.
Structural improvements in the new annotation were also reflected by (1) the appearance of new Pfam/TIGRfam domain hits not present in the original protein dataset and (2) the identification of genes coding for additional members of different protein families. Noteworthy, among novel protein domains are a domain typically found in some subunits of several DNA polymerases (PF04042), a domain found in phospholipid methyltransferases (PF04191) and another present in panthotenate kinase proteins (PF03630, see section below). On the other hand, point (2) is very well exemplified by the subunits of the Gal/GalNAc lectins. In E. histolytica these lectins are composed of three different subunits: a 170 kDa heavy subunit (Hgl), a 150 kDa intermediate subunit (Igl) and a 31–35 kDa light subunit (Lgl) [21], [22]. In agreement with the current number of Hgl and Lgl genes in the new annotation, studies of pulse-field gel electrophoresis have shown that there are five hgl and six lgl genes in the genome [22]. However, only four Hgl genes, one of them truncated, and four Lgl genes are part of the old dataset.
Particular effort was directed towards the improvement of functional annotation (summarized in Table 1B) by the incorporation of additional 974 enzyme commission (EC) numbers and 531 Pfam/TIGRfam domains. Gene ontology (GO) terms were automatically assigned from Pfam HMM searches refreshing and updating the assignments from InterPro evidence used in the old annotation. The advantage of using hits from Pfam HMM searches is that results can then be filtered not just by e-value but also by trusted cutoff scores, giving a more accurate estimation than InterPro searches and therefore, a more confident GO assignment. In addition to automatic EC number and GO term assignments, functional annotation has been manually curated for 2,130 genes. A total of 3,468 genes have been assigned GO terms, of which 3,216 have a molecular function term. We have distributed the specific terms in a total of 30 molecular function GO-Slim categories (Table S1). No difference was observed in the representation of GO categories in the protein families with respect to that of singletons.
E. histolytica predicted proteins were organized into protein families to facilitate the review of their functional annotation, visualizing relationships between proteins and allowing annotators to examine related genes as a group. Our family clustering method produces groups of proteins sharing protein domains conserved across the proteome, and consequently, related biochemical function, as described in Methods [23], [24]. For example, based on our clustering criteria, all proteins containing a single RhoGAP domain (PF00620) fall within the same family irrespectively of their length.
A total of 897 protein families containing 4,564 proteins (56% of the proteome) were identified from the 8,201 predicted polypeptides in the new annotation, leaving 3,637 “orphan” proteins. Among the families, 247 clusters (479 proteins) have no homology to any known Pfam or TIGRfam domain, and harbor potentially novel domains (91 of these families contain five members or more). On average, E. histolytica families contain five proteins, ranging from two to 149 members (Fig. 3A). We identified seven families with more than 50 members encoding proteins such as small GTP binding proteins, BspA-like leucine-rich repeat proteins, kinase domain-containing proteins, WD domain-containing proteins, a large family of uncharacterized hypothetical proteins, a RNA recognition motif domain-containing protein family and a RhoGAP domain-containing protein family (see Table S2 for the complete list of families).
Interestingly, a number of protein families appear to be physically linked to transposable elements. Table 2 shows the top 27 families that present this type of association (for the entire repertoire of genes see Table S3). For example, a cluster of 31 members of the Hsp70 protein family appears associated 30% of the time with TEs within 1 kb of the gene context. Hsp70 proteins are molecular chaperones that assist a large variety of protein folding processes in the cell by the transient association between their substrate-binding domain and the short hydrophobic peptide segments present in their target proteins. Hsp70 s are highly conserved and are known to be induced by a variety of stresses [25]. It has been previously reported that multiple natural TE insertions in Drosophila reduce the level of expression of hsp70 genes by insertion nearby gene promoter regions [26]. The characteristics of the hsp70 promoter in the fly may make it a suitable target for transposition leading to the generation of novel alleles. In this sense, TEs could be playing an adaptive role in microevolution by gene amplification and also manipulating the expression of genes critical for the parasite fitness [27].
Another family showing a high correlation with transposable elements is the large BspA-like surface protein family [28], [29]. Initially, Davis et al. identified 89 genes coding for BspA-like proteins in the genome of E. histolytica, containing a leucine-rich repeat motif (LRRs). LRRs serve as recognition motifs for surface proteins in bacteria and other eukaryotes [30] and have been shown to be involved in binding to fibronectin. E. histolytica BspA-like proteins have unique LRR-like repeats that resemble, to certain extent, to the Treponema pallidum LLRs (LrrA proteins) [28], that appear to have a role in attachment and penetration to host tissues [31], suggesting they may be involved in attachment to the host cells. Our analysis identified 116 BspA-like genes in the genome, 41 of them associated with transposable elements. The core domain of the BspA-like proteins contains 23 amino acids with the consensus P[T/S][T/S][V/I/L]xx[I/L]GxxCFxxCxxLxx[I/L]x[I/L], and these units form tandem blocks that can contain two or more core motifs represented from 1 to 21 times in a single molecule, leading to a great variability in the protein length in the family. Most of the proteins in the family contain a novel 50 amino acids N-terminal domain that is preserved in 85 members of this cluster. A closer examination of those genes encoding proteins lacking the N-terminal domain showed they are probably truncated by the insertion of transposable elements, primarily SINE and LINE elements at their 5′ end. BspA-like proteins are located on the surface of E. histolytica [28] however no classic membrane-targeting signal is present in the proteins. Therefore, it is tempting to speculate that the conserved N-terminal domain of these proteins might function as either an export signal or serve as a membrane-anchor domain or that export involves a non-classical transport mechanism, independent of the ER–Golgi pathway, similar to those that have been detected in yeast and mammalian cells [32]. Details on the motifs and domain structure are shown in Figure S2.
A third worthy of note family associated with TEs is the AIG family of proteins, comprising 29 members distributed in 3 clusters, of which 18 genes are in close proximity to repetitive elements (Table 2). AIG1 proteins are associated with resistance to bacteria [33]. Interestingly, comparative gene expression studies have shown that AIG1 proteins as well as heat shock proteins have significantly reduced expression levels in E. dispar [34], when compared to E. histolytica. This observation leads us to speculate that transposable elements inserted in the neighborhood of these genes could lead to the enhanced expression of these genes and ultimately could be related to the increased virulence. Indeed we have previously shown that LINEs and SINEs are involved in genome rearrangements driving in consequence genomic evolution [10]. It is tempting to speculate that the amplification of the AIG family was mediated by the close association of TEs, but the observation that non-virulent E. dispar contains the same number of genes without the TE association seems to indicate that this is not the case. We are currently analyzing all gene family/transposable element associations in the context of comparative genomics with other Entamoeba species (manuscript in preparation).
Close examination of the functional annotation of protein families and singleton proteins revealed that a total of 2,981 (65%) genes within the families were annotated as encoding proteins with putative functions and 1,583 genes are hypothetical proteins (34%, Fig. 3B). Of a total of 1,088 genes that have EST support in the whole genome, 705 are genes within protein families. In contrast, singletons had a larger proportion of hypothetical genes (76%) and a smaller portion of genes with a known or putative function (24%), and half the number of genes supported by EST evidence (383).
As mentioned above, about 20% of the E. histolytica genome consists of transposable elements. These repeats show a tendency to insert close to each other forming large TE clusters. We have previously shown that these repeat clusters are frequently found at syntenic breakpoints between E. histolytica and E. dispar suggesting that they could contribute to parasite genome instability and, consequently, to the evolution of these species [10]. It is also possible that the highly repetitive nature of this genome led to genome duplications. In order to evaluate this possibility we analyzed the presence of additional rearrangements within the genome by searching for segmental duplications using DAGchainer as explained in Methods [20]. We observed the presence of four different types of segmental duplications, named D1-D4, spanning seven to ten genes each (Fig. 4).
The first duplication (D1, Fig. 4A) spans a 16.6 kb region containing up to 8 hypothetical protein coding genes. These duplications are approximately 94% identical at the nucleotide level. All D1-type duplications are flanked by 2.3 kb inverted repeats (IR) not found in the rest of the genome. Nucleotide composition analysis revealed that D1-IRs are highly AT-rich (84.3%) compared to the average content of those regions 71.4% and they are 95% identical at the nucleotide level. A genome wide survey of D1-duplications led to the identification of four complete and two partial copies of this element in the genome. It is interesting to mention that all the scaffolds containing the four complete duplications have similar sizes (16.6 kb on average) and are spanned almost in their entire length by their respective segmental duplications. The two partial D1-duplications are located in shorter scaffolds of 14.4 kb and 6.6 kb, respectively.
The second duplication (D2, Fig. 4B) is 12.5 kb long and contains up to eight duplicated hypothetical protein coding genes depending on the duplication. Comparative analysis showed that these duplications are more than 80% identical at the nucleotide level with an average of 92%. Similar to D1-type duplications, D2 are frequently flanked by 1.2 kb IRs, composed of two fragments derived from the TEs EhERE1 and EhLINE2. D2-IRs share 92.6% identity at the nucleotide level and are also very AT-rich (85%AT). The organization of the duplications is not conserved in all copies across the genome, with some copies flanked by IRs composed of either EhERE1 or EhLINE2 fragments, while in others we could not identify any IR.
D3-type duplications are 7.4 kb long and 83% identical at the nucleotide level. Although frequently found nearby TEs (mostly EhLINE1), none of the eight identified genome duplications are flanked by IRs as D1- and D2-type duplications. D3 presents a very unique gene content that suggest that the segment could present a unique functionality, represented in Fig. 4C. A total of seven protein coding genes are arranged in the same orientation, and include a putative serine-threonine kinase similar to ARK1, a human protein that participates in cell cycle regulation; an endonuclease V domain-containing protein coding gene potentially involved in DNA repair; a putative secreted hypothetical protein coding gene; a tandem duplicated gene coding for a putative protein containing a type-1 glutamine amido transferase-like domain and a GDSL-like lipase/acylhydrolase domain-containing protein coding gene. Interestingly, D3-type duplications are found at or in close proximity to the end of scaffolds, and therefore, they could potentially be located at subtelomeric regions. However, in spite of a thorough analysis we could not identify any repetitive telomeric/subtelomeric motif in these regions.
Lastly, the 10 kb long D4 (Fig. 4D) shares more than 85% identity at the nucleotide level and spans up to 9 hypothetical protein coding and one putative dUTP hydrolase-coding genes. Most D4-type duplications have TEs inserted nearby, but no flanking IRs were identified.
The presence of these duplications is not likely to be an artifact of the assembly due to the fact that they are also appear duplicated in E. dispar. It is possible that some of these duplications, that in some cases span full scaffolds represent different copies of one of the several extrachromosomal elements known to exist in Entamoeba species, as described by Dhar et al [35].
Our work has led to the identification of 460 novel putative protein coding genes not present in the OGA, 16% of which have some functional annotation. One of these genes codes for a putative pantothenate kinase (EHI_183060) the first enzyme in the biosynthesis of coenzyme A from pantothenate. Although the coding genomic region was present in the original assembly, the gene had not been predicted and therefore, it was missing from the previous annotation. Only the enzymes phophopantothenoyl-cysteine decarboxilase (EC 4.1.1.36), phosphopantothenoyl-cysteine synthase (EC 6.3.2.5), and dephospho-CoA kinase (EC 2.7.1.24), responsible for the second, third and last of the five consecutive enzymatic reactions, had been previously identified in the OGA (EHI_164490, EHI_092330, EHI_040840). However, the lack of candidate enzymes for the remaining two biochemical reactions of this pathway raised the question whether E. histolytica can synthesize coenzyme A from pantothenate [36]. Our de novo gene prediction for a putative pantothenate kinase plus the identification of a candidate gene for the forth step of this pathway, a putative pantetheine-phosphate adenylyltransferase (EC 2.7.7.3), indicates that the whole set of metabolic reactions required to synthesize coenzyme A from pantothenate is present in this amoeba. Interestingly, the enzymes that participate in this pathway resemble those from eubacteria but not higher eukaryotes. Indeed, the second and third sets of reactions are catalyzed by a single enzyme present in two copies (EHI_164490, EHI_092330), while the fourth and fifth steps are carried out by independent enzymes, EHI_006680 and EHI_040840, respectively. In higher eukaryotes the last two reactions are carried out by the same enzyme [37].
Another gene not present in the OGA (EHI_141410) codes for a protein with a predicted molecular weight of 44.6 kDa similar to subunit p50 of the DNA polymerase delta, a key enzyme for chromosomal DNA replication in higher eukaryotes. In mammals, it has been shown that p50 is tightly associated with p125, the catalytic subunit of these types of DNA polymerases. Accordingly, a gene coding for a putative 124.4 kDa catalytic subunit of the DNA polymerase delta (EHI_006690), is also present in the NGA. These results are in agreement with a previous work showing that the sensitivity to different inhibitors of the DNA polymerase activity of E. histolytica resembles that of mammalian DNA alpha, delta and epsilon polymerases [38].
In addition, a gene coding for a protein containing a Yos1-like Pfam domain is also absent from OGA (EHI_178640). This putative protein has similarity to other members of the Yos1 family, involved in protein transport between the endoplasmic reticulum and the Golgi apparatus [39].
Comparative analysis between the two annotation datasets also allowed us to identify genes present in their complete form in NGA but truncated in OGA. Example of these genes are two copies of a gene coding for a putative pyridine nucleotide transhydrogenase, EHI_055400 and EHI_014030, the latter identical to a gene previously cloned by Clark et al. [40], which exists as a single truncated copy in the OGA. Another example is a 605 bp gene coding for a putative phospholipid methyltransferase protein (EHI_153710) similar to Schizosaccharomyces pombe cho1 (35% identity; e-value = 4×10−21), an enzyme that participates in the synthesis of phosphatidylcholine via the methylation of phosphatidylethanolamine. A coding sequence containing only the last 222 bp of this gene is present in the OGA.
Our reannotation effort has focused mostly on the improvement of the assembly and the gene content and structure of the E. histolytica genome. The new assembly, annotation and analysis of the genome has incorporated many updates and enhancements to the structural and functional assignments of the original gene predictions, including an improved assembly, removal of spurious genes, improved gene structures and functional assignments, and generation of gene families.
Regardless of the advancement of the computational methods and of the exponentially growing amount of data that could be used for automated genome annotation, only experimental evidence from expression data will conclusively validate the accuracy of computationally assigned functions done at the genome-wide level. Nevertheless, in order to provide a sound bases to drive research, genome annotations have to be maintained and revised, either by expert annotators in the field and/or community involvement. Additional sequence information will allow the further refinement of gene structures and a deeper understanding of the genome architecture, while the functional annotation will be enriched both by the availability of new experimental data and from expression and other kinds of analyses to characterize each gene and its function fully.
This reannotation effort will be the base for the future analysis and annotation of new E. histolytica genomes from patient isolates, a project recently approved under the NIAID supported program Genome Sequence Centers for Infectious Disease, GSCID (http://gsc.jcvi.org/).
|
10.1371/journal.ppat.1002402 | Follicular Dendritic Cell-Specific Prion Protein (PrPc) Expression Alone Is Sufficient to Sustain Prion Infection in the Spleen | Prion diseases are characterised by the accumulation of PrPSc, an abnormally folded isoform of the cellular prion protein (PrPC), in affected tissues. Following peripheral exposure high levels of prion-specific PrPSc accumulate first upon follicular dendritic cells (FDC) in lymphoid tissues before spreading to the CNS. Expression of PrPC is mandatory for cells to sustain prion infection and FDC appear to express high levels. However, whether FDC actively replicate prions or simply acquire them from other infected cells is uncertain. In the attempts to-date to establish the role of FDC in prion pathogenesis it was not possible to dissociate the Prnp expression of FDC from that of the nervous system and all other non-haematopoietic lineages. This is important as FDC may simply acquire prions after synthesis by other infected cells. To establish the role of FDC in prion pathogenesis transgenic mice were created in which PrPC expression was specifically “switched on” or “off” only on FDC. We show that PrPC-expression only on FDC is sufficient to sustain prion replication in the spleen. Furthermore, prion replication is blocked in the spleen when PrPC-expression is specifically ablated only on FDC. These data definitively demonstrate that FDC are the essential sites of prion replication in lymphoid tissues. The demonstration that Prnp-ablation only on FDC blocked splenic prion accumulation without apparent consequences for FDC status represents a novel opportunity to prevent neuroinvasion by modulation of PrPC expression on FDC.
| Prion diseases are infectious neurological disorders and are considered to be caused by an abnormally folded infectious protein termed PrPSc. Soon after infection prions accumulate first upon follicular dendritic cells (FDC) in lymphoid tissues before spreading to the brain where they cause damage to nerve cells. Cells must express the normal cellular prion protein PrPC to become infected with prions. However, whether FDC are infected with prions or simply acquire them from other infected cells is unknown. To establish the role of FDC in prion disease PrPC expression was specifically “switched on” or “off” only on FDC. We show that PrPC-expressing FDC alone are sufficient to sustain prion replication in the spleen. Furthermore, prion replication is blocked in the spleen when PrPC-expression is switched off only on FDC. These data definitively demonstrate that FDC are the essential sites of prion replication in lymphoid tissues.
| Prion diseases (Transmissible spongiform encephalopathies; TSE) are sub-acute neurodegenerative diseases that affect both humans and animals. Many prion diseases, including natural sheep scrapie, bovine spongiform encephalopathy, chronic wasting disease in mule deer and elk, and kuru and variant Creutzfeldt-Jakob disease in humans, are acquired by peripheral exposure (eg: orally or via lesions to skin or mucous membranes). After peripheral exposure prions accumulate first upon follicular dendritic cells (FDC) as they make their journey from the site of infection to the CNS (a process termed, neuroinvasion) [1]–[7]. FDC are a unique subset of stromal cells resident within the primary B cell follicles and germinal centres of lymphoid tissues [8]. Prion accumulation upon FDC is critical for efficient disease pathogenesis as in their absence neuroinvasion are impaired [1]–[4]. From the lymphoid tissues prions invade the CNS via the peripheral nervous system [9].
During prion disease aggregations of PrPSc, an abnormally folded isoform of the cellular prion protein (PrPC) accumulate in affected tissues. Prion infectivity co-purifies with PrPSc [10] and is considered to constitute the major, if not sole, component of infectious agent [11]. Host cells must express cellular PrPC to sustain prion infection [12] and FDC appear to express high levels of PrPC on the cell membrane in uninfected mice [13], [14]. Although prion neuroinvasion from peripheral sites of exposure is dependent upon the presence of FDC in lymphoid tissues, it is not known whether FDC actually replicate prions themselves. FDC characteristically trap and retain native antigen on their surfaces for long periods in the form of immune complexes, consisting of antigen-antibody and/or complement components. Prions are also considered to be acquired by FDC as complement-opsonized immune complexes [15]–[18]. Thus, during prion infection FDC might simply trap and retain PrPSc-containing immune complexes on their surfaces following synthesis by other infected cells such as neurones.
Many cell types including classical DC, lymphocytes, mast cells, platelets, reticulocytes and epithelial cells secrete membrane vesicles termed exosomes that are enriched in cell-specific protein [19], [20]. Although the functions of exosomes are uncertain FDC can bind them on their surfaces. These microvesicles permit FDC to passively acquire and display proteins on their surfaces that they do not express at the mRNA level [21]. Studies have shown that prions only accumulated in the spleens of mice in which the FDC-containing stromal compartment expressed PrPC [13], [14]. However, in each of those studies it was not possible to dissociate the Prnp expression status of the FDC from that of the nervous system and all other host-derived non-haematopoietic and stromal cell populations [13], [14], [22]. This is important as prion infection can occur within inflammatory PrPC-expressing stromal cells that are distinct from FDC [23]. Furthermore, as both PrPC and PrPSc can be released from cells in association with exosomes [20] FDC may passively acquire PrPC and prions after release in exosomes from other infected cells [24], [25].
No therapies are available to treat prions diseases. A thorough characterization of the host cells that are infected by prions is imperative for the identification of candidate molecular targets for therapeutic intervention, the development of useful pre-clinical diagnostics and to aid our understanding of the risk of transmission. To definitively determine the role of FDC in prion pathogenesis, two unique compound transgenic mouse models were created in which PrPC expression was specifically “switched on” or “switched off” only on FDC. These mice were then used to establish: i) whether FDC express PrPC or simply acquire it from other host cells; and ii) whether FDC amplify prions, or simply acquire them from other infected host cells. Our data clearly show that PrPC-expressing FDC alone are sufficient to sustain prion replication in the spleen. Furthermore, prion replication in the spleen is blocked in mice in which PrPC-expression is specifically ablated only on FDC.
To study FDC-specific gene function transgenic mice were used that expressed Cre recombinase under the control of the Cr2 locus (CD21-Cre mice) which directs expression in FDC and mature B cells [26], [27]. First the cellular specificity of the Cre recombination was assessed by crossing the CD21-Cre mice with the ROSA26flox/flox reporter strain [28]. Histological analysis showed efficient LacZ expression indicative of Cre-mediated gene recombination in FDC and B cell follicles in the spleens, lymph nodes and Peyer's patches of CD21-Cre ROSA26flox/flox mice (Figure 1A, B). No recombination was observed in FDC and mature B cells in the spleens of ROSA26flox/flox reporter mice that lacked Cre expression (Figure 1B). Unlike lymphocytes, FDC do not derive from bone marrow precursors [29]. As a consequence, it is possible to mix-and-match the genotype of FDC and lymphocytes by grafting bone marrow cells from donor mice into recipients of a different genetic background [13], [14], [22]. To restrict Cre-expression to FDC, adult CD21-Cre ROSA26flox/flox mice were lethally γ-irradiated and 24 h later reconstituted with bone marrow from Cre-deficient C57BL/6 wild-type (WT) mice (termed WT→CD21-Cre ROSA26flox/flox mice) and tissues from six mice from each group analysed 100 days after transfusion. Using this approach, in these mice all B cells lack Cre-expression as they derive from the WT donor bone marrow, whereas the FDC express Cre as they are host-derived. Analysis of the cellular sites of LacZ expression in WT→CD21-Cre ROSA26flox/flox mice confirmed that Cre-mediated recombination was associated with FDC (Figure 1B). No other cellular sites of Cre-mediated recombination were observed in the spleens of WT→CD21-Cre ROSA26flox/flox mice. Furthermore, no other cellular sites of Cre-mediated recombination were observed in a wide range of non-lymphoid peripheral tissues from CD21-Cre ROSA26flox/flox and WT→CD21-Cre ROSA26flox/flox (heart, liver, kidney, pancreas, ear, tongue, skeletal, muscle, ovary, uterus, bladder, testes, epididymis, sciatic nerve and spinal cord; data not shown). These data clearly demonstrate that CD21-Cre mice are a useful tool to study FDC-specific gene expression and function.
Cre toxicity can occur in some Cre transgenic mouse lines whereby Cre recombinase causes mis-recombination, DNA damage and death of Cre-expressing cells [30]. However, immunohistochemical (IHC) analysis of spleens from CD21-Cre ROSA26flox/flox mice and WT→CD21-Cre ROSA26flox/flox mice showed no significant effect of Cre-expression on the status of FDC networks and B cell follicles when compared to spleens from WT control mice and ROSA26flox/flox mice that lacked Cre expression (Figure 1C). Furthermore, the expression of Cre recombinase under the control of the Cr2 locus had no observable effect on CD21/35 expression (Figure 1C).
Next, mice were created in which Prnp expression (which encodes PrPC) was restricted only to FDC. To do so, CD21-Cre mice were first bred onto a PrPC-deficient (Prnp-/-) background. The resulting CD21-Cre Prnp-/- mice were then crossed with Prnpstop/- mice in which a floxed β-geo stop cassette was inserted into intron 2 of the Prnp gene upstream of exon 3 [31]. In the progeny CD21-Cre Prnpstop/- mice, PrPC is only expressed in cells expressing Cre recombinase (CD21-expressing FDC and mature B cells). To restrict the Prnp-expression to FDC, CD21-Cre Prnpstop/- mice were lethally γ-irradiated and grafted with bone marrow from Cre-deficient Prnpstop/- mice (Prnpstop/-→CD21-Cre Prnpstop/- mice). We also performed bone marrow transfers from CD21-Cre Prnpstop/- donors into CD21-Cre Prnpstop/- recipients (CD21-Cre Prnpstop/-→CD21-Cre Prnpstop/- mice), CD21-Cre Prnpstop/- donors into Cre-deficient Prnpstop/- mice (CD21-Cre Prnpstop/-→Prnpstop/- mice) and Prnp+/- donors into Prnp+/- recipients (Prnp+/-→Prnp+/- mice) as controls (Figure 2A).
Spleens, tails and blood from six mice from each group were examined 100 days after bone marrow transfusion. PCR analysis of DNA isolated from the tails, blood and spleens of mice in each group was used to confirm the presence of Cre (Figure 2B, upper panel) and Cre-mediated DNA recombination (Figure 2B, lower panel) within the stromal, haematopoietic or both compartments (respectively). The detection of Cre in the tail and spleen but not blood of the Prnpstop/-→CD21-Cre Prnpstop/- mice confirmed the restriction of the Cre-expression to the stromal but not haematopoietic compartments of these mice. In addition, PCR analysis also confirmed that in these mice efficient Cre-mediated recombination of the Prnpstop/- allele was restricted to the FDC-containing stromal compartment of the spleen (Figure 2B). In Prnpstop/-→CD21-Cre Prnpstop/- mice the recombined Prnpstop/- allele (Prnpstop(R)) was detected in the spleen, but not blood and tail. Thus these data indicate that in the spleens of Prnpstop/-→CD21-Cre Prnpstop/- mice Cre-mediated recombination is restricted to FDC and not B cells.
As anticipated, in the spleens of Prnp+/-→Prnp+/- control mice high levels of PrPC expression were observed upon FDC and tyrosine hydroxylase (TH)-positive sympathetic nerves (Figure 2C). In contrast, in the spleens of Prnpstop/-→CD21-Cre Prnpstop/- mice PrPC was only expressed on FDC (Figure 2C). In the absence of Cre-recombinase expression by FDC and peripheral nerves in CD21-Cre Prnpstop/-→Prnpstop/- mice, PrPC expression was not expressed by either cell population (Figure 2C).
Morphometric analysis confirmed that the amount of the PrPC expression co-localized upon the surfaces of FDC in the spleens of Prnpstop/-→CD21-Cre Prnpstop/- mice was not significantly different from that observed upon FDC in spleens from Prnp+/-→Prnp+/- control mice (P<0.69, n = 48 FDC/group; Figure 2D). In contrast, in the absence of Cre-recombinase expression by FDC in CD21-Cre Prnpstop/-→Prnpstop/- mice, PrPC expression was substantially lower than that observed upon FDC in spleens from Prnp+/-→Prnp+/- control mice (P<1×10-25, n = 48; Figure 2D). Morphometric analysis also confirmed that PrPC expression upon the surfaces of sympathetic nerves in the spleens of Prnpstop/-→CD21-Cre Prnpstop/-, CD21-Cre Prnpstop/-→ CD21-Cre Prnpstop/- and CD21-Cre Prnpstop/-→Prnpstop/- mice was significantly ablated when compared to that observed upon sympathetic nerves in spleens from Prnp+/-→Prnp+/- control mice (p<1×10−25, n = 48 sympathetic nerves/group). Together, these data confirm that in the spleens of Prnpstop/-→CD21-Cre Prnpstop/- mice PrPC expression is specifically restricted to FDC, whereas in spleens from CD21-Cre Prnpstop/-→Prnpstop/- mice, FDC lack PrPC expression. FDC can passively acquire the expression of some surface molecules including MHC class II and complement component C4 [21], [32]. However, these data confirm that FDC express high levels of cellular PrPC on their surfaces and do not simply acquire it from neighbouring cells.
IHC analysis confirmed that the microarchitecture (Figure 3A), size (P = 0.755, n = 32; Figure 3B) and number (P = 0.249, n = 32; Figure 3C) of the FDC networks in spleens from mice with Prnp-expression restricted to FDC (Prnpstop/-→CD21-Cre Prnpstop/- mice) were normal when compared to control mice. Other studies have shown that the density of sympathetic nerves can significantly influence the amount of prion accumulation within in the spleen [33]. Quantitative analysis of the relative positioning of FDC and sympathetic nerves showed there were no significant differences in average distance between these cell populations in spleens from each mouse group (Figure 3D & E; P<0.932, n = 48).
Next, we determined the effect of FDC-restricted Prnp-expression on prion replication in the spleen. In this study, the normal cellular form of the prion protein is referred to as PrPC, and two distinct terms (PrPSc or PrPd) are used to describe the disease-specific, abnormal accumulations of PrP that are characteristically found only in prion-affected tissues and considered a reliable biochemical marker for the presence of infectious prions [10]. Disease-specific PrP (PrPd) accumulations are relatively resistant to proteinase K (PK) digestion, whereas cellular PrPC is destroyed. Where we were able to confirm this resistance by treatment of samples with PK and subsequent paraffin-embedded tissue (PET) immunoblot analysis [34], PrPSc is used as a biochemical marker for the presence of prions. Unfortunately, treatment of tissue sections with PK destroys the microarchitecture. Therefore, for IHC analysis tissue sections were fixed and pre-treated to enhance the detection of the disease-specific abnormal accumulations of PrP (PrPd), whereas cellular PrPC is denatured by these treatments [4]. We have repeatedly shown in a series of studies that these PrPd-accumulations occur only in prion-infected tissues, and correlate closely with the presence of ME7 scrapie prions [1], [4], [13], [35]–[37].
Within weeks after i.p. exposure of WT mice to ME7 scrapie prions, strong accumulations of prion-specific PrPSc occur upon FDCs within the spleen and are sustained until the terminal stages of disease [1], [13], [35]. Here, mice were injected i.p. with ME7 scrapie prions and spleens from 4 mice from each group collected 35, 70 and 105 days after exposure. In spleens from control mice (Prnp+/-→Prnp+/- mice) heavy PrPd accumulations, consistent with localisation upon FDC, were detected at 70 days after i.p. injection with the scrapie agent and had increased in intensity by 105 days after infection (Figure 4A & B). PET immunoblot confirmed the presence of PrPSc upon the surfaces of the FDC in spleens from control mice (Figure 4C). Furthermore, in the spleens of Prnpstop/-→CD21-Cre Prnpstop/- mice in which cellular PrPC was expressed only on FDC, heavy PrPSc accumulations were likewise maintained upon FDC (Figure 4A & B). In contrast, in the absence of PrPC expression by FDC in the spleens of CD21-Cre Prnpstop/-→Prnpstop/- mice, no PrPSc accumulations were observed upon FDC. In the spleens of mice with PrPC-deficient FDC, if PrP was detected at all, it was only occasionally observed within tingible body macrophages (Figure 4A and B, arrowheads; Figure S1). We also analysed prion infectivity levels in spleens collected 70 days after infection from control mice (Prnp+/-→Prnp+/- mice) and Prnpstop/-→CD21-Cre Prnpstop/- mice in which cellular PrPC was expressed only on FDC (Figure S2; n = 3/group). As anticipated high levels of prion infectivity were observed in each control spleen. Furthermore, consistent with data above our analysis showed that PrPc expression only of FDC was sufficient to sustain high levels of prion infectivity within the spleen (Figure S2). These data demonstrate that PrPC expression only on FDC is sufficient to sustain prion replication in the spleen. In the absence of PrPC expression on FDC the prions appeared to be scavenged by tingible body macrophages resident within the B cell follicles.
Next, mice were created in which Prnp expression was specifically ablated in FDC. To do so, CD21-Cre Prnp-/- mice were crossed with mice carrying a “floxed” Prnp gene (Prnpflox/flox mice; [31]). In the progeny CD21-Cre Prnpflox/- mice, Prnp expression is conditionally ablated in cells expressing Cre recombinase (CD21-expressing FDC and mature B cells). To restrict the Prnp-ablation to FDC, CD21-Cre Prnpflox/- mice were lethally γ-irradiated and grafted with bone marrow from Cre-deficient Prnpflox/- mice (Prnpflox/-→CD21-Cre Prnpflox/- mice). We also performed bone marrow transfers from CD21-Cre Prnpflox/- donors into CD21-Cre Prnpflox/- recipients (CD21-Cre Prnpflox/- mice→CD21-Cre Prnpflox/- mice), CD21-Cre Prnpflox/- donors into Cre-deficient Prnpflox/- mice (CD21-Cre Prnpflox/-→Prnpflox/- mice), and Prnp+/- donors into Prnp+/- recipients (Prnp+/-→Prnp+/- mice) as controls (Figure 5A). Spleens, tails and blood from 6 mice from each group were examined 100 days after bone marrow transfusion. PCR analysis of DNA isolated from the spleens, blood and tails of Prnpflox/-→CD21-Cre Prnpflox/- mice confirmed that efficient Cre-mediated DNA recombination and Prnp-ablation was restricted to the FDC-containing stromal compartment of the spleen (Figure 5B). In Prnpflox/-→CD21-Cre Prnpflox/- mice the recombined Prnpstop/- allele (Prnpdeflox) was detected in the spleen, but not blood and tail. Thus these data indicate that in the spleens of Prnpstop/-→CD21-Cre Prnpstop/- mice Cre-mediated recombination and Prnp-ablation is restricted to FDC and not B cells.
IHC analysis showed that in the spleens of Prnpflox/-→CD21-Cre Prnpflox/- mice and CD21-Cre Prnpflox/- mice→CD21-Cre Prnpflox/- mice FDC did not express PrPC whereas high levels were associated with TH-positive sympathetic nerves (Figure 5C). In the absence of Cre-recombinase expression by FDC in CD21-Cre Prnpflox/-→Prnpflox/- mice, high levels of PrPC were expressed by FDC and sympathetic nerves (Figure 5C).
Morphometric analysis confirmed that the magnitude of the PrPC expression co-localized upon the surfaces of FDC in the spleens of Prnpflox/-→CD21-Cre Prnpflox/- mice and CD21-Cre Prnpflox/- mice→CD21-Cre Prnpflox/- mice was substantially lower than that observed upon FDC in spleens from Prnp+/-→Prnp+/- control mice (P<1×10-24 and P<1×10−23, respectively, n = 48 FDC/group) and not significantly different when compared to background levels (Figure 5D). In contrast, in the absence of Cre-recombinase expression by FDC in CD21-Cre Prnpflox/-→Prnpflox/- mice, PrPC expression was not significantly different from the level observed upon FDC in spleens from Prnp+/-→Prnp+/- control mice (P<0.106; Figure 5D). In contrast, morphometric analysis showed that the magnitude of the PrPC expression co-localized upon the surfaces sympathetic nerves in the spleens of Prnpflox/-→CD21-Cre Prnpflox/-, CD21-Cre Prnpflox/-→ CD21-Cre Prnpflox/- and CD21-Cre Prnpflox/-→Prnpflox/- mice was similar to that observed upon sympathetic nerves in spleens from Prnp+/-→Prnp+/- control mice (p = 0.400, n = 48 sympathetic nerves/group). Together, these data confirm that in the spleens of Prnpflox/-→CD21-Cre Prnpflox/- mice the Prnp ablation is specifically restricted to FDC.
Data in the current study definitively demonstrate that FDC express high levels of PrPC but the role PrPC plays in FDC function and homeostasis is not known. IHC analysis showed that the microarchitecture of the FDC networks from Prnp-ablated Prnpflox/-→CD21-Cre Prnpflox/- mice were normal when compared to control mice (Figure 6A). Furthermore, no significant difference was observed in the size (P = 0.750, n = 32) and number (P = 0.713, n = 32 of the FDC networks in spleens from each mouse group (Figure 6B & C, respectively). The relative positioning of the FDC and sympathetic nerves was likewise similar in spleens from each mouse group (Figure 6D & E; P<0.765, n = 48).
FDC characteristically trap and retain native antigen on their surfaces in the form of immune complexes, consisting of antigen-antibody and/or complement components. Antigens trapped on the surface of FDC are considered to promote immunoglobulin-isotype class switching, affinity maturation of naïve IgM+ B cells and the maintenance of immunological memory [38]–[42]. Indeed, prions are also considered to be acquired by FDC as complement-opsonized immune complexes [15]–[18]. To determine whether antigen retention by Prnp-ablated FDC was affected six mice from each group were passively immunized with preformed PAP immune complexes, and 24 h later, the presence of FDC-associated immune complexes identified by IHC (Figure 7) and the presence of peroxidase activity (data not shown). No significant difference in the magnitude of immune complex trapping could be detected between FDC from Prnp-ablated Prnpflox/-→CD21-Cre Prnpflox/- mice and control mice (Figure 7; P = 0.85, n = 40/group). Together, these data demonstrate that Prnp-ablation does not impair FDC status or their ability to trap and retain immune complexes.
Next, the effect of FDC-specific Prnp-ablation on prion replication by FDC was determined. Mice were injected i.p. with ME7 scrapie prions and spleens from 4 mice from each group collected 70 days after exposure. As anticipated, heavy PrPd (Figure 8A and B) and PrPSc (Figure 8C) accumulations consistent with localisation upon FDC were detected in spleens from control mice (Prnp+/-→Prnp+/- mice) and mice in which Prnp was ablated only in mature B cells (CD21-Cre Prnpflox/-→Prnpflox/- mice). In the spleens in which cellular PrPC was ablated only on FDC (Prnpflox/-→CD21-Cre Prnpflox/- mice), or FDC and mature B cells (CD21-Cre Prnpflox/-→CD21-Cre Prnpflox/- mice), no PrP accumulations were observed upon FDC (Figure 8A–C). Consistent with data above (Figure 4), in spleens of mice with PrPC-deficient FDC PrP accumulations were only occasionally observed within tingible body macrophages (Figure 8A and B, arrowheads; Figure S1). We also analysed prion infectivity levels in spleens from Prnpflox/-→CD21-Cre Prnpflox/- mice in which cellular PrPC expression was ablated only on FDC (Figure S2; n = 3). Consistent with data above this analysis showed that in the absence of PrPc expression only on FDC the accumulation of high levels of prion infectivity in the spleen was blocked (Figure S2). Taken together, these data show that in the specific absence of PrPC expression FDC are unable to sustain prion replication upon their surfaces and as a consequence the agent is scavenged by tingible body macrophages.
When mice with PrPC-ablated FDC (Prnpflox/-→CD21-Cre Prnpflox/- mice) were injected intracerebrally (i.c.) with the ME7 scrapie agent strain directly into the CNS all mice succumbed to clinical signs of scrapie approximately 300 days after exposure with incubation periods indistinguishable from those of Prnp+/- control mice [43] (Prnpflox/-→CD21-Cre Prnpflox/-, 297±4 days, n = 4; Prnp+/-, 290±4 days, n = 5; P = 0.386). Histopathological analysis showed that brains from all clinically-affected mice from each group displayed the characteristic spongiform pathology, astrogliosis, microgliosis and PrPd accumulation typically associated with terminal infection with the ME7 scrapie agent (Figure 9A). Following i.c.-injection with the ME7 scrapie agent, high levels of PrPSc accumulate upon FDC and are maintained for the duration of the incubation period [13](Figure 9B and C). However, FDC are not critical for ME7 scrapie pathogenesis when infection is established directly within the CNS [4], [13], [35], [44]. In the spleens from clinically-scrapie affected mice in which PrPC expression was specifically ablated only on FDC (Prnpflox/-→CD21-Cre Prnpflox/- mice), PrPSc replication upon FDC was also blocked. These data how that FDC do not simply trap and retain prions after their release from infected neurones in the CNS. These data also confirm that the Prnp-ablation in Prnpflox/-→CD21-Cre Prnpflox/- mice was specific to FDC and had no effect on prion neuropathogenesis and disease susceptibility when the infection was established directly in the CNS.
Studies in mice show that efficient prion neuroinvasion from peripheral sites of exposure is dependent upon the presence of FDC in lymphoid tissues [1], [4], [35], [44]–[46]. Next, the effect of FDC-specific Prnp-ablation on prion neuroinvasion via the peritoneal route was determined. Unfortunately, due to the advanced ages of the mice in this experiment, some succumbed to ageing-related inter-current illness. As there was a 100 days interval between the time of lethal γ-irradiation/bone marrow reconstitution and prion infection, many mice were approximately 500–600 days old when culled. However, most mice with PrPC-expressing FDC in their spleens succumbed to clinical prion disease after i.p. injection (Prnp+/-→ Prnp+/- control mice, n = 5/7; CD21-Cre Prnpflox/-→Prnpflox/- mice, n = 3/6; Table S1). Histopathological analysis showed that brains from all clinically-affected mice from these groups displayed the characteristic spongiform pathology, astrogliosis, microgliosis and PrPd accumulation typically associated with terminal infection with the ME7 scrapie agent (Figure S3, third and fourth columns). In contrast, none of the mice with PrPC-ablated FDC (Prnpflox/-→CD21-Cre Prnpflox/- mice, n = 0/6; CD21-Cre Prnpflox/-→CD21-Cre Prnpflox/- mice, n = 0/7) succumbed to clinical prion disease during their life-spans (Table S1). Although we cannot exclude the possibility that if the clinically-negative mice with PrPC-ablated FDC mice had lived longer some may have succumbed to clinical prion disease after substantially extended incubation periods, no PrPd or other characteristic histopathological hallmarks of prion disease were detected in their brains (Figure S3, first two columns). Together, these data suggest that in the specific absence of PrPC expression on FDC neuroinvasion following peripheral exposure is impaired.
These data definitively demonstrate that FDC are essential sites of prion replication in lymphoid tissues. In order to precisely establish the role of FDC in prion pathogenesis two unique compound transgenic mouse models were created in which PrPC expression was specifically “switched on” or “off” only on FDC. Our data confirm that FDC express high levels of PrPC and do not simply acquire it from other host cells. Furthermore, we show that following peripheral exposure PrPC-expressing FDC alone are sufficient to sustain high levels of prion replication in the spleen. Accordingly, when PrPC-expression was specifically ablated only on FDC prion replication in the spleen was blocked. These data likewise demonstrate that FDC do not simply acquire prions after their release from other infected host cells. Our analysis showed that the effects of Prnp-ablation on prion replication in the spleen were specific to FDC and had no effect on prion neuropathogenesis when the infection was established directly in the CNS. In the absence of PrPC expression on FDC the PrPSc from the initial inoculum appeared to be scavenged by tingible body macrophages resident within the B cell follicles. Together, these data definitively demonstrate that FDC are the critical early sites of prion replication in lymphoid tissues. This study is the first to demonstrate that the specific ablation of a cellular protein only on FDC, without apparent consequences for FDC status and function, blocks the replication of an important pathogen in the spleen.
FDC reside in the primary B cell follicles and germinal centres of lymphoid tissues and are a completely distinct cell lineage from bone-marrow-derived classical dendritic cells [47]–[49]. FDC possess many slender and convoluted dendritic processes which provide the FDC with an extremely large surface area. This helps the FDC to efficiently trap and retain large amounts of native antigen in the form of immune complexes, consisting of antigen-antibody and/or complement components. The longevity of FDC ensures that antigen is retained upon their surfaces for long periods [50], [51]. Antigens trapped on the surface of FDC are considered to promote immunoglobulin-isotype class switching, affinity maturation of naïve IgM+ B cells and the maintenance of immunological memory [38]–[42]. FDC are also considered to aid the clearance of apoptotic B lymphocytes [52], and play a role in infection with human immunodeficiency virus [53] and the pathogenesis of chronic inflammatory and autoimmune diseases [54] and peripherally-acquired prion infections.
A number of studies have addressed the role of FDC in prion pathogenesis. They show that prion replication in the spleen and subsequent neuroinvasion are both impaired in immunodeficient mice that lack FDC [4], [44], [45], or following their temporary de-differentiation [1], [35], [46]. Although the precise identity of FDC precursor cells is unknown, other studies have exploited their non-haematopoietic-origin to address their role in prion pathogenesis. In these bone marrow chimera studies, mismatches were created in Prnp expression between the FDC-containing stromal and haematopoietic compartments by grafting bone marrow cells from PrP-deficient (Prnp-/-) mice into PrP-expressing wild-type mice, and vice versa [13], [14]. Using this approach FDC and all other stromal cells were derived from the recipient, whereas lymphocytes and other haematopoietic lineages were derived from the donor cells. Following peripheral exposure prion accumulation upon FDC was only detected in the spleens of mice with a Prnp-expressing stromal compartment.
While the above studies clearly show that the presence of FDC is important for prion replication in the spleen, it was not possible to dissociate the Prnp expression status of FDC from that of the nervous system and all other non-haematopoietic host-cell populations and therefore precisely characterise the role of FDC in prion neuroinvasion [13], [14]. This is important for a number of reasons. Firstly, prion infection can occur within inflammatory PrPC-expressing stromal cells that are distinct from FDC [23]. Secondly, the FDC's ability to bind exosomes may have lead to the wrong interpretation to be made in earlier studies describing their ontology [55]. Most evidence indicates that FDC do not derive from haematopoietic precursors [29], [49]. However, the detection of donor bone marrow derived MHC class-I molecules, and other donor-derived antigens, on the surface of FDC in recipient mice was considered evidence of FDC precursor cells within bone marrow [55]. With hindsight these observations are most likely due to the FDC's capacity to acquire exosome-associated antigens from other cell types [21]. Both PrPC and PrPSc can be released from cells in association with exosomes [20]. The possibility, therefore, cannot be excluded that FDC passively acquire prions after their release in exosomes from other infected non-haematopoietic cell populations. Finally, FDC characteristically trap and retain immune complexes on their surfaces. FDC express negligible levels of complement component C4 at the mRNA level but the detection of abundant activated C4 on their surfaces by IHC using mAb FDC-M2 (as used in this study) is indicative of the capture and retention of immune complexes by FDC [32]. Opsonising complement components and cellular CR are likewise considered to play an important role in the retention of prions by FDC [15], [16], [18]. Thus FDC may simply act as concentrating depots for prion-containing complement-opsonized immune complexes.
The practical hurdles that are encountered when attempting to isolate highly purified FDC from lymphoid tissues have made detailed analysis of their pathobiological functions extremely difficult. The main issues include: contamination with other cell types such as B cells and tingible body macrophages which express MFGE8 (FDC-M1), a common marker used to identify FDC [52], [56], low yield, and their dependence on constitutive lymphotoxin β receptor-stimulation to maintain their differentiated state [57]. FDC and mature B cells express high levels of Cr2 which encodes the complement receptors CR2/CR1 (CD21/35) [18], [27]. A previous study used CD21-cre mice to study FDC-specific gene function [27]. In the current study, our data confirm that Cre/loxP-mediated DNA recombination was specific to FDC and mature B cells in CD21-cre mice, and could be restricted to FDC by transfusing the mice with Cre-deficient bone marrow. In some Cre transgenic mouse lines Cre-toxicity is encountered whereby Cre recombinase causes mis-recombination, DNA damage and death of Cre-expressing cells [30]. However, our analysis suggested no significant effect of Cre-expression on the number, size and status of FDC networks and B cell follicles. CD21-Cre mice are therefore a powerful in vivo tool in which to study FDC-specific gene expression and function.
Expression of PrPC is mandatory for host cells to sustain prion infection [43]. In the current study to establish whether FDC actively amplify prions a compound transgenic mouse model was created using the CD21-cre mouse line to specifically “switch on” PrPC expression only on FDC (Prnpstop/-→CD21-Cre Prnpstop/- mice). As a consequence, only FDC in these mice had the potential to be actively infected with and replicate prions. Our analysis showed that expression of PrPC only on FDC was sufficient to sustain high levels of PrPSc accumulation upon FDC in the spleen after peripheral prion exposure. These data definitively demonstrate that FDC are the critical sites of prion replication in lymphoid tissues. Ultrastructural analysis of the cellular compartments within which PrPd localizes upon/within FDC has failed to show any intracellular accumulation. Instead the PrPd appears to be restricted to the plasmalemma of their dendritic processes [58]. This implies that early de novo PrPSc conversion occurs upon the surface of FDC.
A second compound transgenic mouse model was created in which PrPC expression was specifically “switched off” only on FDC (Prnpflox/-→CD21-Cre Prnpflox/- mice). If, as shown above, FDC do actively amplify prions, then one would also expect the specific ablation of PrPC expression only on FDC to block prion replication in the spleen. Our data confirmed this to be the case. As PrPC expression in all other host cells (eg: neurones) in these mice was unaffected, these data clearly show that FDC do not simply acquire prions following release from other infected host cells, even in mice with clinical prion disease in the brain. IHC analysis implied that in the spleens of mice with PrP-deficient FDC the prions appeared to be scavenged by tingible body macrophages resident within the B cell follicles. The lack of detection of PrPd within tingible body macrophages in the spleens of clinically-affected mice with PrP-deficient FDC (Figure 9) clearly demonstrates that these cells are not alternative sites of replication of ME7 scrapie prions. High levels of prions rapidly accumulate within the spleen and other lymphoid tissues within weeks of peripheral exposure. The magnitude of the prion accumulation within the spleen rapidly reaches a plateau level which is maintained for the duration of the disease [13], [44]. The maintenance of this plateau may be the consequence of a competitive state whereby FDC act to amplify prions above the threshold required to achieve neuroinvasion, whereas phagocytic cells such as macrophages act to destroy them [59], [60]. Indeed increased numbers of PrPd-containing tingible body macrophages are found within the B cell follicles of TSE-affected animals [58]. Thus, our data suggest that in the specific absence of PrPC expression by FDC the initial inoculum is phagocytosed and gradually degraded by mononuclear phagocytes such as tingible body macrophages [59], [60]. These data are congruent with data from our earlier study which likewise occasionally detected trace levels of prions from the initial inoculum within tingible body macrophages in the spleens of mice with a PrPC-deficient FDC-containing stromal compartment [13].
The density of sympathetic nerves can significantly influence the amount of prion accumulation in the spleen [33]. In the current study the distribution of TH-positive sympathetic nerves in the spleens of the FDC-specific gene targeted mouse lines was not adversely affected. Furthermore, when prions were injected directly to the brain, FDC-specific Prnp ablation had no influence on the onset of clinical disease or the neuropathology. These data provide strong evidence that the effects of Cre-mediated Prnp ablation on prion replication in the spleen were specific to FDC and not due to unregulated ablation of PrPC expression within the nervous system. In the current study PrPSc accumulation upon PrPC-ablated FDC (Prnpflox/-→CD21-Cre Prnpflox/- mice) was blocked even in spleens from i.c. injected clinically-scrapie affected mice. These data contrast those reported by Crozet and colleagues [61] which used Tg(OvPrP4) mice that express the ovine PRNP gene under the control of the neuron-specific enolase promoter on a murine Prnp-/- background. As a consequence ovine PrPC is expressed only in neurones. In contrast to data in the current study, when Tg(OvPrP4) mice were injected i.c. with a high dose of natural sheep scrapie PrPSc was detected in the germinal centres of their spleens. The reasons for this discrepancy are uncertain. However, the expression of PrPC in the neuronal compartment of Tg(OvPrP4) mice is 2-4X higher than in controls. In the current study in mice in which PrPC was ablated only on FDC (Prnpflox/-→CD21-Cre Prnpflox/- mice) the expression of murine Prnp in Cre-deficient cells such as neurones is controlled by the endogenous Prnp promoter and expressed at similar levels to controls (Figure 5E). In the presence of increased PrPC expression on neurones it is plausible that greater prion replication occurred within the peripheral nervous system, which may have been subsequently trapped on the surface of the FDC and scavenged by macrophages as the prion burden increased. Similarly, hyper-innervation of the spleen likewise leads to increase prion burden in this tissue [33].
In conclusion, our data demonstrate that PrPC-expressing FDC are the essential sites of prion replication in lymphoid tissues. Indeed, PrPC-expression on FDC alone was sufficient to sustain high levels of prion replication. In contrast, the specific ablation of PrPC expression on FDC blocked prion replication. Although FDC have the capacity to bind exosomes and immune complexes which may contain PrPSc, this finding clearly demonstrates that FDC do not simply passively acquire prions from other infected cell populations such as neurones. Previous data show treatments which impair the status or immune complex-trapping function of FDC reduce prion susceptibility after peripheral exposure [1], [16], [35], [46], [62]. The demonstration that Prnp-ablation only on FDC blocked splenic prion replication without apparent consequences for FDC status represents a novel opportunity to prevent neuroinvasion by modulation of PrPC expression on FDC.
All studies using experimental mice and regulatory licences were approved by both The Roslin Institute's and University of Edinburgh's Protocols and Ethics Committees. All animal experiments were carried out under the authority of a UK Home Office Project Licence within the terms and conditions of the strict regulations of the UK Home Office ‘Animals (scientific procedures) Act 1986'. Where necessary, anaesthesia appropriate for the procedure was administered, and all efforts were made to minimize harm and suffering. Mice were humanely culled using by a UK Home Office Schedule One method.
The CD21-Cre [26], ROSA26flox/flox reporter strain [28], Prnp-/- [12] mice and tga20 mice over-expressing PrPc [63] were generated as described previously. Prnpflox/flox mice have loxP sites flanking exon 3 of the Prnp gene [31]. Prnpstop/- mice have a floxed β-geo cassette inserted into intron 2 of the Prnp gene upstream of exon 3 [31]. Mice were maintained under SPF conditions.
Prior to their use in experiments, the genotype of each mouse was confirmed by PCR analysis. DNA was prepared from tails, blood and spleens using the DNeasy blood and tissue kit (Qiagen, Crawley, UK) according to the manufacturer's instructions. Where indicated DNA samples were analysed for presence of Cre, LacZ, Prnp+/+, Prnp-/-, Prnpflox, recombined Prnpflox (Prnpde-flox), Prnpstop and recombined Prnpstop (Prnpstop(R)) using the primers listed in Table 1. PCR products were resolved by electrophoresis through a 1.0% agarose gel containing 0.002% GelRed (Biotium, Cambridge Biosciences Ltd, Cambridge, UK).
Bone-marrow from the femurs and tibias of donor mice was prepared as single-cell suspensions (3×107–4×107 viable cells/ml) in HBSS (Invitrogen, Paisley, UK). Recipient adult (6–8 weeks old) mice were γ-irradiated (950 rad) and 24 h later reconstituted with 100 µl bone-marrow by injection into the tail vein. Recipient mice were used in subsequent experiments as described 100 days after bone marrow reconstitution to allow sufficient time for removal of long-lived B lymphocyte populations and their replacement from the donor bone marrow.
Tissues were first immersed in LacZ fixative [PBS (pH 7.4) containing 2% paraformaldehyde, 0.2% gluteraldehyde, 0.02% Nonidet P40, 0.01% sodium deoxycholate, 5 mM EGTA, 2 mM MgCl2] and washed in LacZ wash buffer [PBS (pH 7.4) containing 0.02% Nonidet P40, 0.01% sodium deoxycholate, 2 mM MgCL2]. Tissues were subsequently incubated in 15% (wt/vol) sucrose in PBS overnight followed by a further overnight incubation in 30% (wt/vol) sucrose in PBS and embedded in Tissue-Tek O.C.T. compound (Bayer PLC, Newbury, UK). Serial sections (thickness 8 µm) were cut on cryostat and stained overnight at 37°C with LacZ staining solution (Glycosynth, Warrington, UK). Staining reaction was stopped by washing in LacZ wash buffer followed by dH2O. Sections were counterstained with nuclear fast red (Vector Laboratories, Peterborough, UK).
For i.c. or i.p. exposure mice were injected with 20 µl of a 1% (v/w) scrapie brain homogenate prepared from mice terminally-affected with ME7 scrapie prions (containing approximately 1×104 i.c. ID50 units). Following exposure, mice were coded and assessed blindly for signs of clinical disease and culled at a standard clinical endpoint [64]. Survival times were recorded for mice that did not develop clinical signs of disease and were culled when they showed signs of intercurrent disease. Scrapie diagnosis was confirmed blindly on coded sections by histopathological assessment of vacuolation in the brain. For the construction of lesion profiles, vacuolar changes were scored in nine grey-matter areas of brain as described [65]. Where indicated, some four mice from each group were culled at the times indicated post injection with scrapie and tissues taken for further analysis. For bioassay of scrapie agent infectivity, individual half spleens were prepared as 10% (wt/vol) homogenates in physiological saline. Groups of four tga20 indicator mice were injected i.c. with 20 µl of each homogenate. The scrapie titre in each sample was determined from the mean incubation period in the indicator mice, by reference to a dose/incubation period response curve for ME7 scrapie-infected spleen tissue serially titrated in tga20 mice using the relationship: y = 9.4533–0.0595x (y, = log ID50 U/20 µl of homogenate; x, incubation period; R2 = 0.9562). As the expression level of cellular PrPc controls the prion disease incubation period, tga20 mice overexpressing PrPc are extremely useful as indicator mice in prion infectivity bioassays as they succumb to disease with much shorter incubation times than conventional mouse strains [63].
Spleens were removed and snap-frozen at the temperature of liquid nitrogen. Serial frozen sections (10 µm in thickness) were cut on a cryostat and immunostained with the following antibodies: FDCs were visualized by staining with mAb 7G6 to detect CR2/CR1 (CD21/CD35; BD Biosciences PharMingen), mAb FDC-M2 to detect C4 (AMS Biotechnology, Oxon, UK) or mAb 8C12 to detect CR1 (CD35; BD Biosciences PharMingen). Cellular PrPC was detected using PrP-specific polyclonal antibody (pAb) 1B3 [66]. B cells were detected using mAb B220 to detect CD45R (Caltag, Towcester, UK), or anti-CD19 (BD biosciences PharMingen). Marginal zone B cells were detected using mAb 1B1 to detect CD1d (BD Biosciences PharMingen). Sympathetic nerves were detected using tyrosine hydroxylase (TH)-specific pAb (Chemicon Europe).
For the detection of disease-specific PrP (PrPd) in spleens and brains, tissues were fixed in periodate-lysine-paraformaldehyde fixative and embedded in paraffin wax. Sections (thickness, 6 µm) were deparaffinised, and pre-treated to enhance the detection of PrPd by hydrated autoclaving (15 min, 121°C, hydration) and subsequent immersion formic acid (98%) for 5 min [67]. Sections were then immunostained with 1B3 PrP-specific pAb. For the detection of EGF-like module-containing mucin-like hormone receptor-like 1 (EMR1)-expressing macrophages, paraffin-embedded spleen sections were micro-waved in citric acid buffer (pH 6.0) for 10 min. Endogenous peroxidase activity was blocked using 1% hydrogen peroxidase in methanol, and macrophages detected using rat mAb F4/80 to detect EMR1 (clone CI:A3-1, AbD Serotec). For the detection of astrocytes, brain sections were immunostained with anti-glial fibrillary acidic protein (GFAP; DAKO, Ely, UK). For the detection of microglia, deparaffinised brain sections were first pre-treated with Target Retrieval Solution (DAKO) and subsequently immunostained with anti-ionized calcium-binding adaptor molecule 1 (Iba-1; Wako Chemicals GmbH, Neuss, Germany). Immunolabelling was revealed using HRP-conjugated to the avidin-biotin complex (Novared kit, Vector laboratories, Peterborough, UK). Paraffin-embedded tissue (PET) immunoblot analysis was used to confirm the PrPd detected by immunohistochemistry was proteinase K (PK)-resistant PrPSc [34]. Membranes were subsequently immunostained with 1B3 PrP-specific pAb.
For light microscopy, following the addition of primary antibodies, biotin-conjugated species-specific secondary antibodies (Stratech, Soham, UK) were applied followed by alkaline phosphatase or HRP coupled to the avidin/biotin complex (Vector Laboratories). Vector Red (Vector Laboratories) and diaminobenzidine (DAB; Sigma Aldrich, Dorset, UK) were used as substrates, respectively, and sections were counterstained with haematoxylin to distinguish cell nuclei. For fluorescent microscopy, following the addition of primary antibody, species-specific secondary antibodies coupled to Alexa Fluor 488 (green), Alexa Fluor 594 (red) dyes or Alexa Fluor 647 (blue) dyes (Invitrogen, Paisley, UK) were used. Sections were mounted in fluorescent mounting medium (DakoCytomation) and examined using a Zeiss LSM5 confocal microscope (Zeiss, Welwyn Garden City, UK).
Digital microscopy images were analyzed using ImageJ software (http://rsbweb.nih.gov/ij/) as described [68]. Intensity thresholds were first applied and then the number of pixels of each colour (black, red, green, yellow) were then automatically counted and presented as a proportion of the total number of pixels in each area under analysis. The preferential co-localisation of fluorochromes was determined by comparisons of the observed distribution of colours with those predicted by the null hypothesis that each element of positive staining was randomly and independently distributed. Values found to be significantly greater than the null hypothesis confirm significant co-localisation of fluorochromes. Spleens from 6 mice from each group were analyzed. From each spleen, 2 sections were studied and on each section data from 4 individual FDC networks collected. Thus, for each mouse group data from a total of 48 individual FDC were analysed. Similarly, data from 48 images from each group were analyzed to determine the preferential co-localisation of fluorochromes upon TH-positive sympathetic nerves within the spleen. A one-way ANOVA test was then used to compare the null hypothesis (that the pixels were randomly distributed) to the observed levels of co-localisation.
To assess antigen trapping by FDC in vivo, mice were passively immunized by intravenous injection with 100 µl preformed PAP immune complexes (Sigma). Spleens were removed 24 h later and the presence of FDC-associated immune complexes identified by IHC.
Data are presented as mean ± SE. Unless indicated otherwise, significant differences between samples in different groups were sought by one-way ANOVA. Values of P<0.05 were accepted as significant.
|
10.1371/journal.pgen.1000028 | Gene Activation Using FLP Recombinase in C. elegans | The FLP enzyme catalyzes recombination between specific target sequences in DNA. Here we use FLP to temporally and spatially control gene expression in the nematode C. elegans. Transcription is blocked by the presence of an “off cassette” between the promoter and the coding region of the desired product. The “off cassette” is composed of a transcriptional terminator flanked by FLP recognition targets (FRT). This sequence can be excised by FLP recombinase to bring together the promoter and the coding region. We have introduced two fluorescent reporters into the system: a red reporter for promoter activity prior to FLP expression and a green reporter for expression of the gene of interest after FLP expression. The constructs are designed using the multisite Gateway system, so that promoters and coding regions can be quickly mixed and matched. We demonstrate that heat-shock-driven FLP recombinase adds temporal control on top of tissue specific expression provided by the transgene promoter. In addition, the temporal switch is permanent, rather than acute, as is usually the case for heat-shock driven transgenes. Finally, FLP expression can be driven by a tissue specific promoter to provide expression in a subset of cells that can only be addressed as the intersection of two available promoters. As a test of the system, we have driven the light chain of tetanus toxin, a protease that cleaves the synaptic vesicle protein synaptobrevin. We show that we can use this to inactivate synaptic transmission in all neurons or a subset of neurons in a FLP-dependent manner.
| Genes turn on and off as a natural part of development. The nematode C. elegans has been an important model system for studying the roles of genes in animal development and physiology. However, worm researchers have had a limited toolkit for controlling gene activation. These drawbacks have been particularly restrictive when studying the function of a gene that has different roles in several cell types or at different times in development. Here we describe a way to turn any gene on at a specific time in specific cells. We provide a set of mix-and-match reagents that give researchers a way to quickly build new combinations of regulatory elements. These reagents will allow researchers to express a single gene in a wide array of temporal or spatial patterns, or to serially express many genes in a single cell. As a proof of principle, we made an artificial worm gene composed of a neurotoxin that would block neurotransmission. When we activated the gene in a small number of neurons in adult animals, these cells ceased to function. We anticipate that this new technique will find a wide variety of uses by the C. elegans community.
| Every widely-used genetic model organism can be manipulated to express genes introduced by the experimenter. In C. elegans it is simple to create a transgene by injecting DNA containing a complete genomic region. In most cases the transgene will be expressed in its native temporal and spatial pattern and will rescue the mutant phenotype. However, if the researcher would like to test the function of a gene at a specific time or in a specific tissue of the worm, one would need a very specific promoter. Due to a limited set of promoters available, it is often impossible to express a gene of interest in a specific cell. More importantly, there is only one temporally inducible promoter available – the heat-shock promoter. This promoter has been a workhorse of the field, but has a major drawback because it is expressed ubiquitously. Three techniques have been developed that provide more precise temporal and spatial control by making gene expression dependent on two independently controllable events.
The Chalfie laboratory has developed one solution by expressing the gene product as two complementary halves. In this case, the full gene product is reconstituted only in the cells that express both promoters. When expressed under the control of two different overlapping promoters, the complete protein is only reconstituted in a small number of cells. They have demonstrated that a two-part GFP can be used to label specific cells [1] and that a two-part caspase can be used to kill specific cells [2]. Their technique can also be applied to the temporal control of gene expression. If one of the promoters is a heat-shock inducible promoter, specific cells can be killed on command. The limitation of the two-part system is that it requires gene products that can reconstitute activity from two halves.
A second combinatorial control technique relies on temperature-dependent degradation of the mRNA of the target gene. This method was independently proposed by several groups, but made practical by Getz, Xu and Fire (A. Fire personal communication)[3].The nonsense mediated decay (NMD) pathway specifically degrades mRNAs with long 3′ untranslated regions containing many introns. Transgenes can be engineered with long 3′ UTRs so that their mRNAs are targeted for degradation. Strains with temperature-sensitive mutations in NMD components cannot degrade these mRNAs at the restrictive temperature. Thus, the transgene is more strongly expressed at the restrictive temperature (when the NMD system is not functioning) than at the permissive temperature (when NMD is actively degrading aberrant mRNAs). This system gives some degree of control over expression levels, although there is a moderate background level of expression even in the “off” state, which has limited its use.
Bacaj and Shaham have developed a method to add heat-shock control to a transgene expressed under a tissue-specific promoter[4]. The method uses a mutant background in which the heat-shock response is defective. By rescuing the mutant defect with tissue-specific promoters, a tissue-specific heat shock response is generated. This method requires using the hsf-1 mutant background and gene expression from the heatshock promoter is still acute. Ideally, a combinatorial expression system would be in the wild-type strain and provide a permanent change in the genotype of the cells of interest.
Here we describe a method that uses FLP recombinase to control transgene expression in C. elegans. In this configuration, the transgene is expressed at the combinatorial intersection of two different promoters: either two spatially restricted promoters, or a spatially restricted and temporally controlled promoter.
The site specific recombinases Cre and FLP have been used in many systems to control gene structure and expression [5]–[10]. These enzymes align tandem copies of the target sequence, perform site-specific recombination, and remove the sequence between the targets as a circular DNA molecule. If the intervening sequence disrupts expression, removal by recombination will allow the transgene to be activated. We designed an “off cassette”, composed of a putative transcriptional terminator, that could be placed between a promoter and a coding region to disrupt expression of the coding region (Figure 1). Expression of FLP recombinase will excise the cassette as a circular DNA molecule. This rearrangement will place the promoter adjacent to the downstream coding regions, converting the transgene to the “on” state. Thus, expression is dependent on both the promoter driving the coding region and the promoter driving expression of FLP.
The constructs are designed to provide a fluorescent readout in either the ‘off’ or ‘on’ state. The FRT-flanked “off cassette” contains the mCherry coding sequence followed by the 3′ genomic region from the let-858 gene. The red-fluorescent mCherry protein acts as a reporter for the promoter activity of the transgene, verifying that the transgene is present and expressing in the expected cell types prior to FLP-induced recombination. The let-858 3′ genomic region provides the poly-adenylation signal for mCherry mRNA as well as a putative transcriptional terminator, preventing expression of downstream sequences. GFP acts as a reporter for transcriptional read-through or reinitiation of transcription in the let-858 genomic region. Recombination of the FRT sequences removes the mCherry coding sequence and the terminator and brings a GFP coding region under control of the promoter, to indicate that the FLP reaction was successful. The coding region for any gene of interest can be fused in frame to the 3′ end of the GFP sequence, thus providing FLP-inducible expression of that protein.
To speed assembly of constructs and to make use of genome reagents generated by other laboratories, we based our constructs on the Multisite Gateway™ in vitro recombination system from Invitrogen [11],[12]. These vectors allow rapid, modular construction of plasmids using a site-specific recombinase from the bacteriophage lambda (Figure 2A). The recombinase target sequences are designed to allow pairwise recombination between specific DNA sequences. The standard multisite system uses three libraries of entry vectors: a promoter library, a cDNA library, and transcriptional terminator library (to be recombined into slots 1, 2 or 3, respectively). Individual components from these libraries can be selected and mixed with a destination vector to generate a desired combination of promoter, cDNA and terminator in a single reaction. Many reagents are already available for the construction of C. elegans expression constructs using this system. The C. elegans promoterome consortium has cloned the 5′ regulatory regions of approximately 6,000 different C. elegans genes into Gateway ‘entry’ vectors compatible with the slot 1 entry vectors [13]. The ORFeome project from the Vidal laboratory has cloned cDNAs from approximately 11,000 genes into slot 2 entry vectors [14].
Because the promoterome constructs are designed to recombine directly to ORFeome constructs, it was not possible to introduce the off-cassette between them. Instead, we built vectors that have the off-cassette in either the promoter slot (slot 1) (Figure 2B) or in the cDNA slot (slot 2) (Figure 2C). The former arrangement requires construction of a promoter-FRT construct, but is compatible with the existing ORFeome library. The latter requires placing the ORF to the third slot, but is compatible with the existing promoterome library.
The ORFeome-compatible format requires that the “off-cassette” be cloned into entry vectors containing various promoters. We generated several potentially useful promoter-“off-cassette” entry clones (Figure 2B, Table 1). These plasmids can then be recombined with any ORFeome clone to produce a large number of different open reading frames under the same FLP-inducible promoter sequence. This arrangement is particularly useful when one is primarily interested in expressing a number of different proteins in a particular cell or tissue.
For the promoterome-compatible constructs, we placed the off-cassette into the position usually occupied by the ORF of interest. We then used standard cloning methods to place several ORFs of interest into slot 3 entry vectors (Figure 2C, Table 2). Any of the set of promoterome clones can be recombined with these clones to express a single ORF in a wide array of tissues in a FLP-dependent manner. This arrangement is particularly useful when one is determining the focus of gene rescue or inactivating different groups of cells.
As a proof of principle, we produced two FLP-inducible constructs that would express GFP-tagged histone in different muscle cells. The first construct was in the ORFeome compatible configuration described in Figure 2B. We used a pharyngeal muscle promoter (Pmyo-2) followed by the “off-cassette” in the first Gateway slot. We added the HIS-11 open reading frame in the second slot and an unc-54 3′ polyadenylation site in the third slot (Figure 3A–D). We injected this plasmid together with a plasmid encoding FLP-recombinase under the control of the hsp-16.48 heat shock promoter [15] and a lin-15(+) co-injection marker to produce a line of transgenic worms. As expected, these worms expressed diffuse red fluorescence in their pharyngeal muscles with no apparent GFP fluorescence (Figure 3A, B). The lack of GFP fluorescence confirms that the let-858 3′ genomic region functions as expected to prevent read-through into the downstream gene. We then exposed the worms to a one hour, 34° heat shock and imaged the worms 2 hours, 3 hours and 15 hours later. Although no GFP was apparent at 2 hours, the heat shock-induced expression of the GFP::HIS-11 fusion protein was visible at three hours (Figure S1) and was strong at 15 hours (Figure 3C, D). From roughly 200 animals observed under a dissecting microscope, every worm examined exhibited green nuclei in muscle cells expressing mCherry. We counted the GFP positive nuclei in five randomly selected adult worms 22 hours after a 34° heat shock. We found strong expression in 98% (123/125) of major pharyngeal muscle cells (pm3-pm8, 25 nuclei per worm). Determining expression in pm1 and pm2 is complicated, since these cells are surrounded by the pm3 muscle cell (see http://www.wormatlas.org/handbook/fig.s/alim9.jpg). Bright mCherry fluorescence obscured expression of the cassette in these cells before heatshock. Thus, we could not determine whether the myo-2 promoter was driving expression of the transgene. After heatshock we did not score GFP expression in any pm1 nuclei (0/15) and only identified weak expression in 7/15 pm2 nuclei. In summary, robust expression due to recombination was observed in all cells in which the transgene was unequivocally expressed.
The second construct was in the promoterome-compatible configuration described in Figure 2C. We generated this construct by recombining a myo-3 promoterome plasmid in the first position, the FRT cassette was placed into the second position and a HIS-11 ORF cloned in front of the unc-54 3′ genomic region in the third position. This construct should express the GFP::HIS-11 fusion in the body muscles of the worm after FLP expression. There was diffuse red but no green fluorescence in the body wall muscles prior to heat shock (Figure 3E, F), but strong induction of nuclear-localized GFP 15 hours after a 1 hour exposure to 34° heat shock (Figure 3G, H). Using a dissection microscope to survey a large number of animals, GFP was detected in body muscle nuclei in all transgenic animals. We examined 5 animals in greater detail and observed expression in body wall muscles, vulval muscles and the anal depressor muscles in all animals, demonstrating that FLP works in all major body muscle types.
If all of the copies of the transgene are recombined, mCherry expression should not occur after FLP expression. However, in all cells examined (using the myo-2, myo-3 and unc-47 promoters), we never saw substantial loss of mCherry expression even after induction of the GFP fusion protein. Although the half-life of mCherry in worm cells is not known, the half life of GFP in body muscle cells is greater than 24 hours [16]. Thus, perdurance of mCherry could be a source of this remaining expression. It is also possible that several copies of the transgene on the extrachromosomal array are refractory to recombination. This could be due to the chromatin structure of the repetitive arrays, or damage to the arrays that occurred during their formation.
One important application for the FLP-on method is to silence neurotransmission in specific neurons in a temporally-controlled manner. The tetanus toxin light chain is a highly specific protease that recognizes and cleaves synaptobrevin [17]. Since synaptobrevin is one of the three SNARE-class proteins required for calcium-mediated release of neurotransmitter [18], expressing tetanus toxin eliminates the ability of a neuron to signal through chemical synapses. The high specificity of tetanus toxin preserves all other functions of the neuron, including electrical coupling through gap junctions. Temporal control of tetanus toxin expression is important for two reasons. First, synaptic transmission is essential for development. Animals lacking acetylcholine neurotransmission (in cha-1 mutants) or all synaptic neurotransmission (in unc-13 deletion alleles) are arrested in the first larval stage. Thus, loss of synaptic transmission in at least some neurons may lead to broad developmental defects that blind the investigator to functions for a neuron in the adult. Using the FLP-on method, expression of tetanus toxin can be activated after the developmental requirement for a neuron. Conversely, early loss of neuronal function can lead to developmental compensation in the nervous system. For example, when particular sensory neurons were ablated in males in the L3 stage the nervous system compensated for their loss; by contrast ablation of these same neurons in the L4 stage led to behavioral abnormalities in the adult [19]. The ability to silence these neurons in adulthood allows one to assay the function of the synaptic connectivity of a neuron in an existing developmentally-normal circuit.
To test if tetanus toxin induction could inactivate neurotransmission, we designed an ORFeome-compatible tetanus toxin expression construct. The tetanus toxin sequence was inserted into slot 2, and the FRT-flanked terminator was placed in slot 1 after the GABA-specific neuron promoter Punc-47. unc-47 encodes the vesicular GABA transporter required to fill synaptic vesicles with GABA, and is expressed in all GABA neurons. We chose this promoter because loss of GABA function produces two distinct phenotypes: a locomotory phenotype and a defecation motor program defect.
Animals lacking GABA neurotransmission exhibit a distinctive locomotory defect [20]. These animals cannot back when touched on the head but rather execute an accordion-like “shrinking” response. This symmetrical contraction of the body muscles occurs due to the lack of contralateral inhibitory inputs from the GABA motor neurons. Prior to heat shock, transgenic worms exhibited normal movement. Animals were exposed to heat shock at 34° for one hour. 24 hours later the animals exhibited a clear shrinking phenotype (Video S1). The structure of the GABA neurons was not affected by toxin expression (Figure S2), suggesting that the toxin was simply silencing synaptic transmission in these neurons. As expected the shrinking phenotype was associated with the presence of the transgene: 99 of 100 shrinker animals carried the Pmyo-2::GFP extrachromosomal array marker. The presence of nonshrinking animals in the population was due to loss of the extrachromosomal array: 29 of the 30 non-shrinker animals had lost the Pmyo-2::GFP transgene marker. The one non-shrinking animal carrying the array lacked the mCherry marker in the VD and DD motor neurons, demonstrating that this animal was a somatic mosaic which lacked the array in the motor neurons.
GABA function is also required for the motor program of the defecation cycle [21],[22]. The defecation motor program requires the AVL and DVB GABA neurons to stimulate contraction of the enteric muscles via a GABA-gated cation channel [23]. Prior to heat shock, transgenic worms expressing mCherry in the AVL and DVB GABA neurons had wild-type enteric muscle contractions during the defecation cycle (n = 10 worms, 10 cycles per worm, Figure 4). After heat shock, transgenic animals lacked the enteric muscle contractions during the defecation cycle (Figure 4), as expected for loss of GABA neurotransmission. Because AVL and DVB are partially redundant, these data suggest that FLP function must be greater than 95% effective. In addition, we were able to see induction of GFP-tagged tetanus toxin by confocal microscopy (not shown). Tetanus toxin expression is continuous using this FLP-on construct; thus, unlike direct heatshock-induced expression of the toxin, the behavioral change is permanent.
FLP-dependent gene expression will have two general uses in C. elegans: to provide spatial specificity and temporal specificity for gene expression. Because it provides expression within the spatial overlap of two promoters, the method essentially squares the number of transgene expression patterns now available. In many cases, especially in the nervous system, this can restrict expression to single cells of the worm.
FLP recombinase can also be used to provide temporal specificity for gene expression. For most purposes temporal control is best provided by activation of the heatshock promoter. Thus, heatshock-driven expression of FLP recombinase will confer temporal specificity to any promoter. This will be particularly useful for expression of dominant negative or constitutively activated gene products that may kill cells before their effects can be assayed. Moreover, acute expression can outflank the criticism most often thrown at genetic analysis –that homeostatic mechanisms will compensate for chronic genetic changes. In this way the heatshock FLP-on method and the cell-specific rescue of hsf-1 method developed by the Shaham laboratory are similar. In the FLP-on method, expression is permanently activated, whereas in the hsf-1 rescue method, expression is acute and depends on the length of the heatshock response. Depending on the circumstances and the gene product being expressed, one method may be more advantageous than the other.
For the neuroscientist, FLP-on constructs provide a method for analyzing the role of a neuron in a circuit by killing a specific cell, inactivating the cell, or activating the cell. By combining our system with the Chalfie split caspase system (Table 1), cells can be killed at the intersection of three promoters. These three promoters could each provide a spatial component, giving extremely tight spatial control to potentially address the few single neuron types that have escaped the two-promoter system. Alternatively, one of the three could provide temporal inducibility, adding heat shock control onto many two-promoter single-cell killing experiments. Silencing the cell can allow specific dissection of the chemical synapses present in the system while leaving the gap junction connections in the network intact. In addition, leaving the cell in place will minimize any developmental perturbations of the circuit that might be caused by removing a neuron by the cell death pathway. In addition to eliminating a cell entirely, or silencing the chemical neurotransmission in a cell, single cells can be electrically silenced or activated on command by expressing halorhodopsin, a light activated chloride pump protein [24], or channelrhodopsin, a light activated cation channel protein [25], at the intersection of two spatial promoters.
For the developmental biologist, FLP-on constructs can provide temporal control of gene expression so that the role of a gene during different developmental periods can be evaluated. This application is limited by the time delay required for FLP expression, recombination and gene expression. We observed a three hour delay from the end of heat shock to the expression of Pmyo-2::GFP. FLP regulation will also be useful for analyzing promoter expression patterns by permanently marking descendents of cells that have expressed a promoter. Combined with the known cell lineage of C. elegans, the expression pattern can quickly pinpoint the expression pattern of a transgene that may come on very briefly and in only a few cells in the embryo. Using traditional GFP reporters, such expression might be missed entirely, or if it is detected it might be very difficult to unambiguously identify the expressing cell among the dividing cells of the embryo. This technique could be combined with forward genetic screens. A FLP-dependent histone GFP reporter (Table 1) will easily identify mutant backgrounds in which a gene is transiently misexpressed during development.
In conclusion, FLP-dependent excision of a transcriptional terminator provides a simple way to make expression of a transgene dependent on the activity of two promoters. Depending on the promoters used, FLP-on constructs can confer combinatorial spatial or temporal control of gene expression in C. elegans. We anticipate that the combination of the wide availability of the Gateway reagents and the imagination of the C. elegans community will yield many new applications.
pWD157 (slot 2 TeTx): Tetanus toxin light chain was PCR amplified from CMV-LC-Tx (Heiner Niemann) using primers containing attB1 and attB2 tails GGGGACAAGTTTGTACAAAAAAGCAGGCTTAATGCCGATCACCATCAACAACTTC and GGGGACCACTTTGTACAAGAAAGCTGGGTTTAAGCGGTACGGTTGTACAGGTT and recombined with the attP1 and attP2 sites in the slot 2 donor vector pDONR221 (Invitrogen) using the BP recombination reaction.
pWD170 (slot 3 TeTx): Tetanus toxin light chain was PCR amplified from CMV-LC-Tx using primers GTATGCCGATCACCATCAACAAC and TTAAGCGGTACGGTTGTACAGG and cloned as a blunt fragment into pMH472 using SrfI. pMH472 is a slot 3 entry vector containing a SrfI site followed by two stop codons and then the unc-54 3′ UTR.
pPD119FRTRFPGFP: A fragment of mRFP1 was PCR amplified and cloned between the SalI and SmaI sites in pPD118.33. pPD118.33 is a Pmyo-2::GFP plasmid (gift of Andrew Fire). Tandem FRT sites [8] were cloned at the junctions as SalI-BamHI and MluI-SmaI double-stranded oligonucleotides.
pWD176 (Pmyo-2-FRT-mCherry-terminator-FRT-GFP-terminator): pPD119FRTRFPGFP was modified to replace mRFP with mCherry. An MluI-KpnI double-stranded oligonucleotide containing an FRT fragment in a different frame was used to replace the second FRT in pPD119FRTRFPGFP. The plasmid was then cut with BamHI and Mlu and a PCR fragment containing mCherry followed by the let-858 3′ end was ligated in as a BamHI-BssHII digested fragment. This produced a myo-2 promoter::FRT::mCherry terminator::FRT::GFP::unc-54 3′ UTR plasmid.
pWD177 (slot 1 Pmyo-2-FRT-mCherry-terminator-FRT): The Pmyo-2::mRFP FRT cassette::GFP fragment from pPD119FRTRFPGFP was PCR amplified using primers containing attB4 and attB1 tails GGGGACAACTTTGTATAGAAAAGTTGCTTGCATGCCTGCAGGTCGAGG and GGGGACTGCTTTTTTGTACAAACTTGTTTTGTATAGTTCGTCCATGCCATG and recombined with the attP4 and attP1 sites in the slot 1 donor vector pDONR P4-P1R (Invitrogen) using the BP recombination reaction to make pWD159. The mRFP cassette was replaced with mCherry using a BamHI-XhoI fragment from pWD176. This plasmid was sequenced with primers T7, M13fwd, mCh r1: CTTTCACTTGAAGCTTCCCATCCC, GFP-I-r2: CTCCAGTGAAAAGTTCTTCTCC, and GFP-I-r1: TTGTGCCCATTAACATCACC.
pWD178 (slot 2 FRT-mCherry-terminator-FRT): The mRFP FRT cassette::GFP fragment from pPD119FRTRFPGFP was PCR amplified using primers containing attB1 and attB2 tails GGGGACAAGTTTGTACAAAAAAGCAGGCTTACGAAGTTCCTATTCTCTAGA and GGGGACCACTTTGTACAAGAAAGCTGGGTTTTTGTATAGTTCGTCCATGCC and recombined with the attP1 and attP2 sites in pDONR221 (Invitrogen) using the BP recombination reaction. The mRFP cassette was replaced with mCherry using a BamHI-XhoI fragment from pWD176. This plasmid was sequenced with primers T7, M13fwd, GFP-I-r2: CTCCAGTGAAAAGTTCTTCTCC, and GFP-I-r1: TTGTGCCCATTAACATCACC
pWD179: The 1.2 kb unc-47 promoter was PCR amplified from plasmid pKS4.1 (K. Schuske) using primers: CGAACGCATGCGGATCCCGGAACAGTCGAAAG and CGAACGTCGACGCATCTGTAATGAAATAAATGTGACGCTG. This sequence was inserted into pWD159 as an Sph-Sal fragment. The mRFP cassette was replaced with mCherry using a SalI-BstZ17I fragment from pWD176. This plasmid was sequenced with primers T7, M13fwd, mCh r1: CTTTCACTTGAAGCTTCCCATCCC, GFP-I-r2: CTCCAGTGAAAAGTTCTTCTCC, and GFP-I-r1: TTGTGCCCATTAACATCACC.
pWD180: The 1.2 kb rab-3 promoter was PCR amplified from N2 genomic DNA using primers: CGAACGCATGCATCTTCAGATGGGAGCAGTGG and CGAACGTCGACGCATCTGAAAATAGGGCTACTGTAGAT. This DNA was inserted into pWD159 as an Sph-Sal fragment. The mRFP cassette was replaced with mCherry using a BamHI-XhoI fragment from pWD176. This plasmid was sequenced with primers T7, M13fwd, mCh r1: CTTTCACTTGAAGCTTCCCATCCC, GFP-I-r2: CTCCAGTGAAAAGTTCTTCTCC, and GFP-I-r1: TTGTGCCCATTAACATCACC.
pWD195 (slot 2 Histone 2B): The his-11 ORF was PCR amplified using primers containing attB1 and attB2 tails GGGGACAAGTTTGTACAAAAAAGCAGGCTTACCACCAAAGCCATCTGCCAAGG and GGGGACCACTTTGTACAAGAAAGCTGGGTATTACTTGCTGGAAGTGTACTTGG using pGH42 (G. Hollopeter) as template. The DNA fragment was recombined with the attP1 and attP2 sites in pDONR221 (Invitrogen) using the BP recombination reaction.
pWD198 (Pmyo-3-FRT-mCherry-FRT-GFP-Histone): A multisite LR reaction was performed using pEntry[4-1] Pmyo-3 (Open Biosystems), pWD178, pGH42 (G. Hollopeter), and pDEST R4-R3 (Invitrogen).
pWD199 (Punc-47-FRT-mCherry-FRT-GFP-TeTx): A multisite LR reaction was performed using pWD179, pWD157, pMH473 (M. Hammarlund), and pDEST R4-R3 (Invitrogen). pMH473 is an attP2-attP3 entry clone carrying the unc-54 3′ polyadenylation site.
pWD200 (Pmyo-2-FRT-mCherry-FRT-GFP-Histone): A multisite LR reaction was performed using pWD177, pWD195, pMH473 and pDEST R4-R3 (Invitrogen).
pWD203 (slot 3 Caspase C-terminus): The caspase 3 C-terminal fragment, leucine zipper and unc-54 3′ UTR from TU#813 [2] was PCR amplified using primers: GGGGACAGCTTTCTTGTACAAAGTGGGAAGTGGTGTTGATGATGACATGGCG and GGGGACAACTTTGTATAATAAAGTTGCCATAGACACTACTCCACTTTC and BP cloned into pDONR P2R-P3 (Invitrogen).
pWD204 (slot 3 Caspase N-terminus) The caspase 3 N-terminal fragment, leucine zipper and unc-54 3′ UTR from TU#814 [2] was PCR amplified using primers containing attB2 and attB3 tails: GGGGACAGCTTTCTTGTACAAAGTGGGAATGGCTAGCGCACAGCTGGAGAAG and GGGGACAACTTTGTATAATAAAGTTGCCATAGACACTACTCCACTTTC and BP cloned into pDONR P2R-P3 (Invitrogen).
pWD79-2RV (Phsp-16-48:FLP): A PCR fragment containing the FLP coding sequence from pOG44 (Stratagene) was cloned as an MluI-NheI fragment into pJL44 (J.L. Bessereau). pJL44 contains the Phsp-16-48 heat-shock promoter and the glh-2 3′ UTR. The FLP coding sequence in pOG44 contains a point mutation which was repaired using a PCR fragment from the FLP coding sequence cloned into pBR322 (Makkuni Jayaram). An artificial intron was introduced into the FLP coding sequence at the EcoRV site using a double-stranded oligo: GTAAGTTTAAACATATATACTAACTAACCCTGATTATTTAAATTTTCAG.
pWD172 (slot 2 FLP-stop): The FLP ORF was PCR amplified from pWD79-2RV using primers containing attB1 and attB2 tails: GGGGACAAGTTTGTACAAAAAAGCAGGCTTAATGCCACAATTTGGTATATTATGT and GGGGACCACTTTGTACAAGAAAGCTGGGTATTATATGCGTCTATTTATGTAGGATG and BP cloned into pDONR221 (Invitrogen). This version of FLP contains a stop codon at the native C terminus.
pWD173 (slot 2 FLP-no stop): The FLP ORF was PCR amplified from pWD79-2RV using primers containing attB2 and attB3 tails: GGGGACAAGTTTGTACAAAAAAGCAGGCTTAATGCCACAATTTGGTATATTATGT and GGGGACCACTTTGTACAAGAAAGCTGGGTATATGCGTCTATTTATGTAGGATG and BP cloned into pDONR221 (Invitrogen). This version of FLP does not contain a stop codon, and can be used to make C-terminally tagged proteins.
Gateway BP and LR in vitro recombination reactions were carried out according to manufacturer instructions.
Strains used in this study: The wild type is Bristol N2.
EG3251 unc-25(e156) III.
EG4859 oxEx1099 was made by injecting: pWD198 (Pmyo-3-FTF-GFP-Histone) at 5 ng/ul, pWD79-2RV (Phsp-16-48:FLP) at 45 ng/ul, pPD118.33 (Pmyo-2::GFP)(A. Fire) at 1 ng/ul, and pL15EK (lin-15(+))[26] at 50 ng/ul into N2 animals.
EG4860 oxEx1100 was made by injecting: pWD199 (Punc-47-FTF-GFP-TeTx) at 5 ng/ul, pWD79-2RV (Phsp-16-48:FLP) at 45 ng/ul, pPD118.33 (Pmyo-2::GFP)(A. Fire) at 1 ng/ul, pL15EK (lin-15(+))[26] at 50 ng/ul into N2 animals.
EG4866 lin-15(n765ts) X oxEx1101 was made by injecting: pWD200 (Pmyo-2-FTF-GFP-Histone)at 2 ng/ul, pWD79-2RV (Phsp-16-48:FLP) at 45 ng/ul, pL15EK (lin-15(+))[26] at 50 ng/ul into MT1642 lin-15(n765ts).
‘FTF’ symbolizes the off cassette composed of FRT-mCherry-terminator-FRT. |
10.1371/journal.ppat.1007044 | The HTLV-1 gp21 fusion peptide inhibits antigen specific T-cell activation in-vitro and in mice | The ability of the Lentivirus HIV-1 to inhibit T-cell activation by its gp41 fusion protein is well documented, yet limited data exists regarding other viral fusion proteins. HIV-1 utilizes membrane binding region of gp41 to inhibit T-cell receptor (TCR) complex activation. Here we examined whether this T-cell suppression strategy is unique to the HIV-1 gp41. We focused on T-cell modulation by the gp21 fusion peptide (FP) of the Human T-lymphotropic Virus 1 (HTLV-1), a Deltaretrovirus that like HIV infects CD4+ T-cells. Using mouse and human in-vitro T-cell models together with in-vivo T-cell hyper activation mouse model, we reveal that HTLV-1’s FP inhibits T-cell activation and unlike the HIV FP, bypasses the TCR complex. HTLV FP inhibition induces a decrease in Th1 and an elevation in Th2 responses observed in mRNA, cytokine and transcription factor profiles. Administration of the HTLV FP in a T-cell hyper activation mouse model of multiple sclerosis alleviated symptoms and delayed disease onset. We further pinpointed the modulatory region within HTLV-1’s FP to the same region previously identified as the HIV-1 FP active region, suggesting that through convergent evolution both viruses have obtained the ability to modulate T-cells using the same region of their fusion protein. Overall, our findings suggest that fusion protein based T-cell modulation may be a common viral trait.
| In order to successfully infect and persist in their hosts, viruses utilize multiple strategies to evade the immune system. HIV utilizes membrane interacting regions of its envelope protein, primarily used to fuse with its target cells, to inhibit T-cell activation. Yet, it is unknown whether this ability is shared with other viruses. In this study we examined the T-cell inhibitory activity of the envelope protein of the Human T-lymphotropic virus 1 (HTLV-1), which infects T-cells. We focused on a functionally conserved region of HTLV’s and HIV’s fusion proteins, the fusion peptide (FP). Here, we reveal that HTLV’s FP inhibits the activity of T-cells in-vitro and in a T-cell hyper activation model in mice. This inhibition is characterized by downregulation of the T-cell Th1/type 1 response, leading to an elevated T-cell Th2/type 2 response observed by transition in the profiles of mRNA, cytokines and regulatory proteins. Furthermore, we demonstrate that the HTLV and HIV FPs inhibit T-cell activation at different levels of the signaling cascade. Although the HTLV FP’s mechanism of T-cell inhibition differs from the HIV’s FP, our findings suggest that FP mediated immune evasion might be a trait shared between different viruses.
| The mutual evolutionary pressure between viruses and their hosts has driven viruses to adopt various immune evasion mechanisms [1–4]. Many evasion strategies of enveloped viruses, such as antigen presentation antagonism and glycan shielding, can be mediated by their fusion glycoproteins (reviewed in [5]). One of the most studied glycoproteins in this aspect is HIV’s gp41, which aside from its crucial role in virus-cell membrane fusion [6, 7], was shown to inhibit T-cell activity. This was proposed to occur during the fusion process using several membrane interacting segments [8–10], including the fusion peptide (FP) [11, 12] (reviewed in [9]). This strategy of modulating the immune response during membrane fusion has only been reported for HIV, although other enveloped viruses infect T-cells through membrane fusion as well [13–16]. We hypothesized that other human enveloped viruses might share HIV’s strategy of immune suppression.
To this aim we examined the immune modulatory ability of the human T-lymphotropic virus-1 (HTLV-1), which exploits CD4+ T-cells as its primary target cell population [17]. As both HTLV-1 and HIV-1 are members of the retroviridae family they share a common ancestor and similar genomic architecture [18, 19]. Their envelope proteins are similarly structured and are composed of two non-covalently bound subunits, gp46/gp21 in HTLV and gp120/gp41 in HIV, which bind cellular receptors and initiate fusion, respectively [20, 21]. Both viruses utilize several proteins to interfere with T-cell activity and manipulate the anti-viral immune response (23–25). HTLV’s p12 and p8 promote the proteosomal degradation of MHC-I and downregulate TCR complex signaling, respectively [22] while HIV’s Nef and Vpu downregulate MHC-I from the cell surface and promote internalization and degradation of CD4 in infected cells [23, 24]. Additionally, HTLV-1 has been previously reported to harbor an immunosuppressive domain (ISD) within its envelope transmembrane subunit gp21 that is conserved between different retroviral envelope proteins [25]. The ISD that is concealed by the envelope’s surface subunit [26, 27], has been reported to inhibit T-cell proliferation [25], to be crucial for viral infection in vivo [27] and to support tumor cells immune escape [26, 28, 29].
Suppression of TCR induced activation by HIV is well characterized and was shown to occur by targeting several TCR complex components via gp41 in the membrane [8, 9, 11, 30]. A membranotropic region of HTLV-1 gp21 is the FP that is concealed within the envelope complex. Following binding of the surface subunit to the cellular receptor, a conformational change exposes the FP leading to its insertion into the plasma membrane and to fusion with the host cell [31, 32]. Therefore, we decided to focus on the FP region as a possible immune suppressor of TCR activation in the membrane.
In this study we utilized in-vitro and in-vivo assays including T-cell proliferation and an experimental autoimmune encephalomyelitis (EAE) mouse model to investigate the ability of the HTLV-1 gp21 FP to interfere with T-cell activity. We reveal that the HTLV FP inhibits T-cell activation downstream of the TCR complex in contrast to the HIV FP that specifically targets the TCRα subunit. Moreover, the HTLV FP markedly reduced manifestation of an in-vivo EAE mouse model. Downregulation of T-cell activity was associated with reduced expression and secretion of Th1-specific cytokines and an elevated expression and secretion of Th2-specific cytokines. This transition in cytokine pattern was correlated to a decreased expression of T-bet and an elevated expression of Gata3, Th1- and Th2- specific transcription factors respectively. Interestingly, the HIV FP had no effect on both T-bet and Gata3 expression levels. This study suggests that in addition to its role in fusion, the HTLV FP interferes with T-cell activation by downregulating the type 1 anti-viral immune response, consequently leading to an elevated type 2 response. Overall, these findings reveal that like HIV, HTLV-1 adopted a similar strategy of immune suppression by it fusion protein, pointing to a possible prevailing trait of human T-cell viruses.
To examine whether other viruses can utilize their FPs to interfere with T-cell activity we initially investigated the immunosuppressive ability of the HTLV FP on primary C57BL/6J mMOG(35–55)-specific T-cells upon activation by antigen presenting cells (APCs). We compared this activity to the well characterized HIV FP and to the bovine leukemia virus (BLV) and Jembrana disease virus (JDV) FPs. BLV and JDV are HTLV and HIV equivalents in cattle, respectively (Table 1). The HTLV FP was found to inhibit T-cell proliferation with equal potency to that of HIV’s FP and significantly stronger than the BLV FP. The JDV FP showed no inhibitory activity (Fig 1A). The HTLV FP was not toxic up to 4-fold higher than the concentrations used in this study (S1A Fig), even when viability was measured following 72 hours incubation of cells with the peptide (S1B Fig).
T-cells can be activated in-vitro either directly through TCRα and β by antigen presentation, downstream to TCRα and β using antibodies against CD3 and CD28 or downstream to the entire TCR complex using the PKC activator PMA together with the Ca+2 ionophore Ionomycin [8, 9]. To examine where in the TCR signaling cascade the HTLV FP exerts its inhibitory activity, we activated T-cells using either APCs, CD3 and CD28 antibodies or PMA and Ionomycin. The HTLV FP inhibited T-cell proliferation induced at all three levels of activation with equal potency, while inhibition by the HIV FP that specifically targets the TCRα subunit, decreased significantly when activation was downstream of TCRα and β (Fig 1B), as previously reported [11]. This indicates that in contrast to HIV-1’s FP, the HTLV FP does not inhibit T-cell activation by targeting the TCR complex.
Following the previous results we tested HTLV FP specificity by assessing its inhibitory activity on activated macrophages. Primary mouse bone marrow derived macrophages (BMDM) were isolated, grown and stimulated using LTA, LPS, or PAM3CSK4, toll like receptor (TLR) 2/6, 4/4, and 2/1 ligands, respectively. The effect of HTLV FP treatment on TNFα and IL6 secretion was measured by ELISA. The HTLV FP had no effect on cytokine secretion from primary mouse BMDM (S2 Fig), demonstrating that the peptide selectively inhibits T-cell activation.
To further characterize the inhibitory mechanism of the HTLV FP, we examined its effect on the expression level of several Th1 and Th2 specific genes that are transcribed upon T-cell activation [33, 34]. C57BL/6J mMOG(35–55)-specific primary T-cells were activated using APCs. RNA was extracted 24hr following activation and mRNA levels were determined using RT-qPCR. The HTLV FP reduced the mRNA levels of the Th1-specific genes IFNG, LTA and the Th1 key mediator STAT4 [35, 36] (Fig 2A). On the other hand, the HTLV FP elevated the mRNA levels of the Th2-specific genes IL4 and IL10 (Fig 2A). Yet, Tumor necrosis factor α (TNF-α), that is expressed by both subsets [37], was not affected (Fig 2A).
To determine whether the observed changes in gene expression can be observed at the protein level, we performed ELISA for selected cytokines. C57BL/6J mMOG(35–55)-specific primary T-cells were activated using APC and supernatants were collected 24hr following activation. The HTLV FP inhibited IFN-γ secretion, elevated IL4 secretion and had no effect on TNF-α secretion from activated T-cells (Fig 2B), further corroborating our RT-qPCR results. These findings suggest that HTLV-1 might utilize its FP to restrict the T-cell antiviral immune response by downregulation of the Th1 and upregulation of the Th2 responses.
As the HTLV FP was shown to inhibit mMOG(35–55)-specific primary T-cell activation in-vitro (Fig 1A and 1B) by downregulating their Th1 response (Fig 2), we examined inhibition of pathogenic MOG(35–55)-specific T-cells in-vivo by the HTLV FP. We tested this in an EAE model, which is a widely used mouse model that mimics chronic MS in humans [38, 39] and is considered a CD4+ Th1-mediated autoimmune disease [40, 41]. We performed an initial experiment in which C57BL/6J mice were immunized with MOG35-55/CFA for EAE induction and were either treated with a single dose of HTLV FP (1mg/kg) or vehicle. Clinical manifestation of EAE for vehicle-treated mice was first observed at 8 days post immunization (DPI), reaching a severe disease at 10 DPI, and was accompanied with a substantial loss of weight (S3A and S3B Fig). Moreover, clinical severity was correlated to reduced locomotion as a result of hind limb ataxia or paralysis (S1 and S2 Movies). In contrast, HTLV FP treated mice showed only mild clinical symptoms that were first observed at 11 DPI and were followed only by minor weight fluctuations (S3A and S3B Fig). Additionally, most of the HTLV FP treated mice exhibited normal locomotory behavior throughout the experiment (S3 and S4 Movies). As disease manifestation in vehicle treated mice was relatively severe, experiment was terminated at 14 DPI due to institutional animal care and use committee (IACUC) limitations and requirements.
In order to examine whether the HTLV FP sequence is crucial for its inhibitory activity we synthesized a scrambled HTLV FP peptide (HTLV Scr), consisting of the same amino acid composition and length as the HTLV FP (Table 1). The effect of both peptides on mMOG(35–55)-specific primary T-cell activation was compared. The HTLV FP inhibited both T-cell proliferation and IFN-γ secretion with higher potency than the HTLV Scr (Fig 3A and 3B), demonstrating that the HTLV FP sequence is critical for its inhibitory activity.
We then performed an additional EAE experiment in which mice were treated with a single dose of HTLV FP (1mg/kg), HTLV Scr (1mg/kg) or vehicle. EAE clinical signs of vehicle and HTLV Scr treated groups ascent between 11 to 16 DPI, a period during which the HTLV FP treated mice showed only mild clinical symptoms and weight loss (Fig 3C and 3D). Moreover, most of the HTLV FP treated mice exhibited better locomotory behavior than the HTLV Scr treated mice throughout the experiment (S5 and S6 Movies). As disease manifestation was milder compared to the previous experiment, we were able to monitor animals for a longer period of time. In order to determine whether the reduction in EAE severity upon HTLV FP treatment actually results from downregulation of pathogenic MOG35-55-reactive T-cells, spleens were harvested at 26 DPI, cultured ex-vivo, and stimulated using MOG35-55. Stimulation with MOG35-55 resulted in a T-cell proliferative response that was significantly lower in HTLV FP treated spleenocytes compared to HTLV Scr treated spleenocytes (Fig 3E). In addition, HTLV FP treatment resulted in significantly lower IFN-γ secretion and significantly higher IL4 secretion compared to both vehicle and HTLV Scr treated groups (Fig 3F and 3G). These results demonstrate that the HTLV FP modulates antigen-specific T-cell activation in-vivo leading to a downregulation of Th1 and upregulation of Th2 responses.
As HTLV-1 is a T-cell infecting human pathogen [17] and since in this study we show that the HTLV FP inhibits T-cell activation in mice both in vitro and in vivo, we aimed to determine whether this inhibition would apply to human T-cells as well. Hence, human peripheral T lymphocytes were isolated from Peripheral blood mononuclear cells (PBMCs), cultured ex-vivo and activated using CD3 and CD28 antibodies. The effect of HTLV FP treatment on secretion of the T-cell activation marker IL2 was measured by ELISA, as well as the Th1- and Th2-specific cytokines IFN-γ and IL10, respectively. HTLV FP treatment resulted in a significant reduction in IFN-γ and IL2 secretion and an elevation in IL10 secretion, while treatment with HTLV Scr had no effect on T-cell activation (Fig 4). These results demonstrate that the modulatory activity of the HTLV FP is not limited to mouse cells or to cells that recognize a certain antigen and further emphasize its sequence specificity.
To determine whether the transition in cytokine pattern driven by the HTLV FP is indicative of a Th1 to Th2 transition we examined the expression of Th1 and Th2 specific transcription factors, T-bet and Gata3 respectively. C57BL/6J mMOG(35–55)-specific primary T-cells were activated using APCs following HTLV FP treatment. RNA was extracted 24hr following activation and mRNA levels were determined using RT-qPCR. HTLV FP treatment resulted in a reduced TBX21 (T-bet) expression and elevated Gata3 expression (Fig 5A and 5B). We next examined the expression of these genes at the protein level via FACS analysis. C57BL/6J mMOG(35–55)-specific primary T-cells were activated using APCs following either HTLV FP or HIV FP treatment, collected 0, 24, 48 and 72 hours following activation and stained for T-bet and Gata3. Initially, we gated on T-bet expressing lymphocytes (S5A Fig). Since our mMOG(35–55)-specific primary T-cells express basal level of T-bet that is elevated upon activation, we focused on the activated subset of lymphocytes (S5B Fig). The HTLV FP reduced T-bet expression 24 and 48 hours following activation, while the HIV FP had no effect on T-bet expression (Fig 5C and 5D). However, after 72 hours no difference was observed (Fig 5E), though T-bet expression of activated cells diminished in comparison to 24h and 48h (S6 Fig). When gating on Gata3 expressing cells (S7 Fig), we found that the HTLV FP elevated Gata3 expression 24, 48 and 72 hours following activation while the HIV FP had no effect on Gata3 expression (Fig 5F–5H). Taken together, these results suggest that HTLV FP administration downregulates the Th1 response leading to a more Th2-like response.
In order to detect the active segment within the HTLV FP, three peptides, designated HTLV FP5-13, HTLV FP9-22 and HTLV FP14-22 (Table 1), were synthesized based on the network protein sequence (NPS) secondary consensus prediction method [42] (Fig 6A). The HTLV FP5-13 peptide encompasses the helical predicted section of the HTLV-1 FP (S8 Fig) and is located at the same region previously found to be active segment of the HIV FP, both at the 5–13 amino acid section [12]. The HTLV FP9-22 peptide consists of two consecutive repeats of the known GxxxG-like dimerization motif [43, 44], while the HTLV FP14-22 peptide contains only one. These HTLV FP derived peptides were then examined for their ability to inhibit T-cell proliferation. This analysis revealed that both HTLV FP1-33 and FP5-13 are significantly more active compared to the FP9-22 and FP14-22 (Fig 6B). In order to compare the ability of the HIV FP5-13 and HTLV FP5-13 to suppress the induction of T-cell activation at different steps of the TCR signaling cascade, C57BL/6J mMOG(35–55)-specific primary T-cells were activated using either APC, CD3 and CD28 antibodies or PMA and Ionomycin. Similar to the HTLV FP, the HTLV FP5-13 inhibited all three levels of activation (Fig 6C). Peptides were not toxic to T-cells at concentrations used in this study (S1A Fig). In contrast, the activity of the HIV FP5-13 significantly diminished when T-cells were activated using CD3 and CD28 antibodies or PMA and Ionomycin (Fig 6C). The secondary structures of the peptides were then determined in a membrane mimetic environment using circular dichroism (CD) (Fig 6D) and analyzed for structure proportions using CDNN. Both the HTLV FP and HTLV FP5-13 exhibited an α-helical structure while the HTLV FP9-22 and HTLV FP14-22 were found to be random coils (Table 2). This result suggests that the loss of inhibitory activity by the HTLV FP9-22 and FP14-22 might be due to loss of secondary structure.
Next we aimed to determine whether T-cell inhibition by the HTLV FP5-13 occurs within the membrane. For that purpose we utilized a D-enantiomer form of the HTLV FP5-13 (designated HTLV FP5-13D) as interactions of peptides and proteins in the membrane have been shown to be chirality independent [45–47]. We activated our C57BL/6J mMOG(35–55)-specific primary T-cells with APCs and examined their proliferative response following treatment with HTLV FP5-13 and HTLV FP5-13D. Both peptides inhibited T-cell proliferation with the same potency (Fig 6E), suggesting that their active site is within the membrane. Overall, these results indicate that the HTLV FP5-13 is the immune modulatory segment of the HTLV FP and that it acts as an α-helix in the membrane.
HIV-1 utilizes the FP of its gp41 fusion protein to downregulate T-cell activation [11]. Yet, it is unknown whether this ability is shared by other viruses. We utilized the FP of the CD4+ T-cell infecting retrovirus HTLV-1 to explore whether other viral FPs might exhibit immune modulating properties. We reveal that the HTLV-1 gp21 FP is a potent suppressor of T-cell activation, demonstrated by its ability to reduce the onset of the EAE multiple sclerosis (MS) model in mice. Comparing both HIV-1 and HTLV FPs reveals that in contrast to HIV-1, the HTLV FP’s inhibitory effect occurs downstream of the TCR complex and is associated with a decrease in Th1 responses and an elevation in Th2 responses.
Activation of T-cells can be induced in-vitro by antigen presentation either through the TCR itself, downstream of the TCR using CD3 and CD28 antibodies or downstream from the entire TCR complex via PMA and Ionomycin [8, 9]. Here we found that in contrast to the HIV FP, the HTLV FP does not exert its inhibitory effect by targeting the TCR complex. This raises a question regarding the specificity of HTLV-1’s FP inhibitory activity to T-cells, yet, the peptide had no effect on the activation level of mouse primary bone marrow derived macrophages, supporting its specificity T-cells.
In order to elucidate HTLV FP’s mechanism of action, we examined its effect on mRNA expression and cytokine secretion levels of several Th1 and Th2 specific genes that are transcribed upon T-cell activation [33, 34]. The HTLV FP inhibited expression and secretion of Th1-specific cytokines that are crucial for the T-cell antiviral response [48, 49], yet, elevated Th2-specific cytokines [50, 51]. These findings suggest a shift in the Th1/Th2 balance promoted by the HTLV FP. A Th1 response evokes cell-mediated immunity, therefore crucial for the eradication of intracellular pathogens such as viruses. On the other hand a Th2 response controls humoral immunity and evokes antibody responses, which govern the elimination of extracellular pathogens [52]. Skewing the Th1/Th2 balance towards a Th2 response is beneficial for viral persistence within its host and several viruses have been shown to utilize this immune modulation strategy [53, 54]. This is also evident by the findings that some antiviral compounds exert their activity by increasing the Th1 response [55]. Studies indicate that HTLV-1 infection induces IFN-γ production that is aimed to eradicate the virus [56, 57]. As the HTLV FP is exposed during membrane fusion [32], our data suggests that the virus might have the ability to utilize this gp21 region to antagonize this initial anti-viral immune response, thus to better persist within its host. This is in line with evidence showing that HTLV-1 harbors Th1 suppressing factors [58]. Additionally, the ISD that was previously identified within retroviral envelope proteins [25], including the HTLV-1 gp21, has been reported to decrease Th1 and increase Th2 cytokine production [59]. As in the case of the ISD, HTLV FP treatment inhibited IL2 secretion [60]. The ISD is proposed to induce this immune modulation by elevation of cAMP concentration and by inhibition of protein kinase C (PKC) [59, 61]. In this study we show that the HTLV FP significantly inhibits T-cell activation through PMA (PKC activator) and Ionomycin [11], indicating some similarities between the HTLV FP and the ISD derived peptide mechanisms of action. Yet, additional work is required in order to elucidate the exact mechanism of HTLV FP immune suppression.
Th1- and Th2-responses are orchestrated by the specific transcription factors T-bet and Gata3, respectively [62–64]. Therefore, changes in the expression level of these proteins seen in this study further suggest a shift in the Th1/Th2 balance. Yet, though significantly elevated, the fold change of Gata3 expression levels in activated versus non-activated cells was low compared to T-bet as seen by FACS analysis. This suggests that the HTLV FP is not directly elevating Gata3 expression but rather inhibiting T-bet. Since Gata3 expression is negatively regulated by T-bet expression and vice versa [63, 64], it is plausible that the remarkably sharp decrease in T-bet expression is sufficient to cause a prolonged elevation in Gata3 expression. This experimental evidence suggests that by downregulating T-bet expression the HTLV FP disrupts the Th1/Th2 balance, thus elevating Gata3 expression. Yet, other T-cell subsets such as T regulatory cells (Treg) are infected by HTLV-1 [56, 65]. Interestingly, although the development and function of Treg is governed by the master regulator FoxP3 [66], this T-cell subset has been shown to express Gata3 as well [67]. Yet, in contrast to its inhibitory effect on T-bet, Gata3 has been shown to be crucial for Treg function and homeostasis by enhancing FoxP3 expression [68–70] as reviewed in [71]. Interestingly, an HTLV-1 derived factor has been shown to induce CCR4 expression through induction of Gata3 in Treg, promoting T-cell migration and proliferation [72]. Since cell-free HTLV-1 virions are poorly infectious [73, 74] the virus mainly spreads from cell to cell through virological synapses [75]. Thus, promoting T-cell migration is of crucial importance for viral transmission and propagation as it allows infected cells to infiltrate healthy tissues eventually supporting transmission from infected to non-infected cells. Overall, HTLV-1 might utilize its FP to modulate the activity of different T-cell subsets through elevation of Gata3 expression in order to support its persistence within hosts.
As the HTLV FP was shown to downregulate Th1-responses in-vitro, we examine its effect on the induction of a Th1-mediated autoimmune disease in-vivo. EAE is induced by immunization with myelin peptides, such as MOG(35–55), emulsified in CFA, yet, it can be induced by adoptive transfer of myelin-specific CD4+ Th1 cells into naïve recipient mice as well [76–82]. In addition, Stat4 and T-bet, transcription factors in the Th1 differentiation pathway, have been shown to be essential for EAE induction [81, 83–85]. In this study, we demonstrate that the HTLV FP downregulates the transcription of Stat4 and T-bet mRNA, as well as inhibiting mMOG(35–55)-specific primary T-cell activation in-vitro, making EAE an ideal in-vivo model. Here we demonstrate that inhibition of EAE clinical signs by the HTLV FP specifically results from downregulation of pathogenic MOG35-55-reactive T-cells. Interestingly, in humans a small percentage of HTLV-1 infected individuals develop a chronic neuroinflammatory disease termed HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP) [86] that has some pathological similarities to MS. In both cases, lymphocytes that infiltrate the CNS secrete pro-inflammatory cytokines such as IFN-γ and TNF-α [87–89] that can induce neurotoxicity at high concentrations [90, 91]. This results in spinal lesions that initially lead to muscular weakness in the lower limbs [87]. Additionally, soluble TNF-α receptor has been suggested as a common marker for monitoring the progression of these diseases [88] that are both characterized by Th1 predominance [41, 87]. In light of our results, the fact that only a small percentage of HTLV-1 infected individuals eventually develop HAM/TSP might be partially attributed to the neuro-protective nature of the HTLV FP, demonstrated here by its ability to specifically downregulate IFN-γ secretion from pathogenic T-cells in EAE and to alleviate and delay disease onset.
Next, we aimed to identify the immune modulatory region within the HTLV FP. NPS secondary structure prediction analysis [42] predicted the HTLV FP5-13 to be alpha helical. This region was found to be alpha helical in CD analysis and inhibited T-cell activation in contrast to the non-alpha helical regions of the HTLV FP. As we assume that the HTLV FP functions in the membrane, it is likely that an alpha helical structure would support its activity as interactions within the membrane are typically mediated by helix-helix interactions [92–94] through dimerization motifs [95–97], such as GxxxG [30, 98–102]. Hence, we concluded that the 5–13 region is the active segment within the HTLV FP. Interestingly, this is the same region previously identified as HIV-1’s FP active segment [12]. Since HIV-1 and HTLV-1 FPs completely differ in sequence, it seems that possibly through convergent evolution both viruses have obtained the ability to downregulate T-cell activation using the same region of their fusion protein.
The membrane environment holds unique characteristics that allow protein-protein interactions that would not be energetically favored in soluble environment [45], such as interactions between L- and D-enantiomer proteins [47]. Such chirality-independence has been utilized for inhibition of HIV cell-cell fusion by the HIV FP D-enantiomer [46], inhibition of T-cell activation by gp41’s loop derived peptides [103] and for inhibition of Tar receptor mediated chemotaxis in E.coli [47]. In these cases, the L- and D-enantiomers had the same potency. Since FPs are membranotropic regions of viral fusion proteins [104], and as HIV’s FP was specifically shown to target the transmembrane domain of the TCR within the membrane [11], we utilized a D-enantiomer form of the HTLV FP5-13. As both L- and D- peptides were found to inhibit T-cell proliferation with the same potency, we concluded that their active site is situated within the membrane. This is in line with evidence showing that upon binding of the envelope’s surface subunit gp46 to its cellular receptors, the HTLV-1 gp21 FP is exposed and then binds and perturbs the membrane eventually leading to fusion [32, 105].
In summary, our findings indicate that FP mediated T-cell immunosuppression is not unique to HIV, and suggest that it might be a more widespread immune evasion strategy utilized by viruses. Yet, it seems that the HTLV-1 and HIV-1 FPs exert their inhibitory activity on T-cells through different mechanisms thus demonstrating that there are distinct manners by which T-cell activation can be overcome. Our findings demonstrate that the HTLV FP has the capacity to downregulate Th1-mediated antiviral immune response, suggesting that the virus might utilize it for T-cell modulation during fusion. Yet, additional studies using HTLV-1 particles or HTLV-infected cells are required in order to be more conclusive. As the HTLV-1 gp21 FP is known to mediate membrane fusion [106], its ability to modulate T-cell activity highlights how viruses have evolved to alter different cellular processes with limited repertoire of proteins.
C57Bl/6J mice were purchased from Jackson Laboratories (Bar Harbor, ME, USA). All mice were 2–3 month-old when used in the experiments.
Antigen-specific T-cells were selected in-vitro [107] from primed lymph node cells derived from C57Bl/6J mice that had been immunized 9 days before with antigen (100μg myelin peptide, MOG35-55) emulsified in complete Freund’s adjuvant (CFA) containing 150μg Mycobacterium tuberculosis (Mt) H37Ra (Difco Laboratories, Detroit, MI). T-cells were maintained in-vitro in medium containing 500 ml RPMI Ca/Mg + heat inactivated FCS (10% final) + 5 ml 200 mM L-Glu (2 mM final) +5 ml 100 M Na pyruvate (1 mM final) + 5 ml Pen Strep antibiotics + 5 ml Eagle-MEM (Biological Industries, Ref 01-340-1B) + interleukin-2 (IL-2), with alternate stimulation with the antigen every 14 days.
Mouse Femora and tibiae BM cells were isolated from C57Bl/6J mice and cultured in RPMI medium containing FBS (10%), L-glutamine (1%), sodium pyruvate (1%), Pen-strep (1%), and 10 ng/ml recombinant CSF-1 (Peprotech). At day 3, half the medium was replaced, and on day 7, cells were used for in vitro assay, in which 2*105 cells were plated per well in a 24-well plate.
Human peripheral T lymphocytes were isolated from whole blood of healthy adult donors by dextran sedimentation and Ficoll (Sigma) gradient separation followed by depletion of B cells using nylon wool column (Unisorb), to which B cells were adsorbed. Cells were incubated in a complete RPMI growth medium (500 ml RPMI Ca/Mg + heat inactivated FCS (10% final) + 5 ml 200 mM L-Glu (2 mM final) +5 ml 100 M Na pyruvate (1 mM final) + 5 ml Pen Strep antibiotics) for more than 2 h, and then non-adherent cells were harvested and transferred to a new plate, resulting in ~90% CD3+ T lymphocytes. Cells were then used for in vitro assay, in which 105 cells were plated per well in a 96-well plate and activated using CD3 and CD28 antibodies.
Peptides were synthesized using the F-moc solid phase method on Rink amide resin (0.65mmol/gr), as previously described [108]. The peptides were purified by reverse phase HPLC (RP-HPLC) to >95% homogeneity on a C4 or C2 column using a linear gradient of 20–70% acetonitrile in 0.1% trifluoroacetic acid (TFA) for 45 minutes. The peptides were subjected to ESI–MS (electrospray ionization mass spectrometry) analysis to confirm their composition.
Antigen-specific T-cells were plated onto round 96-well plates in medium containing RPMI-1640 supplemented with 2.5% fetal calf serum (FCS), 100 U/ml penicillin, 100 μg/ml streptomycin, 50μM β-mercaptoethanol, and 2mM L-glutamine. Each of the 96 wells contained 104 T-cells, 5x105 irradiated (25 gray) antigen presenting cells (APC), and 5μg/ml of MOG p35-55. In addition, the relevant peptide was added. In order to exclude interaction between the examined peptides and the MOG p35-55 antigen, we added the MOG p35-55 antigen to the APC in a test tube, and in a second test tube we added the examined peptides to the T-cells. After 1 hour, we mixed the APC with the T-cells and incubated them for 48h in a 96 well round bottom plate. Then T-cells were pulsed with 1μCi (H3) thymidine, with a specific activity of 5.0 Ci/mmol, for 24 hours, and (H3) thymidine incorporation was measured using a 96-well plate beta-counter. The mean cpm ± SD was calculated for each quadruplicate. In several experiments, cells were activated with pre-coated CD3 and CD28 antibodies (LEAFTM purified anti mouse clones 145-2-C11 and 37.51, respectively from Biolegend) at final concentration of 2μg/ml, or 50ng/mL of PMA (phorbol 12-myristate 13-acetate) together with 1μM of ionomycin (Sigma Chemical Co, Israel).
Antigen-specific T-cells were plated onto round 96-well plates in medium containing RPMI-1640 supplemented with 2.5% fetal calf serum (FCS), 100 U/ml penicillin, 100 μg/ml streptomycin, 50μM β-mercaptoethanol, and 2mM L-glutamine. Each of the 96 wells had a final volume of 200μl and contained 104 T-cells, 5x105 irradiated (25 gray) spleen cells, as APC, and 5μg/ml of MOG p35-55. In addition, the relevant peptide was added. Each treatment was made with quadruplicate. Analysis of IFN-γ, IL-4 and TNFα secretion was performed by ELISA 24 hours after cell activation according to standard protocols from R&D systems.
Mouse Femora and tibiae BM cells were collected and cultured in RPMI medium containing FBS (10%), L-glutamine (1%), sodium pyruvate (1%), Pen-strep (1%), and 10 ng/ml recombinant CSF-1 (Peprotech). On day 7, cells were stimulated by either (i) LTA, (ii) LPS, or (iii) PAM3CSK4 (1 μg/ml), in the presence of the HTLV FP at 10μM. Media was collected either 5 hours following activation (for TNF-α detection) or 22 hour following activation (for IL-6 detection) and secretion levels were determined according to standard protocols from R&D systems.
Human peripheral T lymphocytes were isolated from whole blood of healthy donors and were incubated in a complete RPMI growth medium (500 ml RPMI Ca/Mg + heat inactivated FCS (10% final) + 5 ml 200 mM L-Glu (2 mM final) +5 ml 100 M Na pyruvate (1 mM final) + 5 ml Pen Strep antibiotics). Cells were activated using CD3 and CD28 antibodies, in the presence of relevant peptides at 10μM. Media was collected 48 hours following activation and secretion of IL2 and IFN-γ was determined according to standard protocols from R&D systems.
Antigen-specific T-cells were plated onto round 12-well plates (106 cells/ well) and activated with 5x105 irradiated (25 gray) APC and 5μg/ml of MOG p35-55 in the presence or absence of relevant peptides. Cells were washed with PBS, blocked (5% Donkey serum, 2% BSA and 0.1% Triton in PBS) and fixed with 4% Paraformaldehyde (PFA) 24 hour following activation. Cells were then stained with Gata3-FITC and T-bet-APC fluorochrome-labeled monoclonal mouse antibodies (purchased from Miltenyi Biotec) according to Miltenyi Biotec protocols. Samples were then collected using LSR-II flow cytometer and analyzed with FlowJo cell analysis software.
EAE was induced in 9-week-old wild type and homozygous C57BL/6 female mice (Harlan Laboratories Israel/ Weizmann Institute animal facilities) by injecting a peptide comprising residues 35–55 of mouse myelin oligodendrocyte glycoprotein (MOG35–55; PolyPeptide Laboratories, Strasbourg, France). Mice were injected subcutaneously above the lumbar spinal cord with 100 μl of emulsion containing 200 μg/mouse of the encephalitogenic peptide in complete Freund’s adjuvant (BD-Difco) enriched with 250 μg/mouse of heat-inactivated Mycobacterium tuberculosis (BD-Difco) at 0 days post-induction (DPI). The HTLV FP was dissolved in PBS and added to the emulsion (1mg/kg). Pertussis toxin (Enzo Life Sciences) at a dose of 300 ng per mouse was injected intraperitoneally immediately after the encephalitogenic injection, as well as at 0 DPI. EAE disease was scored using a five-point grading with 0 for no clinical disease; 1, tail weakness; 2, paraparesis (incomplete paralysis of one or two hindlimbs); 3, paraplegia (complete paralysis of one or two hindlimbs); 4, paraplegia with forelimb weakness or paralysis; 5, moribund or dead animals. The mice were examined daily.
Aliquots of 104 cells were distributed onto a 96-well plate in the presence of 1.25–40μM of the relevant peptides for 16 or 72 hours. Following incubation, XTT reaction solution (benzene sulfonic acid hydrate and N-methyl dibenzopyrazine methyl sulfate, mixed in a proportion of 50:1), was added for 2 hours. Optical density was read at 450-nm wavelength. The percentage of toxicity was calculated relative to the control, 104 cells in medium with no peptide added.
Samples sizes were chosen with adequate statistical power on the basis of past experience and literature. Differences between group means were tested using student’s t-test when the experiment contained two groups, or one-way ANOVA (followed by a Tukey post hoc test) when the experiment contained more than two groups. P< 0.05 was considered significant. Analyses were done using GraphPad Prism (data analysis software) version 6.05. (*P≤0.05, **P≤0.01, ***P≤0.001). Results are displayed as mean ±SEM.
All experiments involving animals were conducted under the approval of the IACUC of the Weizmann Institute, permit numbers: 26980516–3 (in-vitro T-cell activation assays) and 29650816–3 (Experimental Autoimmune Encephalomyelitis), which were performed in accordance to their relevant guidelines and regulations. The facility where this research was conducted is accredited by AAALAC and has an approved Office of Laboratory Animal Welfare (OLAW) Assurance (#A5005-01). The facility operates according to the guide for the care and use of laboratory animals 8th edition by the national research council. All procedures were conducted by trained personnel under the supervision of veterinarians and all invasive clinical procedures were performed while animals were anesthetized.
Human peripheral T lymphocytes were isolated from whole blood of healthy adult donors that provided written informed consent under the regulations and authorization of the Weizmann Institutional Review Board, Project 247–2.
|
10.1371/journal.ppat.1001051 | Contribution of Herpesvirus Specific CD8 T Cells to Anti-Viral T Cell Response in Humans | Herpesviruses infect most humans. Their infections can be associated with pathological conditions and significant changes in T cell repertoire but evidences of symbiotic effects of herpesvirus latency have never been demonstrated. We tested the hypothesis that HCMV and EBV-specific CD8 T cells contribute to the heterologous anti-viral immune response. Volume of activated/proliferating virus-specific and total CD8 T cells was evaluated in 50 patients with acute viral infections: 20 with HBV, 12 with Dengue, 12 with Influenza, 3 with Adenovirus infection and 3 with fevers of unknown etiology. Virus-specific (EBV, HCMV, Influenza) pentamer+ and total CD8 T cells were analyzed for activation (CD38/HLA-DR), proliferation (Ki-67/Bcl-2low) and cytokine production. We observed that all acute viral infections trigger an expansion of activated/proliferating CD8 T cells, which differs in size depending on the infection but is invariably inflated by CD8 T cells specific for persistent herpesviruses (HCMV/EBV). CD8 T cells specific for other non-related non persistent viral infection (i.e. Influenza) were not activated. IL-15, which is produced during acute viral infections, is the likely contributing mechanism driving the selective activation of herpesvirus specific CD8 T cells. In addition we were able to show that herpesvirus specific CD8 T cells displayed an increased ability to produce the anti-viral cytokine interferon-γ during the acute phase of heterologous viral infection. Taken together, these data demonstrated that activated herpesvirus specific CD8 T cells inflate the activated/proliferating CD8 T cells population present during acute viral infections in human and can contribute to the heterologous anti-viral T cell response.
| The majority of humans are infected by herpesviruses, such as Epstein-Barr virus and Human Cytomegalovirus, which rarely cause severe pathology but heavily distort the human T cell repertoire. Up to 20% of cytotoxic T cells can be specific to Epstein-Barr and Cytomegalovirus. It is believed that all these herpesvirus specific T cells are needed to control the persistent infection. However, it has not been explored whether these T cells can contribute to the immune response to a new viral infection. To investigate this possibility, we analyzed the volume of activated virus-specific and total T cells in patients with acute hepatitis B, dengue, influenza and adenovirus infections. We observed that all acute viral infections trigger an expansion of activated T cell population, part of which is specific to infecting agent, and the other part to herpesviruses. Our study provides evidence that persistent herpesvirus infections alter the composition of the T cell population which is activated during new acute viral infection.
| Over the course of the human lifetime, we are exposed to and infected by many different organisms which may be eliminated or may persist. The co-existence of microorganisms in humans is mainly perceived to have negative consequences for health and wellbeing, but examples of potential symbiotic relationship between the host and microbes start to be recognized [1], [2].
Classic examples of microorganisms establishing persistent infections in humans are Epstein Barr virus (EBV) and human cytomegalovirus (HCMV) which are both from ubiquitous herpesviridae family of viruses which infect more than 90% of the human populations. These viruses are associated with the development of specific tumors (i.e. Burkitt's Lymphoma) and they can reactivate with significant pathological consequences in immunocompromised hosts [3], [4]. Nevertheless, in most of the cases, herpesvirus infections are subclinical and well tolerated, even though they cause a robust distortion of T cell repertoire [5], [6] with HCMV and EBV-specific CD8 T cell known to represent up to 20% of total CD8 T cell population [7], [8], [9]. Our inherent effort to maintain such a large population of virus-specific T cells, is seen as a necessity to suppress CMV and EBV reactivation in humans [9], [10], [11]. This would imply that the sole function of herpesvirus specific memory effector CD8 T cells is to act against CMV and EBV infected cells. However, evidence in animal models have shown that effector or memory CD8 T cells can provide immune protection against infection with unrelated intracellular pathogens through production of Interferon γ (IFN-γ) [12]. Such data open the possibility that the large population of HCMV and EBV-specific CD8 T cells present in humans might contribute to the immunological response against other pathogens.
Thus, we set out to evaluate whether CD8 T cells specific for herpesviruses can contribute to the anti-viral T cell response triggered by heterologous acute viral infection in humans. CD8 T cell responses to acute viral infections were analyzed sequentially (from onset to recovery) by measuring the population of activated/proliferating CD8 T cells in patients with acute Hepatitis B Virus (HBV), influenza, dengue and adenovirus infections. The combination of activation and proliferation markers (CD38, HLA-DR, Ki-67 and Bcl-2) expressed by CD8 T cells have been recently proposed to identify the whole population of virus-specific effector CD8 T cells induced by viral infection [13]. These results were obtained in subjects receiving attenuated virus vaccines (Smallpox and Yellow Fever), and activation (CD38/HLA-DR) and proliferation markers (Ki-67/Bcl-2 low) were only expressed by CD8 T cells specific for the vaccine but not by CD8 T cells of different specificities.
In contrast, we demonstrate here that acute symptomatic viral infections trigger an expansion of activated/proliferating CD8 T cell populations of variable sizes, comprising CD8 T cells specific for the infecting virus but these populations are also invariably inflated by CD8 T cells specific for persistent herpesvirus infections. The increased sensitivity of HCMV and EBV-specific CD8 T cells to IL-15 is the likely explanation of this in vivo observation. In addition, HCMV and EBV specific CD8 T cells demonstrate, at the peak of acute infection, an increased ability to secrete IFN-γ suggesting that they might functionally contribute to the heterologous acute anti-viral immunity.
We initially evaluated the size and the expansion kinetics of CD8 T cell population during acute hepatitis B infection. The frequency and quantity of CD8 T cells expressing CD38/HLA-DR and Ki-67/Bcl-2 phenotypic markers was analyzed in 20 patients with acute hepatitis B. Samples were collected at multiple time points from onset of disease (HBsAg+, ALT>1000 U/L) to full recovery (HBsAg- at least 1 month after onset).
A remarkably large expansion of activated CD8 T cell pool was detected. CD38/HLA-DR markers were expressed by approximately a quarter of total CD8 T cells (mean 23%, range 12–68%) at the onset of clinical hepatitis. The frequency of CD38/HLA-DR+ CD8 T cells decreased consistently at the second time point (8–10 days later, mean 12%; range 4–22%) and at the time of recovery it returned to the normal level (mean 3%, range 0.9–10%), detectable in healthy controls (Figure 1, left panel). CD8 T cells co-expressing Ki-67 and low Bcl-2 followed identical kinetics. The peak of Ki-67/Bcl-2 low CD8 T cells was detected at the onset of disease (mean 14%, range 4.5–27% of total CD8 T) and contracted abruptly after 10 days (mean 5%, range 0.8–11%) and at the resolution minimal proliferation was detected (mean 0.8%, range 0.4–1.6%) (Figure 1, right panel).
To analyze whether the population of activated/proliferating CD8 T cells included HBV-specific CD8 T cells, HBV-specific pentamers were used to directly visualize these cells in 5 HLA-A201+ patients. Figure 2 A and B shows that the expression of activation markers of HBV-specific CD8 T cells followed the kinetics of expression of the total CD8 T cell population. HBV-pentamer+ CD8 T cells expressed activation markers and proliferated at the onset of disease but not at the recovery phase (Figure 2 A and B). These results demonstrate that HBV-specific CD8 T cells are represented within the total population of activated/proliferating total CD8 T cells.
Then we tested whether CD8 T cell specific for other common viruses (CMV, EBV, influenza) quantitatively contribute to the total pool of activated/proliferating CD8 T cells. A set of HCMV, EBV or Influenza pentamers (Supplementary Table S1) was used to detect CD8 T cells specific for these common infections. We visualized a sizeable ex vivo frequency of HCMV, Influenza and EBV-specific CD8 T cells in 13 acute hepatitis B patients and their expression of CD38/HLA-DR and Ki-67 was tested at the onset of acute hepatitis and after recovery.
A remarkably different profile of CD8 T cell activation was detected in relation to the CD8 T cell specificity. While influenza-specific CD8 T cells were neither activated (8 out of 8 patients) nor proliferating (5 out of 5 tested patients) at all time points (Figure 3 A and Supplementary Figure S1), HCMV and EBV-specific CD8 T cells were activated (HCMV mean 12.5%; EBV mean 30%) and proliferating (HCMV mean 4.9%; EBV mean 8%) (Tables 1–2 and Figure 3 A) in all the acute HBV patients where such cells were detectable. The expression of CD38/HLA-DR and Ki-67 markers in HCMV and EBV-specific CD8 T cells followed the same expression kinetics of total and HBV-specific CD8 T cells and contracted after recovery as shown on Figure 3 B (patient 12).
The differential phenotype of CD8 T cells specific for different viruses during acute hepatitis B was well represented in a patient (patient 10, Supplementary Figure S1) where the different CD8 T cells specificities co-exist in different activation states. As already shown in Figure 2, at the peak of acute hepatitis, HBV-specific CD8 T cells are mostly activated (77%) and proliferating (65%). At the same time point, a proportion of HCMV-specific CD8 T cells are also expressing activation (20%) and proliferation (6%) markers, while influenza-specific are in a complete resting phenotype. Taken together, these data demonstrated that, at least in acute hepatitis B infection, a sizable proportion of CD8 T cells specific for persistent viruses are activated during acute heterologous infection.
Evidence of activation of unrelated virus specific CD8 T cells has been also reported in HIV infection [14], but our results clearly differ from the ones obtained in attenuated virus vaccine recipients [13], where activation of Influenza, HCMV and EBV was not reported. Thus we tested whether our observation was peculiar to acute HBV infection or whether it represents a common feature in other acute viral infections in human.
Samples from patients with acute Dengue (n 12), Influenza A (n 12), Adenovirus (n 3) infections were collected at the onset of disease (represented in these patients by fever >38° C), after 5–7 days and after recovery (∼21 days) and frequency of CD8 T cell population expressing activation (CD38/HLA-DR) and proliferation (Ki-67) markers was measured (Figure 4 A). Differences in the magnitude and kinetics among diseases with different etiology were found. Adenoviral infection elicited a minimal activation of CD8 T cell population (mean 3.5%), which is only slightly higher than that of healthy individuals (mean of 5 healthy controls 2.4%). In addition, the peak frequency of activated total CD8 T cells in dengue and influenza infections is detected 5–7 days after onset of fever unlike that of HBV, where the peak frequency is seen at the onset of disease (Figure 4 A). These different profiles are compatible with the fact that the onset of disease in acute hepatitis (jaundice) is represented by liver injury and coincides with the peak of adaptive immune response [15], [16] while dengue and influenza infections trigger a strong innate immune reaction (febrile status being a clinical manifestation) and thus, in these infections, full maturation of virus-specific adaptive immunity is expected to peak ∼5–7days after infection.
Nevertheless, despite the lower quantity of total activated CD8 T cells in dengue, influenza, adenovirus patients as well as in 3 subjects with fever of unidentified etiology, the CD38/HLA-DR expression profile on HCMV or EBV specific CD8 T cells was similar to that in acute HBV infection. Figure 4 B summarizes the results obtained in the patients with the indicated pathologies, where a sizeable ex vivo frequency of HCMV or EBV specific CD8 T cells was detected. HCMV and EBV specific CD8 T cells (lower panel) express activation markers to a level even higher to what is detected in the global CD8 T cells populations (upper panel). Unfortunately, the paucity of the cells obtained in these patients didn't allow us to analyze also Ki-67 expression on HCMV and EBV specific CD8 T cells, but overall these results demonstrate that activation of CD8 T cells specific for persistent viral infection (HCMV-EBV) is a constitutive feature of acute anti-viral immunity in human.
Interestingly, we were able to study a patient with acute influenza infection in whom, influenza-specific CD8 T cell expansion didn't coincide with the profile of total activated CD8 T cells (Figure 4 C, P14). In this subject, the influenza-specific CD8 T cells (specific for matrix protein 58–66 epitope) could be visualized only 5 days after onset of symptoms. In contrast, HCMV-specific (pp65 123–131) CD8 T cell frequency was comparatively constant at different time points (1.7% onset; 1.8% +5 days; 1.4% +14 days), and already co-expressed activation markers (30%) at the onset of disease (Figure 4 C). Thus, in this patient, the whole CD38/HLA-DR+ population before the expansion of the CD8 T cells specific for the acutely infected virus seems to be composed of herpesvirus-specific activated CD8 T cells.
We investigated the possible mechanisms of the selected activation of HCMV and EBV- specific CD8 T cells in patients with heterologous acute viral infections.
CD8 T cell cross-reactivity, reactivation of the HCMV or EBV infection and/or activation mediated by cytokines can be implicated in this phenomenon.
Cross-reactivity between HCMV or EBV-specific CD8 T cells with epitopes present in the acute heterologous virus infection seems unlikely. The cross-reactive potential of HCMV-specific CD8 T cells is very uncommon [7] and our data do not support cross-reactive mechanisms either. We could detect activation of CD8 T cells specific for two distinct immediate early and latent EBV epitopes (HLA-B8 RAKFKQLL, BZLF-1 190–197, and HLA-B8 FLRGRAYGL, EBNA-3A 193–201) in an acute HBV patient (Supplementary Figure S2). If cross-reactivity was responsible of this activation, it would require that both epitopes share sequence or structural similarity with HBV virus, an unlikely scenario, based on a sequence similarity search (NCBI PubMed BLAST), which demonstrated no sequence overlap (lowest E value obtained = 11) between these EBV epitopes and HBV proteome.
Reactivation of HCMV and EBV could be a plausible cause, and it might explain why CD8 T cells specific for Influenza are not activated in acute HBV infections. To investigate this possibility, HCMV and EBV DNA levels were tested longitudinally in the serum. However, we did not find any evidence of HCMV or EBV reactivations. HCMV-DNA and EBV-DNA titers were below the level of detection in all patients (HBV, Dengue, Influenza, Adenovirus and fever of unidentified etiology) from the onset of acute heterologous viral infections to recovery (data not shown). Importantly, although HCMV and EBV reactivations are usually associated with the expansion of HCMV/EBV- specific CD8 T cells [10], [17], [18] significant changes in the EBV or HCMV specific T cells quantity were not observed through the course of acute infections (Figure 3 B and Supplementary Figure S2).
We therefore analyzed whether cytokines produced during acute viral infections [19] can be responsible for the differential expression of activation markers by EBV-, HCMV- and influenza-specific CD8 T cells.
PBMC or purified CD8 T cells of healthy subjects containing resting EBV, HCMV and Influenza specific CD8 T cells were incubated with different concentrations of IL-15, IL-2, IL-7, IFN-γ, IFN-α and TNF-α and the expression of HLA-DR and CD38 on EBV, HCMV and Influenza specific CD8 T was analyzed at different intervals (Figure 5 A). We detected that after 24 and 48 hours of incubation, IL-15 (at 1 and 10 ng/ml) induced CD38/HLA-DR expression in HCMV and EBV specific CD8 T cells while the other inflammatory cytokines did not activate EBV and HCMV specific CD8 T cells. Similar to the in vivo findings, influenza-specific CD8 T cells were not or only weakly activated by addition of any of the tested cytokines (Figure 5 A and B). Prolonged incubation times (3 to 5 days) did not alter the activation profile. Figure 5 B shows the results obtained in one healthy subject where HCMV, EBV and Influenza specific CD8 T cells were simultaneously detected. Incubation of total PBMC with IL-15 induces expression of CD38/HLA-DR molecules in EBV and HCMV specific CD8 T cells (44% and 37% respectively) but only in few influenza-specific CD8 T cells (7%). The specific effect of IL-15 on HCMV and EBV-specific CD8 T cells was confirmed in other healthy subjects where individual specificities were detected (HCMV = n4; EBV = n4). Similar results were obtained incubating total PBMC or CD3+ CD8+ purified cells (not shown). Thus, IL-15, a cytokine that has been shown to induce T cell activation in mice [20] and human [21], [22] and is known to be produced during acute viral infections ([19] and personal data) induces preferential CD38, HLA-DR up-regulation of HCMV and EBV-specific CD8 T cells rather than influenza-specific ones.
Having observed that a proportion of HCMV and EBV-specific CD8 T cells are activated during heterologous acute viral infection, we sought to analyze their functional profile. The limited quantity of cells available in patients with acute viral infections precludes an extensive evaluation of the functional profile directly in our patient sample. Thus, since IL-15 mimics the differential activation state of HCMV, EBV and influenza-specific CD8 detected in patients with acute viral infections, we performed a series of functional experiments using PBMC of healthy individuals activated with IL-15.
We first tested whether IL-15 can differentially trigger T cell activation in HCMV, EBV and Influenza specific CD8 T cells in vitro. PBMC of healthy individuals were incubated with or without IL-15 for 48 hours and HCMV, EBV and influenza-specific CD8 T cells were tested for their ability to produce anti-viral cytokines (IFN-gamma, IL-2 and TNF-alpha) using intracellular cytokine staining. Note that the cytokines measurement on CD8 T cells specific for the different viruses requires their visualization with the specific HLA-class I/peptides pentameric complex (pentamers). The pentamer staining can potentially trigger T cell stimulation through direct interaction of the TCR with the synthetic MHC-class I peptide complexes of the pentamers[23], [24], [25]. Thus, to distinguish whether IL-15 can directly trigger T cell activation (TCR-independent stimulation) or enhance the T cell activation triggered by MHC/peptide pentamer (TCR-dependent stimulation), the intracellular production of IFN-gamma on CD8 T cells was analyzed adding the MHC/peptide pentamers either before (TCR-dependent stimulation) or after (TCR-independent stimulation) the incubation time of intracellular cytokine staining. A schematic representation of the experimental design is presented in Figure 6 A.
In accordance with previous studies [21], [22], IL-15 elicited a spontaneous production of IFN-gamma on T cells. However, the level of IFN-gamma production was modest and present in CD8 cells irrespective of their specificity. Dot plots displayed in Figure 6 A illustrate these results obtained in one representative subject. Increased production of other cytokines (IL-2, TNF-alfa) was less striking (not shown).
In contrast, we observed that HCMV and EBV-specific CD8 T cells incubated with IL-15 and stained with MHC/peptide pentamer at the beginning of the intracellular cytokine assay showed an increased ability to produce IFN-gamma. More than 70% of IL-15 pulsed HCMV and EBV-specific CD8 T cells produced high quantity of IFN-gamma while in the absence of IL-15, MHC-pentamer staining stimulate only a minority of HCMV and EBV-specific CD8 cells (Figure 6 A–B). Importantly, IL-15 incubation has a modest effect on Influenza specific CD8 T cells (Figure 6 A–B). Thus, our in vitro experiments showed that IL-15 is not only able to preferentially activate HCMV and EBV-specific CD8 T cells, but can also modulate their functional responsiveness to the TCR-dependent stimulation mediated by MHC-pentamer staining.
Having defined a different functional profile on in vitro activated HCMV and EBV-specific CD8 T cells, we tested whether such features could be detected in vivo.
In line with the experiments in vitro, MHC-peptide pentamer stimulation was detected preferentially on HCMV and EBV-specific CD8 cells present during the acute phase of HBV infection (Figure 6 C and D).
Figure 6 C shows the results obtained in a representative patient (P24) with acute hepatitis and with a sizeable population of activated HCMV-specific CD8 T cells (20%). While the spontaneous production of IFN-gamma was identical in HCMV-specific CD8 cells present at the onset and at recovery of acute hepatitis B (Figure 6 C- unstimulated), 12% of HCMV specific CD8 cells present at the onset of acute hepatitis B against only 4% of the ones present at recovery produced IFN-gamma after MHC-pentamer stimulation (Figure 6 C). In addition to the higher frequency of IFN-gamma producing cells, the amount of the cytokine produced during the onset was higher than that during the resolution, as visualized by the difference in mean fluorescence intensity (MFI) (1152 at onset and 507 resolution).
Figure 6 B shows the cumulative results obtained in 6 subjects with detectable HCMV (P9, P10, P7, P24), EBV- (P13) and influenza-specific CD8 (P24, P11) at the onset and recovery of acute hepatitis B. Bars indicate the % increase of IFN-gamma producing CD8 T cells at onset of acute hepatitis in comparison with recovery.
Thus, persistent virus specific CD8 T cells produce more anti-viral cytokines after TCR-mediated activation during acute phase of heterologous viral infection.
We demonstrate here that activation of CD8 T cells specific for persistent viral infection (HCMV-EBV) is a constitutive feature of acute anti-viral immunity in human.
Our conclusions differ from the ones obtained in attenuated virus recipients [13], which have suggested that activated (CD38/HLA-DR+) and proliferating (Ki-67+) CD8 T cells are exclusively constituted of CD8 T cells specific for the acutely infecting virus. However, our patients with activated/proliferating HCMV and EBV responses had a symptomatic viral infection with a high level of inflammation, whereas those subjects vaccinated with attenuated viruses, by definition, should not exhibit any pathology of acute infection.
Of note, the presence of activated HCMV and EBV specific CD8 was also detected during other pathological human viral infections [14], [26], [27] further supporting our conclusion that activation of CD8 T cells specific for persistent infection is a consistent phenomenon during symptomatic viral infections.
In contrast to HCMV and EBV-specific CD8 cells, we observed that CD8 T cells specific for influenza were not activated during the acute phase of heterologous acute viral infection. Thus, our data show that memory CD8 cells specific for persistent and non-persistent viruses not only differs in term of phenotypic profile in healthy individuals [28], but respond differently to the pathological condition triggered by an heterologous acute viral infection.
We can only speculate about the causes of the variable behavior of CD8 T cells specific for the different pathogens. A plausible explanation is that, while influenza-specific CD8 T cells are true memory CD8 cells without any recent encounter to their specific ligand, EBV and HCMV specific CD8 cells might experience a continuous or repetitive exposure to the specific antigens.
The accumulation over time of herpesvirus-specific CD8 T cells in healthy subjects [29], [30] and work in animal model of HCMV infection [31], have suggested that EBV and HCMV antigens are constantly available for T cell stimulation. The Ag-exposure might modulate the functional state of HCMV and EBV-specific CD8 cells and program them to respond to cytokines produced during acute viral infections. The differential functional state of herpesvirus specific CD8 T cells when compared with influenza-specific was confirmed by our in vitro data. We clearly demonstrate that IL-15 triggers in vitro the activation/proliferation of HCMV, EBV specific rather than influenza-specific CD8 T cells.
Based on these in vitro data, we favor the idea that the detection of activated/proliferating HCMV and EBV specific CD8 T cell is mediated principally by the presence of IL-15 during acute phase of viral infections. This makes HCMV/EBV reactivation or indeed cross-reactivity a less likely explanation for this phenomenon. However, it is important to stress that this causative link is hypothetical since the level of IL-15 required to activate HCMV/EBV in vitro (1–10 ng/ml) is higher than what we detected in the serum of the patients in this study (always lower then 50 pg/ml in any viral infection, data not shown). Such inconsistency should be taken into account, even though the serum cytokine levels cannot define their actual concentrations in the target organ or lymph node.
We cannot exclude that a reactivation of EBV/HCMV infection is occurring in our patient population and thus directly driving the HCMV or EBV-specific CD8 T cell activation. We couldn't demonstrate any virological evidence of HCMV and EBV reactivation, but the negative virological tests do not exclude a HCMV and/or EBV viral reactivation is present elsewhere outside the blood compartment and is immediately curtailed by activated HCMV and EBV specific CD8 T cells. A similar scenario was suggested to occur in patients with acute Hantavirus infection where an increased EBV-DNA titers were found only in subjects without measurable EBV-specific T cell response [26]. However, it has been reported that HCMV and EBV reactivation is associated with the expansion of HCMV/EBV specific CD8 T cells [10], [17], [18], which was not observed in any of our patients (Figure 3 B and Supplementary Figure S2).
What appears clear from our data is that the contribution of the activated/proliferating HCMV/EBV specific CD8 T to the size of activated total CD8 T cells is not negligible, but at the contrary can alter the quantitative measurement of anti-viral CD8 T response during acute viral infections. A mean of, respectively, 30% and 12.5% of EBV and HCMV-specific CD8 T cells express activation markers during the acute phase of different viral infections and since the combined population of both HCMV-EBV specific CD8 T cells might exceed 20% of total CD8 T cells [7], [9] it is plausible to conclude that EBV/HCMV-specific CD8 T cells can inflate the number of total activated CD8 T cells.
The presence of activated/proliferating CD8 T cells specific for HCMV and EBV during the early phases of different acute viral infection raises several questions. First, it will be interesting to evaluate whether CD8 T cells specific for other persistent viruses (i.e. HSV1 and 2) can actually behave like HCMV or EBV specific CD8 T cells and thus further contribute to the anti-viral CD8 T cell acute response.
A further question might address the biological significance of the herpesvirus specific CD8 T cell activation during heterologous acute viral infections.
There is a possibility that the activation/proliferation state of HCMV/EBV specific CD8 T cells counteracts the potential attrition exerted by the expansion of CD8 T cells specific for the acutely infecting virus [32] and therefore might be important for preventing the reactivation of HCMV/EBV infection. In this regard, our data differ from previous reports in acute HBV infected patients [33], since we did not observe any significant loss of HCMV or EBV specific CD8 T cells. On the contrary, HCMV and EBV-specific CD8 T cell frequency was remarkably constant during the different phases of acute heterologous viral infections and the observed mild proliferation of HCMV and EBV specific CD8 T cells (Figure 3 A and Supplementary Figure S2) might represent a compensatory mechanism counteracting the attrition exerted by the expansion of CD8 T cell specific for the acutely infected virus [30], [34].
In addition, the observation that activation of HCMV and EBV specific CD8 T cells present during the acute phase of heterologous viral infections is associated with a functional increase in the MHC-pentamer mediated CD8 T cell activation further supports the idea that such events might have a broader biological significance.
We can only speculate about the physiological significance of the increased MHC-pentamer mediated CD8 T cell activation. However, a plausible interpretation is that the HCMV and EBV-specific CD8 cells during acute heterologous viral infection are less dependent to possible co-stimulatory effect mediated by additional molecules provided by their target during T cell recognition. Alternatively, the increased response to pentamer-mediated staining might indicate a lower requirement of MHC-class I complexes necessary for T cell activation [24]. These various possibilities will need further investigation, but what our data clearly demonstrate is that functional differences in the ability to produce IFN-gamma are present in different phases of heterologous acute viral infection.
The increased likelihood of activated HCMV, EBV specific CD8 T cells to produce antiviral cytokines after recognition of HCMV and EBV antigens might be beneficial not only in the control of HCMV/EBV reactivation but can actively contribute to the global anti-viral immune response. Evidences in animal model have already shown that T cell activation of non-antigen specific T cells can contribute to the early response against pathogens [12], [35].
On the other hand, the detected hyper-responsiveness of HCMV and EBV- specific CD8 T cells can have an impact on immunopathogenesis of the viral infections [36]. Heterologous immunity have been observed to alter pathogenesis of different viral diseases [37], [38].
In conclusion, we show that the CD8 T cell population activated during acute viral infection is not constituted exclusively by CD8 T cells specific for the newly infected virus. On the contrary, this population is inflated by the presence of activated T cells specific for herpesvirus, directly demonstrating the ability of persistent virus infections to leave a functional imprint on the acute anti-viral T cell response in humans with functional consequences that will require further elucidation.
Samples were taken from patients or healthy volunteers attending clinics in Singapore (Dengue, Influenza, Adenovirus infections, fevers of unidentified etiology and healthy volunteers) and Italy (HBV infection). Local Review board and Ethical Committees approved the study. Total number of patients is 50: HBV 20, Influenza 12, Dengue 12, Adenovirus 3, patients with fevers of unidentified etiology 3. Number of healthy volunteers enrolled is 5. Age of the subjects ranged from 20 to 54 years old. Patients were selected on the basis of fever >38°C (Dengue, Influenza, Adenovirus) or jaundice (HBV). Diagnosis of dengue (detection of dengue virus by PCR), influenza A (+ isolation of influenza A from nasal swab), adenovirus (isolation of the virus from nasal swab), and HBV (HBsAg +, anti-HBc IgM+ and HBV-DNA+) was performed within 5 days from selection. Acute hepatitis B patients were all HBsAg+ and had ALT>1000 U/L, at the disease onset. All healthy volunteers were asymptomatic. Peripheral blood mononuclear cell isolation from whole heparinized or EDTA blood with Ficoll-Hypaque was performed within 4 hours of drawing. PBMC were analyzed immediately or frozen for subsequent analysis.
HBV patients: HBsAg, HBeAg, anti-HBs, anti-HBc IgG and IgM, anti-HBe, anti-HDV, anti-HCV, anti-HIV-1 and -2 were determined by commercial enzyme immunoassay kits (Abbott Labs, IL, USA; Ortho Clinical Diagnostic, Johnson & Johnson, DiaSorin, Vercelli, Italy). HBV-DNA was quantified by PCR (Cobas Amplicor test; Roche Diagnostic, Basel, CH) and CMV-DNA was quantified with artus-CMV-LC PCR (Qiagen, Qiagen Gmbh, Hilden), EBV-DNA was tested with EBV R-gene DNA extraction and quantification kit (Argene, Varilhes, France). Dengue detection was performed by RNA isolation from serum samples using RNA extraction kit followed by reverse-transcription into cDNA (Superscript III First Strand kit, Invitrogen, California, USA). The cDNA was PCR amplified for detection of the virus, for determination of serotype and for quantification of viral load as previously published [39]. The serum and nasal swab samples were tested for the presence of Influenza A and Adenovirus using RT-PCR (Superscript III First Strand kit, Invitrogen, California, USA) and direct immunofluorescence assay on nasal swabs (Light Diagnostic Influenza A antibody FITC reagent, Millipore, Billerica, MA). Amino acid sequence alignment was done using BLAST from NCBI PubMed (http://blast.ncbi.nlm.nih.gov/Blast.cgi).
HLA-peptide pentameric complexes (pentamers) were purchased from Proimmune (Oxford, UK). Anti-CD8 (PE-Cy7 and APC-Cy7), anti-CD3 (perCP and perCP-Cy5.5) anti-CD38 (APC), anti-HLA-DR (Pe-Cy7), anti-KI-67 (FITC and PE), anti-Bcl-2 (FITC), anti-IFN-γ (FITC and APC), anti-IL2 (FITC, PE and APC), and isotype control antibodies were purchased from BD Biosciences, San Jose, CA.
Titrated pentamers (PE) were added to 50 µl of purified PBMC (2×106 cells total) for 15 min at 25°C in the dark, washed and then panel of titrated antibodies for surface markers were added to pentamer stained or total PBMC. The cells were then fixed and permeabilized using Cytofix/Cytoperm solution (BD Biosciences, San Jose, CA). After washing, intracellular staining was performed for intracellular markers (Ki-67, Bcl-2). Cells were then washed 3 times, and fixed with 1% formaldehyde before acquisition on a FACS Canto flow cytometer. Compensation was checked regularly using FASC Diva software. Compensation controls were individually determined for each experimental setup.
PBMC were stained with the relevant pentamers (MHC/peptide pentamer stimulated), or left unstained (unstimulated), washed and then incubated for 5 h with 10 µg/ml brefeldin A (Sigma-Aldrich, St. Louis, MO). Following incubation, the unstimualted cells were stained with relevant pentamers and MHC/peptide pentamer-stimulated were left in PBS. Then cells were stained with anti-CD8 and anti-CD3 mAbs for 20 min at 4°C then fixed and permeabilized using Cytofix/Cytoperm solution. Finally, cells were stained with anti-IFN-γ and anti-IL-2 for 30 min on ice, washed, and fixed with 1% formaldehyde before acquisition on a FACS Canto flow cytometer. For analysis of anti-virus-specific CD8 T activation in vitro, freshly isolated PBMC or purified CD8+ T cells were incubated in vitro at 2×106/ml with or without cytokines (IL-7, IL-2, IL-15, IFN-γ, IFN- α, TNF- α, purchased from RnD Systems, Minneapolis, MN). The cells were collected at indicated time points, and the intracellular cytokine staining was performed as described above.
This study was conducted according to the principles expressed in the Declaration of Helsinki. The study was approved by the Institutional Review Board of Singapore National Healthcare Group Ethical Domain and Azienda Ospedaliera Universitaria di Parma Ethical Committee hospitals. All patients provided written informed consent for the collection of samples and subsequent analysis.
|
10.1371/journal.pntd.0002521 | Laboratory-Based Prospective Surveillance for Community Outbreaks of Shigella spp. in Argentina | To implement effective control measures, timely outbreak detection is essential. Shigella is the most common cause of bacterial diarrhea in Argentina. Highly resistant clones of Shigella have emerged, and outbreaks have been recognized in closed settings and in whole communities. We hereby report our experience with an evolving, integrated, laboratory-based, near real-time surveillance system operating in six contiguous provinces of Argentina during April 2009 to March 2012.
To detect localized shigellosis outbreaks timely, we used the prospective space-time permutation scan statistic algorithm of SaTScan, embedded in WHONET software. Twenty three laboratories sent updated Shigella data on a weekly basis to the National Reference Laboratory. Cluster detection analysis was performed at several taxonomic levels: for all Shigella spp., for serotypes within species and for antimicrobial resistance phenotypes within species. Shigella isolates associated with statistically significant signals (clusters in time/space with recurrence interval ≥365 days) were subtyped by pulsed field gel electrophoresis (PFGE) using PulseNet protocols.
In three years of active surveillance, our system detected 32 statistically significant events, 26 of them identified before hospital staff was aware of any unexpected increase in the number of Shigella isolates. Twenty-six signals were investigated by PFGE, which confirmed a close relationship among the isolates for 22 events (84.6%). Seven events were investigated epidemiologically, which revealed links among the patients. Seventeen events were found at the resistance profile level. The system detected events of public health importance: infrequent resistance profiles, long-lasting and/or re-emergent clusters and events important for their duration or size, which were reported to local public health authorities.
The WHONET-SaTScan system may serve as a model for surveillance and can be applied to other pathogens, implemented by other networks, and scaled up to national and international levels for early detection and control of outbreaks.
| Shigellosis causes dysentery and kills an estimated 1.1 million people per year worldwide, 60% of them children under the age of 5. The infectious agent is Shigella spp, transmitted from person to person by fecal-oral route or via ingestion of contaminated food or water. Having a system for early detection of outbreaks would be very useful for implementing control measures that help reduce the number of affected patients, economic losses and prevent the dissemination of antimicrobial resistance. We present the application of a space-time permutation scan statistic implemented within the free SaTScan software for laboratory based surveillance of Shigella cases in six provinces from Argentina. SaTScan was applied on the data loaded into WHONET databases (an electronic laboratory data system used world-wide) in the six provinces from April 2009 to March 2012. The project allowed the identification of 32 events, including several of particular public health importance for their duration or number of affected patients. It also strengthened the relationship between the laboratory and epidemiology staff. In conclusion, the combination of WHONET laboratory data and SaTScan analysis can detect important community outbreaks of antimicrobial-resistant shigellosis in a timely manner, to make a difference to public health.
| In view of the increasing movement of people, animals, and food products around the globe, new strategies and collaborations are urgently needed to detect the emergence of microbial threats and implement effective control measures. National and regional electronic laboratory-based surveillance collaborations based on routine clinical laboratory test results, as recommended by the WHO-Global Foodborne Infections Network [1] and the WHO-Advisory Group on Integrated Surveillance of Antimicrobial Resistance [2], offer the potential for real-time monitoring of evolving microbial populations. Sophisticated technologies for differentiating among strains and for processing information have proliferated and been incorporated into surveillance [3]. However, advances in organizational aspects such as timeliness of data entry and analysis and integration of local site-of-care laboratories into national and international surveillance networks have developed more slowly [4]. Statistical analysis of laboratory data for the detection of disease outbreaks in the community or in hospitals has also lagged, and existing statistical approaches have tended to focus on temporal trends [5], [6], [7], [8], largely ignoring the geographic component of pathogen population dynamics.
To be most useful for public health purposes, laboratory-based surveillance should 1) be specific, i.e. be capable of distinguishing (a) among species and preferably variants within species and (b) among antimicrobial resistance profiles within those taxonomic groups; 2) have timely electronic data entry; 3) integrate multiple laboratories using uniform protocols and a uniform database; 4) be linked to and used by the agencies responsible for disease control; and 5) implement statistical methods for detecting departures from background levels in both time and space, rather than relying solely on visual inspection of data.
Of the bacterial pathogens causing diarrhea, Shigella spp. is one of the most prevalent and most consistently associated with dysentery and persistent diarrhea [9]. Shigellosis kills an estimated 1.1 million people per year worldwide, 60% of them children under the age of 5 [10], and can result in reduced growth in children who survive. Shigella species appear highly adaptable to selective pressure and have developed resistance to a number of antimicrobials with patterns of resistance varying temporally and geographically with antimicrobial usage patterns [11], [12], [13], [14]. Highly resistant clones of Shigella have emerged in Argentina [15], [16], [17]. Recently, the unique Shigella flexneri serotype X variant, which emerged in China in 2001, has rapidly spread, including through Argentina [18], undergoing frequent serotype switching and acquiring resistance to multiple antimicrobials in the process [19].
We report here our experience with an evolving, integrated, laboratory-based, near real-time surveillance system now operating in six contiguous provinces of Argentina, building on a prior retrospective study [20]. It represents real-world surveillance rather than a static system pre-defined in a formal protocol. This system includes all of the desired laboratory-based surveillance system elements listed above and may serve as a model for surveillance not only of Shigella spp. but of other community-acquired pathogens in Argentina and elsewhere.
The microbiology data used for shigellosis prospective surveillance came from a subset of the national Argentine network for monitoring antimicrobial resistance, WHONET- Argentina, which was established in 1986 by the Ministry of Health through the Dr. Carlos G. Malbrán National Institute of Infectious Diseases (INEI). The group is named after WHONET, a free software for the management of microbiology laboratory data developed by the WHO Collaborating Centre for Surveillance of Antimicrobial Resistance promoted by the World Health Organization. Led by the National Reference Laboratory at INEI, WHONET-Argentina currently includes 89 clinical laboratories representing all geographic jurisdictions and captures detailed data on human pathogens and their susceptibility profiles from routine diagnostic specimens.
WHONET-Argentina hospitals were selected to participate in this surveillance initiative on the basis of the completeness and timeliness of data entered into the national WHONET database in previous years, a commitment to send updated data on a weekly basis, and Shigella serotyping ability. An additional criterion was the strength of the respective local public health systems in integrating laboratory, epidemiology, food science, and environmental health efforts for the investigation of outbreaks and sources of infection.
On the other hand, surveillance of Shigella infections at the national level is currently conducted by provincial hospitals that report the number of cases weekly to the Ministry of Health.
The period covered by this report was from April 1, 2009 through March 31, 2012. The number of laboratories and provinces participating increased during this period (Table 1), with growth in laboratories' historical data, capacity for complete and timely data entry, and ability to meet the other inclusion criteria. For the first year, seven hospitals in the three contiguous provinces of La Pampa (LP), Neuquén (NQ), and Río Negro (RN) participated. In March 2010, four satellite clinics in the same three provinces were added. In January 2012, 12 additional laboratories from the three additional contiguous provinces of Córdoba (CBA), Mendoza (MZA), and San Luís (SL) were incorporated (Table 1). Thus, by the end of the evaluation period, 23 laboratories in six provinces were included (Figure 1).
To detect localized shigellosis outbreaks in near real-time, we used the non-parametric space-time permutation scan statistic [28], as was previously used in a pilot study of historical data in the WHONET-Argentina database . The method searches for statistically significant clusters of Shigella cases in space and time, using cylindrical scanning windows with a circular base of variable location and radius representing geographical space and a variable height representing the number of days in a potential cluster (ending on the day for which the analysis is being done). It does not require population-at-risk data and makes minimal assumptions about the time, location, or size of outbreaks. The method adjusts for any purely geographical variation in disease incidence, whether due to urban vs. rural, north vs. south, dry vs. humid conditions, etc. There is no need to adjust for any of these variables explicitly. Rather, the adjustment is done non-parametrically, using a permutation-based approach that is conditional on the total area counts summed over all days. Similarly, the method adjusts for any purely temporal trends in the data, such as seasonal variation or day-of-week effects. This is done non-parametrically, by day, by using a permutation-based approach that is conditional on the total daily counts summed over all geographic areas. With a seasonal infection such as shigellosis, this ability to automatically adjust for seasonal variation is critical; without such an adjustment it would be difficult to distinguish between epidemiologically important outbreaks and the expected annual increase in shigellosis cases observed each summer. In contrast to other cluster detection methods that adjust for seasonal variation using a parametric model, the space-time permutation scan statistic does not require multiple years of historical data. The paper by Kulldorff et al. [28] describes the method in more detail.
In looking for clusters, we set the maximum temporal cluster length at 30 days, meaning clusters of 1, 2, 3, and up to 30 days' duration could be detected. Because data were updated on a weekly rather than daily basis and because of potential delays in data entry, data transmission, or data availability (e.g. in the ascertainment of organism serotype), prospective analyses were run not only for the last date in the dataset but also for each day in the prior several weeks to ensure that recent clusters were not missed. In each of these day-specific analyses, the prior 365 days of data were used as a historical baseline. From April 2009 to August 2011 we used SaTScan 7.0.1 to search for single hospital clusters; from September 2011 through March 2012, we used SaTScan 9.1.1 to search for single or multi-hospital clusters with a maximum radius of 200 kilometers (Table 1) (Clusters of that apparent or real size could occur if a contaminated food were distributed regionally or if an area of transmission were located between two participating hospitals, leading some patients to go to one hospital and others to the other).
Per analysis run, the statistical inference is adjusted for the multiple testing inherent in the many potential cluster locations, sizes, and time lengths, and is expressed in terms of a recurrence interval [29]. If a detected group of cases is determined to have a recurrence interval of 400 days, for example, this could be an event of epidemiologic significance. But it could be simply due to chance -during any 400-day period, the expected number of signals of that strength or stronger is one, when the null hypothesis of no clusters is true. Thus, the higher the recurrence interval, the less likely that the observed clustering could be attributed to random variation. In this study, we considered a grouping of cases with a recurrence interval of 365 days or longer to be a statistical “signal” and worth communicating (after basic data quality checking) to relevant local or provincial authorities for possible epidemiologic and molecular investigation. Using 365 days as our recurrence interval threshold for notifying public health authorities means that, on average, we could have expected to see one false positive signal per year within each analysis level (genus, species, serotype, and resistance profile). Since statistically significant space-time groupings of cases may be caused by chance or by organizational or procedural factors such as changes in hospital participation, specimen collection practices, or laboratory testing procedures, it is important that statistical signals be evaluated through traditional epidemiologic means before concluding that they are indications of true disease outbreaks.
Calculations for the space-time permutation scan statistic were done using the free SaTScan™ software [30], as imbedded in the free WHONET software.
In the rest of this paper, we use the term “signal” to refer to the detection by SaTScan of a group of cases clustered in space and time with a recurrence interval of 365 days or more. We use the term “event” to refer to a group of signals overlapping in space and time, which may represent a single potential disease outbreak. The term “cluster” is used generically, referring to a group of cases regardless of whether or not these cases are truly related to each other epidemiologically.
When a signal was detected in weekly analyses, a report including data on each patient was sent by INEI analysts to laboratory and epidemiology personnel in the affected hospital and province. These messages were usually sent on Fridays, but later if there were delays in analysis.
The decision by local authorities of whether to investigate a signal depended in part on the recurrence interval and in part on other criteria such as number of patients, location details, and timing and specificity of the signal (homogeneous clusters detected by resistance phenotype or serotype were considered to be more reliable than heterogeneous clusters detected at the Shigella genus level). Since the purpose of this real-world surveillance system was to support local public health, no uniform signal investigation protocol was imposed; investigations variously included patient or clinician surveys, additional sampling of either contacts or suspected sources (food or water) for Shigella spp., and trace-back of suspected food products.
The events identified by SaTScan confirmed as outbreaks with investigation by local authorities and the confirmation of strains' relatedness by PFGE were analyzed and informed to National Surveillance System.
There were 32 statistically significant events: 2, 12, 3 and 2 of S. flexneri serotypes 1, 2, 3, and AA 479 respectively; 11 of S. sonnei; and 2 of S. boydii serotype 2 (Table 2 and Table 3). Seventeen of the 32 events were found at the resistance profile level (and some at higher levels also). All 6 participating provinces had events. During the study period, 21 of the events occurred primarily during the Argentine summer months of October–March, while 11 occurred primarily in April–September. The median event duration was 48.5 days (minimum: 3, maximum: 94). The median number of patients in a signal was 21.5 (minimum: 2, maximum: 41).
Twenty-six of the 32 events were investigated further by PFGE analysis, which confirmed a close relationship among the isolates (with the first and second enzyme) for 22 (84.6%) of the 26. In contrast, in Events 10, 11, 12, and 28 (all involving S. sonnei, PFGE showed diverse genetic subtypes, and we considered these events to be largely chance concentrations of cases, not potential diseases outbreaks. Seven of the events, Events 5, 7, 14, 16, 17, 31, and 32, were investigated epidemiologically, which revealed links among the patients (Table 4), consistent with the PFGE findings for these events.
Twenty-six of the 28 events considered to represent or possibly represent disease outbreaks were detected before hospital staff was aware of any increase in the number of Shigella isolates. The others were Events 14 and 17. In Event 14, S. flexneri 2 in Rio Negro, the hospital bacteriologist suspected the outbreak before the appearance of the first WHONET- SaTScan signal, because the majority of isolates were from patients from the same family. Event 17 was a cluster of S. flexneri 3 associated with a wedding in Neuquén province, which was attended by 150 people, including some from Chile. More than half the participants became ill, and this outbreak was reported immediately to public health authorities at both national and international levels before the laboratory results could be incorporated into the WHONET database and analyzed.
The system detected events of public health importance. For example, Events 16 and 21, both in Rio Negro, involved S. sonnei resistant to both AMP and SXT, an infrequent resistance profile. A number of long-lasting and/or re-emergent clusters were also detected, represented by 4 pairs of related events: PFGE patterns were very similar for Events 1 and 6, S. flexneri 2 in La Pampa (Figure 2), and for Events 13 and 14, S. flexneri 2 in Rio Negro. Events in each pair were separated from each other by at least 3 months. Events 10 and 11 of S. sonnei in La Pampa were separated by one month and included two persistent patterns and single subtypes circulating simultaneously. Events 16 and 21, S. sonnei with the unusual AMP-SXT resistance phenotype, were centered in two cities (Viedma and Bariloche) in Rio Negro province. The predominant PFGE patterns in the latter two events differed by only one band, which is not considered a significant difference in PFGE analysis for Shigella when person-to-person transmission is a prominent feature and the outbreak persists in time [26] Event 16 lasted from December 2010 to March 2011, while Event 21 occurred in April to May 2011.
Two other events are worthy of mention due to their duration or size. Event 7, caused by SXT-nonsusceptible S. sonnei in La Pampa province, lasted from December 2009 to March 2010, with 34 isolates. Among 23 cases of diarrhoea studied trough clinician surveys, 14 (60.9%) were epidemiologically linked. Seven of 14 SXT-non-susceptible S. sonnei isolates analyzed by PFGE shared a new pattern in the national database and the other 7 were closely related to this pattern with similarities between 91.4% to 97.4%, 1 to 3 bands of difference. (Figure 3). No common source was identified, but most of the cases were associated with two neighboring households found to be epidemiologically linked and to have deficient sanitary conditions. Figure 4 shows the time series of cases, for the period from December 2009 to March 2012, to highlight the detection and evolution of event 7. Event 17, the wedding outbreak in Neuquen province, was noteworthy in that it was detected by our surveillance system on the basis of only 7 isolates when in fact the total number ill, according to public health investigation, was closer to 75. The 7 S. flexneri 3 isolates from the patients showed indistinguishable PFGE patterns.
In general, no common source could be confirmed in the events, even though food and water samples were analyzed in several instances; this may be due to the difficulty for the isolation of Shigella from this kind of sample. Nevertheless, the epidemiologic studies could determine sanitary deficient conditions and probable routes of transmission, mainly from person to person. Furthermore, deficiencies in the conditions for food conservation and elaboration were identified in some events. On this basis, control and prevention measures included recommendations on hygiene and food handling, as well as a notification to the International Health Regulation for an event that affected patients from Chile and Argentina (Table 4).
In three years of active, near real-time surveillance, building on an earlier, purely retrospective pilot study [28], our system detected 32 shigellosis events ( Table 2 and 3 ). Independent suspicion or discovery of only 2 of the 28 events considered suspect of outbreak occurred prior to detection of the first signal by WHONET-SaTScan. Of the 26 events for which we have PFGE evidence, 22 appeared to represent groups of truly genetically interrelated cases, including 9 new patterns (1 of them closely related to pattern identified previously) and 13 subtypes identified before in the NDB , with supporting evidence of epidemiologic linkage for 7.
The 32 detected events represented a broad range of the Shigella variants circulating in Argentina and were distributed among the six participating provinces. Some were of particular public health importance because of long duration or number of patients, e.g. Events 7 and 16, or because of a distinct resistance profile, e.g. Events 16 and 21, both S. sonnei AMP-SXT. Four pairs of events that were related according to PFGE patterns may have represented additional long-lasting outbreaks.
The two known outbreaks in these six provinces that were not detected by WHONET-SaTScan could not have been found by the system. One, a plasmid-conferred cefpodoxime-resistant cluster of S. sonnei in February-March 2011, could not have been detected because cefpodoxime resistance was not one of the phenotypes analyzed at the time. However, it did show up as a non-statistically significant cluster of SXT-AMP-nonsusceptible resistant S. sonnei, with RI = 116. The other, of Shigella flexneri AA479, appeared in one of the new provinces in August–September 2011, but this was before the new laboratories' data were incorporated and analyzed and before the variant was given a specific code in WHONET.
In Argentina, under the auspices of the international laboratory network PulseNet, it has been possible to maintain active surveillance using PFGE to detect circulating clones [16], [31]. When PFGE results are communicated to local public health agencies, they inform investigation into possible sources of contamination and their persistence over time [32]. PFGE results of the type we often saw one predominant pattern with other closely related genetic subtypes in the same event are common in Shigella spp., particularly in long-lasting events with person-to-person transmission, such as Events 7 and 16. Others have reported widespread outbreaks in which this mode of transmission was confirmed by molecular typing results and epidemiologic data [33], [34]. Where person-to-person spread is a prominent feature of the outbreak, more variability is expected [26], [27]. In each of two outbreaks with point-source exposures, event 17 and the outbreak of S. flexneri AA479 in the provinces that had not yet been incorporated into the surveillance system, the PFGE patterns were indistinguishable.
There are several limitations to this study. As is often the case in evaluations of surveillance systems, there was no known set of outbreaks that could serve as a gold standard against which to compare all the events. Some events were studied by public health local authorities, while all the cases were reported to the national surveillance system. Therefore, it was not possible to calculate such measures as sensitivity, specificity, or negative predictive value. PFGE testing did stand in as a strong validation method, and 22 (84.6%) of 26 events for which PFGE was done showed evidence of close genetic relatedness. However, close relatedness by itself does not prove that isolates belong to a single outbreak, and accompanying epidemiologic studies were not carried out for every event, so we cannot claim to know the positive predictive value of our system. Also, the system was not pre-specified in a static protocol, having changed in important ways over the three years: the inclusion of additional laboratories, changes in parameter settings, and changes in the kinds/phenotypes of Shigella for which clusters could be detected (Table 1). The results may have looked somewhat different had we included all 23 laboratories and the final parameter settings and Shigella variants from the start. Finally, we did not compare results of different cluster detection methods applied to the same data. We selected the space-time permutation test because it does not require population-at-risk (denominator) data and it adjusts for purely spatial and purely temporal variation and can do so without multiple years of historical data. It would be worthwhile in a future methods-oriented endeavour to compare the performance of this method with other cluster detection methods.
In large measure, this still-evolving, real-world, laboratory-based surveillance system satisfies criteria for public health utility, including that it 1) be specific, 2) have timely electronic data entry, 3) integrate multiple laboratories using uniform protocols and databases, 4) be used by the agencies responsible for disease control, and 5) implement statistical methods for detecting departures from background levels in both time and space. We have detected clusters of shigellosis of public health importance, which have been confirmed by PFGE as consisting of closely related clones, and informed local public health efforts. This WHONET-SaTScan system of data organization and analysis could represent a good complementary tool for national surveillance system, for early outbreak detection in real time, signalling the importance to investigate some events , and could be applied to other pathogens, implemented by other networks of laboratories, and scaled up to national and international levels.
|
10.1371/journal.pgen.1001293 | Mapping a New Spontaneous Preterm Birth Susceptibility Gene, IGF1R, Using Linkage, Haplotype Sharing, and Association Analysis | Preterm birth is the major cause of neonatal death and serious morbidity. Most preterm births are due to spontaneous onset of labor without a known cause or effective prevention. Both maternal and fetal genomes influence the predisposition to spontaneous preterm birth (SPTB), but the susceptibility loci remain to be defined. We utilized a combination of unique population structures, family-based linkage analysis, and subsequent case-control association to identify a susceptibility haplotype for SPTB. Clinically well-characterized SPTB families from northern Finland, a subisolate founded by a relatively small founder population that has subsequently experienced a number of bottlenecks, were selected for the initial discovery sample. Genome-wide linkage analysis using a high-density single-nucleotide polymorphism (SNP) array in seven large northern Finnish non-consanginous families identified a locus on 15q26.3 (HLOD 4.68). This region contains the IGF1R gene, which encodes the type 1 insulin-like growth factor receptor IGF-1R. Haplotype segregation analysis revealed that a 55 kb 12-SNP core segment within the IGF1R gene was shared identical-by-state (IBS) in five families. A follow-up case-control study in an independent sample representing the more general Finnish population showed an association of a 6-SNP IGF1R haplotype with SPTB in the fetuses, providing further evidence for IGF1R as a SPTB predisposition gene (frequency in cases versus controls 0.11 versus 0.05, P = 0.001, odds ratio 2.3). This study demonstrates the identification of a predisposing, low-frequency haplotype in a multifactorial trait using a well-characterized population and a combination of family and case-control designs. Our findings support the identification of the novel susceptibility gene IGF1R for predisposition by the fetal genome to being born preterm.
| Preterm birth is the major cause of infant deaths and life-long neurologic and cardiopulmonary morbidity. More than 10% of births in the United States occur prematurely, and the rate is increasing without known effective prevention. Previous premature birth increases the risk 3-fold in subsequent pregnancies. We report here, for the first time to our knowledge, a genome-wide study on susceptibility to spontaneous preterm birth in singleton pregnancies. To detect novel regions of the genome associated with preterm birth, we performed linkage analysis on seven carefully selected large families with recurrent spontaneous premature births. When we studied the fetuses, evidence was found for linkage of a region on chromosome 15 with spontaneous preterm birth, with the highest linkage signals occurring within a single gene, IGF1R. Evidence of the involvement of this gene in the etiology of preterm birth was further strengthened by subsequent haplotype segregation analysis and case-control analysis of an independent patient population. The IGF1R gene encodes insulin-like growth factor receptor 1 (IGF-1R), an important protein that potentially regulates signaling cascades involved in the onset of labor. Our analyses are unique in providing evidence that fetal IGF1R influences the risk of spontaneous preterm labor, leading to preterm birth.
| Preterm birth, defined as birth before 37 wk of gestation, accounts for an estimated 2 million annual deaths worldwide and is the major cause of serious morbidity in infants born preterm. Currently, approximately 12% of all births in the United States are premature. The serious acute diseases of prematurely born infants are principally caused by functional immaturity. Common life-long diseases that result in deteriorating quality of life among individuals born preterm include a chronic respiratory disease called bronchopulmonary dysplasia; retinopathy of prematurity, which is the most common cause of blindness in infants; cerebral palsy; and cognitive disorders [1]. The majority of preterm births (approximately 70%) occur after spontaneous onset of labor; nearly 50% of these cases are preceded by rupture of fetal membranes. Apart from excessive uterine distension in multiple pregnancies or certain fetal malformations, and severe maternal diseases such as sepsis and abdominal trauma, no obvious environmental risk factors can be identified in most preterm births. Activation of spontaneous preterm labor and preterm birth is thought to result from the action of multiple pathways and mechanisms, including endocrine dysfunction or ascending intrauterine infection and inflammation that can lead to the induction of labor-producing mediators [2]. Despite ongoing research efforts, there is no effective medication for the prevention of spontaneous preterm birth (SPTB).
A history of SPTB of a single fetus is a strong predictor of its recurrence in families [3]. Approximately 20% of mothers with a preterm delivery have another baby born preterm [4], suggesting that factors that are stable over time, such as genetics, affect birth timing [5]. Mothers and daughters [6] and sisters [7] share the risk of delivering preterm. Twin studies suggest a heritability estimate of about 30% [8]–[10]. Both the fetal and maternal genome, as well as gene–gene and gene–environment interactions are likely to influence predisposition to SPTB. Several studies using fetal or maternal DNA have reported associations of individual gene polymorphisms [11]. These studies have focused on genes involved in infection, inflammation, and innate immunity; e.g. those encoding the cytokines tumor necrosis factor alpha and interleukins 4, 6, and 10, and mannose-binding lectin [12]–[21]. However, most of these associations were not replicated in subsequent studies and across populations. So far, only case-control candidate gene studies have been conducted for SPTB. Genome-wide methods of identifying genes a priori may reveal genes not considered to be obvious candidates, which are potentially important unexplored sources of variability in preterm birth. These studies may contribute to defining the risk of SPTB and developing potential preventive interventions.
The overall aim of the present approach was to define major genes that influence the susceptibility to SPTB. In this first report, we describe a SNP-based genome-wide linkage and haplotype segregation analysis of recurrent familial SPTB using a strictly defined phenotype and carefully selected families, followed by case-control association analysis of a study population independent of the subjects used for the linkage scan. The linkage scan was performed with seven large families originating from northern Finland, where the population is characterized by genetic homogeneity, making it advantageous for gene-mapping studies [22]. With the phenotype defined as being born preterm, significant linkage signals (HLODmax = 4.68) were obtained for chromosome locus 15q26.3 in a region harboring IGF1R (MIM *147370), the gene encoding insulin-like growth factor receptor 1 (IGF-1R). Haplotype segregation analysis performed for markers encompassing the IGF1R gene revealed prominent identical-by-descent (IBD) within-family and identical-by-state (IBS) between-family haplotype sharing among affected relatives. Evidence of the involvement of IGF1R in the etiology of SPTB was further strengthened by case-control analysis of an independent cohort located in northern and southern Finland, with a 6-SNP IGF1R haplotype overrepresented in SPTB infants. In summary, evidence from our linkage, haplotype sharing, and association analyses implicated IGF1R as a candidate gene for susceptibility to SPTB.
The overall study design is illustrated in Figure 1.
We selected the families for the linkage analysis from a total 120,000 births that took place in a single regional hospital in northern Finland; this region is characterized by a homogeneous and stable population with a low prevalence of prematurity (approximately 5.5–6.5% of all births). We chose mothers with recurrent SPTB using very stringent criteria, with known risk factors for SPTB and elective preterm births without labor among the exclusion criteria. We identified 120 mothers with at least two spontaneous singleton preterm deliveries. Family interviews revealed 20 large families with multiple relatives affected by SPTB. According to a genealogical study, these families were non-consanginous. Finally, families with apparent maternal inheritance of SPTB were chosen for the analysis.
We conducted parametric linkage analysis using seven large northern Finnish families with recurrent SPTB. Because genetic factors acting either on the fetus or the mother may influence SPTB, we used two settings in this study: affected fetus or infant phenotype (being spontaneously born preterm as the phenotype, n = 41) and affected mother phenotype (giving spontaneous preterm birth as the phenotype, n = 21). The pedigrees of the families are shown in Figure S1.
Before the analysis, markers were linkage-disequilibrium (LD) pruned to exclude high-LD SNPs, leaving 6377 markers with an average distance of 0.43 Mb between consecutive markers. We considered a heterogeneity logarithm of odds (HLOD) score of >2 as an initial signal of linkage. When we studied the affected fetus phenotype, analysis of the pruned marker set revealed HLOD scores of >2 on six autosomes, as depicted in Figure 2. The maximum HLOD score (HLODmax) of 2.59 was detected for SNP rs2715416 on chromosome locus 15q26.3 (θ = 0.04, α = 1.00). Figure 2 also shows the linkage signals for mother-based analysis, with an HLODmax of 1.53 at rs11167102 on chromosome locus 8q24.3 (θ = 0.00, α = 1.00).
For the affected fetus phenotype, we performed fine mapping using the unpruned marker set on the regions flanking (approximately 5 Mb) the six initial linkage signals. We considered an HLOD of >3 as a further sign of linkage. Table 1 reveals that the three markers with an HLOD of >3 were all on the same chromosome locus 15q26.3 and shows the HLOD scores in the fine-scale analysis for each of the seven families with recurrent SPTB. The HLODmax of 4.68 was obtained at rs2684811. Because one preterm infant not fulfilling the stringent criteria of SPTB (preterm infant from a twin pregnancy) was included among the premature births, we repeated the linkage analyses while excluding this infant from the affected individuals, with little effect on the results. All three markers with the highest HLOD scores are located within a single gene IGF1R. Interestingly, the markers that yielded the second- and third-highest HLOD scores in the mother-based analysis (SNPs rs329292 and rs11247268 with HLOD scores of 1.51 and 1.48 respectively; Figure 2) were located on the same chromosomal region (15q26.3) as the marker with HLODmax in the infants, with a distance of approximately 2.2 Mb between the markers. Because of the colocalized linkage signals in both the fetal- and mother-based analysis in the region including IGF1R, we chose to explore this gene in greater detail.
We performed haplotype segregation analysis using the SNPs flanking the highest linkage peak within IGF1R in the six linked families 24, 70, 126, 150, 185 and 253. Family 210 did not undergo haplotype segregation analysis due to an absence of linkage in this family (Table 1). The analysis was performed for 30 SNPs covering a 330 kb region encompassing the entire IGF1R gene. We aimed to resolve whether one or more distinguishable haplotype segments inferred from these SNPs cosegregated with SPTB. While not suitable as such for statistical evaluation, segregation analysis is a useful tool for an empirical hypothesis-generating approach [23]. The model used for the segregation analysis was initially designed to be best-fit to the mode of inheritance (MOI) used in the linkage analysis (dominant MOI, allowing for the presence of healthy carriers as transmitters of a disease-cosegregating haplotype). However, haplotype-sharing analysis offers the advantage over parametric linkage analysis of dissecting the genetic data regardless of true MOI.
IGF1R haplotypes for family 70 with the highest linkage signal are represented in Figure 3, and Figure S2 shows these haplotypes for all of the linked families, including family 70. In all six families, and considering the haplotypes comprising 30 SNPs spanning the entire IGF1R gene, we observed prominent IBD within-family haplotype sharing among the affected relatives (Figure 3 and Figure S2). A single disease-cosegregating haplotype was shared IBD within family in families 70, 126, 253 and 185. In the remaining two families, 24 and 150, an IBD-shared haplotype was identified in a subset of members of each family, and a second haplotype derived from another carrier was identified in a different subset of members. On the whole, the segregation analysis supported the assumed model very precisely, because it predicted the carriership of a disease-cosegregating haplotype in 34 out of all 38 affected individuals (absent from individuals 70-2, 150-15, 150-19 and 150-24), whereas only six nonaffected family members were deemed to be carriers of a disease-cosegregating haplotype (present in individuals 24-1, 24-5, 24-11, 24-9, 150-8, and 150-26) (Figure 3 and Figure S2). The recombinations that we observed, particularly that in the unaffected male 253-5, also fit the segregation model well. Three of the families (126, 253, and 185) showed a complete haplotype-disease cosegregation (Figure S2). However, unexpected male carriers were identified in four families (spouses of females 24-2 and 70-1, and males 253-1 and 150-3). Families 24 and 150 were consistent with a mixed maternal-paternal bilineal transmission pattern, and family 70 exhibited nearly complete penetrance with mixed unilineal maternal–paternal transmission. The segregation patterns were consistent with complete penetrance in families 126, 253, and 185; with 100% maternal transmission in families 126 and 185; and unexpected 100% paternal transmission in family 253.
An attempt to identify a similar IBD- or IBS-shared chromosomal segment among families led us to discover two core haplotypes with overlapping locations (illustrated in Figure S2). A 55-kb 12-SNP haplotype, CAGACGATACTC (core I, comprising the interval between SNPs rs1879612–rs2715416), was shared IBS among five families: all of families 24, 70, 126 and 253 and part of 150. Another 79-kb 11-SNP core haplotype, ATGTGTAATGT (core II, SNPs rs2684761–rs3743259), was shared IBS between part of family 150 and the whole of family 185. The disease-segregating chromosomes with the core II haplotype were maternal in 100% of preterm-born individuals carrying these chromosomes, while haplotypes with core I were maternal in only about half (47%) of the cases.
All of the linked families shared one of the core haplotype segments, I or II. To demonstrate further the relevance of these haplotypes, we compared the haplotype frequencies in the affected and unaffected members of the linked families with those of an independent Finnish reference population from the Nordic Database Lund-Malmö dataset [24]. These Finnish reference samples (referred to as Nordic–Finn, n = 955) are derived from a region in Bothnia in which there is evidence of western late settlement. The people in this region are genetically close to those in the region of western early settlement representing the origin of the linkage families [22]. Thus, these samples represented a good control and were also better matched with our study population than HapMap CEU individuals (CEPH; Utah residents with ancestry from northern and western Europe). Frequencies of haplotype core segments I and II were 0.30 and 0.33, respectively, in the affected members of the linkage families. Frequencies of core I and II haplotypes were 0.17 and 0.12, respectively, in the unaffected members of the linkage families, and 0.18 and 0.06 in the Nordic–Finn population. Thus, frequencies of the core haplotypes in the reference population were close to those of the unaffected members of the linkage families, while these frequencies were increased in the affected members of the linkage families. The higher frequency of haplotype core II in the unaffected members of the linkage families compared to Nordic–Finn individuals is explained by the high overall incidence of this haplotype in the largest linked family (Family 150; Figure S2). In terms of the LD patterns (Figure S3) in the unaffected individuals, the genetic profile of our study population was representative of the Nordic–Finn reference population. Furthermore, there was a short two-SNP major-allele haplotype AT at rs11630259–rs1357112 shared IBS between core segments I and II in 100% of the affected inviduals (n = 38) and predicted carriers (n = 18) (Figure S2), whereas the same haplotype was present in only 62% of the unaffected family members (n = 37) and 72% of the Nordic-Finn individuals. Whether this particular segment sharing reflected the critical location of disease predisposition or arose by chance, could not be statistically evaluated in the current setting. Unfortunately, these two SNPs, which are not in LD with the surrounding region SNPs (Figure S3) and thus could not be imputed from other SNPs, failed to settle in the Sequenom iPLEX association platform and therefore could not be included in the following case-control association study.
We additionally performed a post hoc linkage analysis in which we used the transmission information obtained from the segregation analysis model, with both unaffected carriers and individuals born preterm defined as affected (naffected = 55). An overall increase in the HLOD scores was observed within IGF1R, with an HLODmax of 3.81 at rs2715416 after pruning and an HLODmax of 5.15 at rs2684811 without pruning, further supporting the view that this genomic region may harbor a true susceptibility gene.
To validate the potential linkage and association between IGF1R and SPTB, we enrolled a new Finnish population comprising cases originating from the northern (Oulu) and southern (Helsinki) regions of the country; these cases were independent of the families used for the linkage analysis. Tagging SNPs covering the entire IGF1R gene were determined using HapMap data from the CEU population. Among the 20 SNPs studied, two (rs7165181 and rs4966038) showed a weak association (P<0.05) in the infants but not in the mothers (Table 2). Even so, a 55-kb region of six SNPs in LD (Figure 4) spanning these two SNPs and extending to rs2715416 (which yielded the original linkage signal) showed statistically significant haplotype association with SPTB in the infants exclusively (Table 3). Similar associations were evident in both recurrent and sporadic SPTB. The signals obtained from independent sets of study populations using linkage, segregation and association approaches were colocalized within the same region of IGF1R (Figure 5 and Table S1) and were consistently observed in the infants but not the mothers; i.e., when the affected phenotype was being born preterm instead of giving birth to preterm infant(s). The birthweights and gestational ages of preterm infants carrying the associating haplotype (n = 71; 2,025±626 g; 32.6±3.0 wk; mean ± standard deviation) did not differ significantly from those of preterm infants without this haplotype (n = 263; 2,105±616 g; 32.3±2.9 wk; P values of 0.64 and 0.35, respectively). Similar to our control population, the predicted frequency of the associating haplotype was 0.052 in the HapMap CEU population. In HapMap CHB (Han Chinese in Beijing, China) and JPT (Japanese in Tokyo, Japan) populations, allele and haplotype distibutions were completely different from our controls, as well as from the HapMap CEU population. The CHB and JPT populations completely lacked the associating haplotype. We were not able to estimate the frequency of this haplotype in the rest of the HapMap populations, including YRI (Yoruba in Ibadan, Nigeria), due to unavailable genotype data for part of the SNPs in these populations.
According to epidemiological studies, there is evidence for a heritable predisposition to preterm birth with both maternal and fetal contribution [11]. To our knowledge, the present report describes the first genome-wide investigation and the first linkage study to identify genomic regions associated with SPTB. We detected significant parametric HLOD scores in the infants for three intronic markers (rs1521480, HLOD = 3.63; rs4966936, HLOD = 3.63; and rs2684811, HLOD = 4.68) on chromosome 15q26.3 within a single candidate gene, IGF1R, encoding the type 1 insulin-like growth factor receptor IGF-1R (Table 1).
We identified one major disease-cosegregating haplotype (core I) in all but one of the six linked families (Figure 3 and Figure S2). Rather than IBD, this sharing is likely to be IBS, because of the high frequency of this haplotype in the population, a view consistent with the genealogical survey, which suggested no common ancestry among the linked families. The occurrence of an alternative shared haplotype segment (core II) in all of family 185 and part of family 150 may reflect allelic heterogeneity or absence of true linkage in this subgroup, which would be typical of complex phenotypes even in an isolated population [25]. Taken together, our data on linkage and disease segregation of IGF1R SNPs are consistent with a role for this genomic region in SPTB under dominant MOI with incomplete penetrance allowing for healthy carriers or disease transmitters under etiological heterogeneity (the existence of phenocopies). However, because the segregation pattern supports the model of disease-cosegregating IGF1R haplotype transmission via both healthy male and female carriers, our initial MOI based on pure unilineal maternal transmission is likely to be oversimplified.
We examined the gene encoding IGF-1R in a separate investigation with a Finnish population originating from two regions of the country, which revealed an association of a fetal IGF1R haplotype with SPTB (Table 3 and Figure 5). Both the pattern of maximal haplotype sharing in the linked families and the region of association observed in the case-control study independently placed SPTB susceptibility on the same segment within IGF1R. As a whole, these analyses provide evidence that the fetal IGF1R influences the risk of spontaneous preterm labor leading to preterm birth.
IGF-1R is a heterotetramer composed of two extracellular alpha subunits containing a ligand-binding site for IGFs and two membrane-spanning beta subunits harboring intracellular tyrosine kinase activity involved in a variety of cellular functions [26], [27]. Upon activation by IGF-1 or IGF-2, IGF-1R participates in regulation of the cell cycle. Accordingly, certain IGF1R mutations result in intrauterine growth restriction, whereas polymorphisms of this ubiquitously expressed gene may not influence fetal growth [28]–[32]. The roles of IGF-1R in normal and pathological growth and differentiation and aging involve interactions with the ubiquitous growth factors, hormones, and proinflammatory cytokines that are considered to be mediators of the labor process [33]–[35]. Several IGF-binding proteins that regulate IGF-1R–dependent signaling cascades have been studied in the context of SPTB, and some of them have been implicated in preterm labor [36]–[38]. However, these studies did not involve IGF1R and thus the extension of studies involving IGF-1R (particularly the ligand-binding alpha subunit encoded by exons 1–10) to the process of labor, with special consideration of fetal involvement in the endocrine and paracrine control of the preterm labor process, is clearly indicated. The range of regulatory roles performed by the IGF system is consistent with our view that IGF1R influences susceptibility to SPTB. Furthermore, a recent study revealed a parent-of-origin-specific methylated site within intron 2 of IGF1R [39], which localizes to the same region that was identified in our haplotype segregation and case-control association analyses. This site was predominantly methylated on the maternally derived chromosome and may be involved in the imprinting process. This finding suggests a possible epigenetic mechanism through which IGF1R may be involved in the preterm labor process. In our linkage families, the core II IGF1R haplotype was maternally transmitted to the individuals born preterm, whereas the core I haplotype had a mixed transmission. This is an important aspect to consider in future studies involving IGF1R and preterm birth.
In the present study, we obtained no clear linkage or association signals when the outcome phenotype was giving preterm birth (affected mother phenotype). Because the number of affected mothers was small (n = 21), the power to detect regions of linkage was limited. Although the maternal HLODmax was <2, which we considered to be below an initial threshold suggesting linkage in this study, we identified the regions yielding the highest maternal linkage signals (HLODs of 1.53 and 1.51 for markers rs11167102 and rs329292 on chromosomes 8q24.3 and 15q26.3 respectively). An interesting finding was that the SNP with the maternal HLOD score of 1.51 was located on the same region as the one with the HLODmax obtained from the infants, separated by 2 Mb (Figure 5). To conclude, the lack of association in the mothers and the observation of both maternal and paternal transmission in the segregation analysis unexpectedly did not support a major maternal contribution in IGF1R-mediated genetic susceptibility to SPTB. Our discovery of a fetal gene was unforeseen, but does not exclude the role of maternal effects via other susceptibility genes. Interestingly, a recent study suggested that the fetal genome plays a more significant role than the maternal genome in individuals of European ancestry [40]. However, other studies have pointed out the importance of the maternal genome in genetic predisposition to SPTB [41], [42].
Our study has several strengths compared to previous genetic studies of preterm birth. These candidate gene studies have been limited in several respects, such as small or mixed populations that may represent only the mother or the fetus, lack of replication, inconsistencies in phenotype definitions and incomplete coverage of variation within a gene [11]. In contrast, our study first focused on detecting novel genes involved in SPTB using a nonhypothesis-driven genome-wide approach. We evaluated both the maternal and fetal contribution. Additionally, careful attention was paid to selection of the families for the linkage analysis to ensure a precise phenotype definition. Lastly, we replicated the finding of a linked and disease-cosegregating region in an independent case-control cohort covering a large part of the entire country. The effect size we observed in the infants fell within the range of power calculations described in Materials and Methods. Although the Finnish population represents a higher genetic similarity than most populations, a clear substructure caused by population history is known to exist within the country [22]. However, the linkage and segregation analyses were performed using a relatively homogenous study population, which is amenable to using a search for shared genomic segments [25]. Our case-control cohort comprised two regions (northern Finland and southern Finland/Helsinki) known to represent slightly distinct patterns of genetic variation; the Helsinki region is known to be representative of a genetically more diverse population [22], [43]. The fact that the association was observed across these two subpopulations gives further credence to our findings.
Our results constitute proof of principle for using a genome-wide SNP linkage scan to elucidate a complex heterogenic trait via stringent initial selection of a limited set of individuals. Despite the important strengths mentioned above, our study has some limitations. The result remains yet to be confirmed and extended by independent case-control analyses in populations with larger sample sizes than in the current study, and by studying additional SNPs covering the large gene more extensively. It is possible that while the ethnically homogenous nature of our Finnish linkage families can facilitate detection of genetic factors, the relatively high degree of genetic similarity among the Finnish population compared to others may mean that this finding cannot be generalized to more outbred populations. The frequency of the associating IGF1R haplotype in the HapMap CEU population was similar (0.052) to the frequency of controls in our case-control analysis (Table 3), while this haplotype was completely lacking from HapMap CHB and JPT populations, reflecting population-specific variation within this genomic segment. In the current setting, it cannot be evaluated whether the observed association was due to a haplotypic effect or another linked polymorphic site that was not analyzed in the current study. Therefore, this region needs to be more thoroughly investigated to find the causative sequence variant(s) and also to determine whether other populations may have an IGF1R-mediated risk of SPTB. Furthermore, although we discussed here only regions with an HLOD of >3, other regions with initial linkage signals on chromosomes 2, 4, 10, 12, and 13 should not be ignored as candidates for the risk of spontaneous preterm labor leading to preterm birth.
In conclusion, our genetic linkage and haplotype segregation analysis mapped the novel fetal SPTB susceptibility gene IGF1R on chromosome 15q26.3; this was further confirmed by association in an independent case-control population replicate. This result was unexpected, because the studies on the etiology of SPTB have been focused on cytokine-mediated inflammatory signaling as a possible route of predisposition by the maternal genome. Future clarification of the molecular mechanism of a growth factor pathway with considerable influence on the heritable proportion of the risk of spontaneous premature labor and birth is likely to open up a new potential avenue for preventive therapies.
Written informed consent was obtained from the participants and the study was approved separately by the Ethics Committee of Oulu University Hospital and that of Helsinki University Central Hospital.
We processed the Affymetrix array data files using PLINK, v. 1.02 [46]. We used a 1 Mb-to-1 cM converted map, as justified in [47]. Before linkage analysis, the markers were pruned using a linkage disequilibrium (LD) r2 threshold of 0.35 with PLINK. Markers with either minor allele frequency (MAF)<0.08, genotyping failure >0.1 or Mendelian errors were excluded, as well as those violating the Hardy–Weinberg equilibrium (HWE, P<0.001). Data processing after removal of the high-LD SNPs yielded a selection of 6,377 autosomal SNPs for the genome-wide linkage analysis. We rechecked the genotype data for any Mendelian inconsistencies using PedCheck [48] before proceeding to linkage analysis.
We studied two main outcome phenotypes for SPTB: (1) being spontaneously born preterm (affected fetus/infant, naffected = 41), and (2) giving spontaneous preterm birth (affected mother, naffected = 21) (pedigrees in Figure S1). A parametric two-point linkage analysis of SPTB as a dichotomous trait was performed with ANALYZE, v. 1.9.3 BETA, which is a linkage and LD analysis package that uses FASTLINK 4.1P to calculate the pedigree likelihoods and is capable of managing both extended pedigrees and large numbers of markers [49]. We used a dominant low-penetrance model, assuming a disease allele frequency of 0.001 and penetrances of 0.001, 0.001, and 0 for the homozygotes, heterozygotes and wild-type homozygotes, respectively. This kind of model is nearly convergent with a model-free analysis, because it minimizes the effect of misspecifying the true MOI while retaining the highest power of a parametric analysis to detect linkage [50], [51]. Parametric linkage in the presence of heterogeneity was assessed using heterogeneity LOD (HLOD) scores and their accompanying estimates of the proportion of linked families (α) and recombination fraction (θ).
We applied two-point linkage analysis to avoid bias that could result in multipoint analyses due to missing parental genotypes and markers residing in LD [52]. Our linkage analysis strategy was to perform an analysis with an LD-pruned marker set and then to screen regions with initial linkage signals (HLOD>2) with a denser marker map including all markers with MAF>0.08 on the region. Using high-density SNPs in linkage analysis has the advantages of a low error rate, high per-marker call rates, and higher information content when compared to microsatellites [53]. In two-point analysis, LD does not increase type 1 error [54].
When a marker showed HLOD>2 (six positions), we selected the genomic region for further analysis. This fine-scale linkage analysis was performed using the unpruned marker set on the flanking region (∼5 Mb) of each of the initial linkage signals with the same parametric model as the original scan. We considered an HLOD score of >3 as a signal of linkage.
We used PedPhase version 2.0 utilizing integer linear programming (ILP) to find the minimum-recombinant haplotype configuration in the linked families 24, 70, 126, 150, 185 and 253 for the SNPs flanking the best linkage peak [55]. Deceased or untyped individuals were also included if their haplotypes could be reconstructed from their relatives. The potential presence of disease-cosegregating haplotypes (“affected haplotypes”, affected fetus phenotype) was evaluated using a model compatible with the assumption of maternal unilineal inheritance with incomplete penetrance, utilizing the scheme of the most likely segregation pattern to search for a minimum number of affected haplotypes. The model assumed dominant MOI, allowing for the presence of healthy carriers as transmitters of an affected haplotype.
We used an independent Finnish reference population of the Nordic Database Lund-Malmö dataset [24]. These samples (referred to as Nordic–Finn, n = 955) are derived from a region in Bothnia in which there is evidence of western late settlement. We used Beagle 3.1 [56] to infer the haplotype phases for unrelated individuals in the Nordic–Finn samples.
IGF1R is a large, >300 kb gene with nearly 2,000 SNPs (NCBI B36 assembly dbSNP b126), but little frequent exonic variation; it has relatively weak intragenic LD and is not readily divided into discrete blocks of limited haplotype diversity. Using HapMap data (release 23a/phaseII) from the CEU population to determine tagging SNPs (tSNPs) covering the entire IGF1R gene (chr15: 97,004,502–97,329,396), we obtained a list of 100 tSNPs (pairwise tagging, r2 cutoff 0.8, MAF>0.1). Given the number of individuals available for the study and the power to detect associations, it was reasonable to set up a single iPLEX set, which can include up to approximately 30 compatible SNPs. To select SNPs, we visualized the gene region's LD pattern in the HapMap CEU population in the Haploview program v. 4.1 [57] and selected one SNP with the highest MAF from each haplotype block designated by this program. We also included the coding SNP rs2229765 (Glu1043Glu, also referred to as 3174 G>A) because individuals carrying at least one A allele are reported to have lower levels of free plasma IGF-1 than GG homozygotes [58], suggesting that this polymorphism may have a functional effect. After testing and validation, a selection of 20 SNPs (listed in Table 2 and details in Table S1) remained for the case-control association study.
We performed statistical tests using Haploview v. 4.1 [57]. We took multiple testing into account by performing permutations with 10,000 replicates and considered a corrected P value of <0.05 to be statistically significant. For analysis of haplotypes with birthweight and gestational age, we inferred phased IGF1R haplotypes from unphased data using Beagle 3.2 [56]. To analyze potential associations between haplotypes and birthweight or gestational age, the nonparametric Mann-Whitney U-test was used with Predictive Analytics SoftWare (PASW) statistics, version 17.0.3 (IBM SPSS, Inc.). Power consideration of the case-control study: the sample size provides 80% power (alpha = 0.0025, allowing for multiple testing of 20 SNPs) to detect the risk allele carrier relative risks of approximately 2.1–2.7 in the infants and 2.4–3.1 in the mothers for allele frequencies ranging from 0.2 down to 0.05, considering SPTB as a discrete trait, assuming causal SNP, a prevalence of 0.05, and using the allelic 1 df test [59].
All of the chromosomal positions refer to NCBI Build 36 of the human genome.
International HapMap Project: http://www.hapmap.org
National Center for Biotechnology: http://www.ncbi.nlm.nih.gov
Nordic Database: http://www.nordicdb.org
Genetic Power Calculator: http://pngu.mgh.harvard.edu/~purcell/gpc
|
10.1371/journal.pntd.0004532 | Community Rates of IgG4 Antibodies to Ascaris Haemoglobin Reflect Changes in Community Egg Loads Following Mass Drug Administration | Conventional diagnostic methods for human ascariasis are based on the detection of Ascaris lumbricoides eggs in stool samples. However, studies of ascariasis in pigs have shown that the prevalence and the number of eggs detected in the stool do not correlate well with exposure of the herd to the parasite. On the other hand, an ELISA test measuring antibodies to Ascaris suum haemoglobin (AsHb) has been shown to be useful for estimating transmission intensity on pig farms. In this study, we further characterized the AsHb antigen and screened samples from a population-based study conducted in an area that is endemic for Ascaris lumbricoides in Indonesia to assess changes in AsHb antibody rates and levels in humans following mass drug administration (MDA).
We developed and evaluated an ELISA to detect human IgG4 antibodies to AsHb. We tested 1066 plasma samples collected at different times from 599 subjects who lived in a village in rural Indonesia that was highly endemic for ascariasis. The community received 6 rounds of MDA for lymphatic filariasis with albendazole plus diethylcarbamazine between 2002 and 2007. While the AsHb antibody assay was not sensitive for detecting all individuals with Ascaris eggs in their stools, the percentage of seropositive individuals decreased rapidly following MDA. Reductions in antibody rates reflected decreased mean egg output per person both at the community level and in different age groups. Two years after the last round of MDA the community egg output and antibody prevalence rate were reduced by 81.6% and 78.9% respectively compared to baseline levels.
IgG4 antibody levels to AsHb appear to reflect recent exposure to Ascaris. The antibody prevalence rate may be a useful indicator for Ascaris transmission intensity in communities that can be used to assess the impact of control measures on the force of transmission.
| Ascariasis is a neglected tropical disease caused by the intestinal nematode Ascaris lumbricoides that affects hundreds of millions of people in the developing world. Current methods for diagnosis of this infection are based on detecting eggs in the stool that are excreted by adult Ascaris worms. However, these methods have limited sensitivity for recent infections, and they do not detect infections with immature parasite stages that do not always result in the establishment of adult worms in the human intestine. We have previously shown that an assay for antibodies to Ascaris hemoglobin in pig serum is useful for assessing transmission of Ascaris infections on pig farms. In this study, we developed and evaluated a similar antibody assay that is based on the detection of human IgG4 antibodies to Ascaris haemoglobin (AsHb). Community antibody rates decreased rapidly following mass drug administration of the anthelmintic drug albendazole, and this decrease reflected reduced Ascaris egg excretion at the community level. This antibody test may be a useful tool for assessing the impact of control measures on the transmission of new Ascaris infections in endemic populations.
| An estimated 1.45 billion people worldwide are infected with three soil transmitted helminths (STH) Ascaris lumbricoides, Trichuris trichiura, or hookworms [1]. Among the STH, infections with A. lumbricoides are most common and its public health impact is estimated to be approximately 1.31 million daily adjusted life years (DALYs) [1]. Current STH control programs are focused on morbidity control through community based deworming of school-aged children by annual or semiannual administration of a single dose of anthelmintics such as albendazole or mebendazole [2]. In addition, the large elimination programs for lymphatic filariasis and onchocerciasis provide anthelminthics to entire at risk populations that are also effective against A. lumbricoides [3, 4].
The intensity of A. lumbricoides infection is routinely measured by the number of eggs per gram (EPG) in stool by the Kato-Katz fecal thick-smear technique [5]. This method is useful for quickly assessing infection prevalence rates for evaluating the efficacy of control programs. However, the sensitivity of the Kato Katz smear is reduced in areas with low infection rates and intensities [6]. Furthermore, it is a time-consuming and cumbersome technique, and eggs in stool do not necessarily correlate well with the intensity of exposure to new infections with migrating stages of the parasite that are major contributors to morbidity caused by the parasite [7]. Therefore, it would be useful to have a practical method to measure the intensity of exposure to new infections at the population level. Research on the value of antibody assays for STH infections has been limited to date apart from interesting work on strongyloidiasis [8–10]. One reason for this is that people did not consider antibody testing to be a priority for common infections that can be diagnosed by microscopy. Also, it is commonly believed that antibody tests will not be able to distinguish between current and past infections. Only a handful of studies have evaluated antibody tests for ascariasis. Most of these studies used crude somatic antigen preparations or excretion/secretion (E/S) products derived from larvae or adult parasites cultivated in vitro. These antigens are difficult to procure and generally lack specificity [11–15].
Recently, Vlaminck et al., [16] evaluated a serological test for the detection of ascariasis in fattening pigs based on the detection of IgG antibodies to A. suum haemoglobin (AsHb) in serum samples by ELISA. The AsHb antigen is highly produced and secreted by both the adult and larval stages [17, 18]. Validation studies with samples from naturally and experimentally infected pigs showed that the antibody assay was superior to detection of eggs in feces for detecting exposure to the infection. A subsequent study showed that antibody reactivity to AsHb correlated with liver pathology caused by migrating A. suum larvae, and high antibody rates in pig herds were associated with low growth rates (reduced farm productivity) [19].
There are many parallels between ascariasis in pigs and in humans that is caused by the closely related species A. suum and A. lumbricoides, respectively. Therefore, the purpose of the present study was to investigate the potential value of anti-AsHb antibody testing for community diagnosis of human ascariasis.
Adult A. suum parasites were collected with permission from the intestines of infected pigs that were being processed as part of the normal work at a local abattoir in Ghent, Belgium. The fresh worms were snap frozen in liquid nitrogen for subsequent storage at -80°C. Tissue homogenization and mRNA extraction and cDNA construction were essentially performed as described in Rosa et al [20]. The AsHb antigen was purified according to the protocol described by Vlaminck et al [17]. The protein sequence of AsHb was obtained from the Entrez Protein Database of the National Center for Biotechnology Information (NCBI), USA (http://www.ncbi.nlm.nih.gov/protein) with the following accession number: AAA29374.1 (A. suum). Protein orthologs of AsHb in other nematode species were identified by a BLASTP search on WormBase ParaSite (http://parasite.wormbase.org/) against the available protein sequence databases for the following species (A. suum (PRJEB80881); A. lumbricoides (PRJEB4950), Toxocara canis (PRJEB533), Strongyloides stercoralis (PRJEB528), Necator americanus (PRJNA72135), Ancylostoma ceylanicum (PRJNA72583), Enterobius vermicularis (PRJEB503), Brugia malayi (PRJNA10729), Wuchereria bancrofti (PRJEB536), Loa loa (PRJNA60051), and Trichuris trichiura (PRJEB535)). The presence of signal peptides was detected using SignalP 4.1 software [21].
The AsHb product was amplified from parasite cDNA by reverse transcriptase- polymerase chain reaction (RT-PCR) using the primer pair AsHbFw (CACCATGCGCTCATTGCTATTATTATCG) and AsHbRv (TCAGTGTTGCTCTTCCTTATGC) and according to the amplification protocol described in Vlaminck et al [17]. The amplified PCR product was cloned into pET100/D-TOPO vector and transformed into One Shot TOP10 competent cells according to the manufacturers protocol (Invitrogen, Carlsbad, CA, USA). Positive colonies were analyzed using PCR and the PCR products were sequenced. One positive transformant was selected and the plasmid was purified using the PureLink HQ Mini Plasmid Purification Kit (Invitrogen). The plasmid DNA was used as template for the amplification of AsHb using the same primers as previously mentioned. The PCR product was sequenced to verify that the insert was in-frame for expression.
The AsHb-containing vector construct was transformed into BL21 Star (DE3) One Shot cells and cells were grown in Luria Broth (Miller) (Sigma, St. Louis, MO, USA) containing 50μg/ml carbenicillin (Sigma). Overnight cultures of the transformed bacteria were diluted 1:100 in LB + carbenicillin and grown at 37°C with shaking to an optical density of approximately 0.5–0.8 at 600nm. Protein expression was induced by addition of isopropylthiogalactoside to the culture medium to a final concentration of 1 mM. After 4h of incubation at 37°C under vigorous agitation (250 rpm), E. coli cells were pelleted by centrifugation and suspended in 1:25 of the initial culture volume of ice-cold RIPA lysis and extraction buffer (G Biosciences, St. Louis, MO, USA), incubated on shaking device at room temperature (RT) for 30 min and then centrifuged for 10 min at 4.500 g. The pellet was suspended again in RIPA lysis and extraction buffer and the previous incubation and centrifugation step was repeated twice more. The final pellet was dissolved in binding buffer (50mM Phosphate buffer, 0.5M NaCl, 10mM imidazole and 7M guanidine hydrochloride, pH 8.0). The recombinant protein was then bound to a HIS-Select Cobalt Affinity Gel column (Sigma) followed by a column wash with 5x column volumes of binding buffer and eluted from the column using elution buffer (50mM sodium phosphate buffer, 500mM NaCl, 7M guanidine hydrochloride, 250mM imidazole, pH 8.0). The eluate was dialyzed overnight against PBS in a Slide-A-Lyzer Dialysis Cassette, 7K MWCO (Thermo Fisher Scientific, Pittsburg, PA, USA) and subsequently concentrated on a Millipore centrifugal filter unit (Millipore, Billerica, MA, USA). Protein dye binding and BCA protein assay (Thermo Fisher Scientific) were used to determine the protein concentration.
Plasma samples were collected from individuals living in Mainang village on Alor Island (Province of East Nusa Tenggara, Timor, Indonesia) as part of a study of the impact of annual MDA with diethylcarbamazine (DEC) combined with albendazole (ALB) on Brugia timori and STH infections [22]. Only the most prevalent geohelminths, A. lumbricoides, hookworm and T. trichiura were analyzed in this study, since other species such as Hymenolepis spp. and S. stercoralis were only found in a few cases [23]. Baseline prevalence for A. lumbricoides, T. trichiura and hookworm in the whole community were 32.2%, 25.3%, and 9.4% respectively [23]. The STH infection rates and egg densities were assessed with single Kato Katz smears in 2002 (prior to any mass drug administration, MDA) and in 2009, two years after the sixth and last annual round of MDA. For these years, community egg output was determined as the sum of EPG of all people divided by the total number of tested individuals. Prevalence rates for STH in stool in 2004 and 2007 were assessed by the formalin ether enrichment method. Additionally, at baseline, 17.6% of the individuals had B. timori mf in their blood [23].
Plasma samples used in the present study were collected in 2002 prior to the first round of MDA, after 2 rounds of MDA (2004), just prior to the sixth and final round of MDA (2007), and 2 years after the last round of MDA (2009).
Other plasma samples used in this study were from people living in the East Sepik region of Papua New Guinea. Hookworm infection rates in this study area were over 90% and no A. lumbricoides infections were detected by Kato Katz examination of one stool sample per subject in any of these study communities. Non-endemic control plasma samples were from American subjects in St. Louis, MO, USA.
Protein samples were denatured and reduced in LDS 4x sample buffer (Thermo Fisher scientific) and separated by SDS-PAGE using Bolt 4–12% Bis-Tris Plus minigels (Thermo Fisher Scientific) under reducing conditions and either stained with SimplyBlue Safestain Coomassie stain (Invitrogen) for the visualization of the proteins or transferred onto nitrocellulose membranes for immunostaining as previously described [24]. After blotting, nitrocellulose membranes were blocked at room temperature for 1hr with 5% non-fat dry milk (BioRad, Hercules, CA, USA) in PBS, and then incubated with human plasma diluted 1:50 in PBS + Tween20 (0.05%) (PBST) and incubated at room temperature (RT) for 2h. After incubation with a secondary antibody (mouse anti human IgG4 pFc-HRP (Southern biotech, Birmingham, AL, USA), blots were washed with PBST and antibody binding was detected using CN/DAB Substrate kit (Thermo Fisher Scientific).
AsHb was deglycosylated with PNGase F according to the manufacturer’s protocol (New England Biolabs, Ipswich, MA, USA). Briefly, 10μg of AsHb was mixed with 10x denature buffer and H2O to make a 20μl total reaction volume that was denatured at 100°C for 10 min. This was followed by addition of 10X Glycobuffer 2, 10% NP-40 and 1μl of PNGase F (500,000 U/ml) and H2O to obtain a total reaction volume of 40μl that was incubated at 37°C for 2h. Finally, the deglycosylated product (dAsHb) was passed over a detergent binding spin column (Thermo Fisher Scientific) to remove the detergent that was added during the deglycosylation reaction. RNAseB treated with PNGase F was included as deglycosylation control.
For mass spectrometric (MS) analysis, N-glycans released with PNGase-F from 5μg of AsHb were labeled with 2-aminobenzoic acid (anthranilic acid, AA), as described [25]. MALDI-TOF-MS was performed in the negative ion reflector mode on an Ultraflextreme instrument (Bruker Daltonics, Germany) using DHB as matrix, as described [26]. Putative glycan structures were assigned on the basis monosaccharide compositions deduced from the observed m/z values.
Phosphorylcholine (PC) specific monoclonal antibodies (TEPC-15) (Sigma) were used to detect the presence of PC in AsHb and dAsHb by Western blot and ELISA. Anti-PC antibodies were detected with HRP conjugated anti-mouse IgA (Sigma) and nitro-blue tetrazolium/5-bromo-4-chloro-3′-indolyphosphate substrate (Sigma). PC linked to bovine serum albumin was used as a positive control.
Previous serological experiments performed by Santra et al., 2001 [13] and Chatterjee et al., 1996 [14] showed that human IgG4 responses to a fractionated adult E/S antigen of Ascaris were superior in reactivity and also showed less cross reactivity than IgG1, IgG2 and IgG3 subclass antibodies in sera from patients infected with hookworm, Trichuris and Strongyloides. This, in combination with the results from earlier experiments performed in the lab, led to the decision to use the IgG4 subclass antibody as detecting antibody in the immunological assays described in this study.
Antigen (AsHb, rAsHb or dAsHb) was coated at a concentration of 1μg/ml overnight at 4°C on Nunc Maxisorp flat-bottomed 96 well plates (Sigma) in 100μl coating buffer (0.05M carbonate/bicarbonate buffer pH9.6). Following incubation, plates were washed 3 times with wash buffer (PBST: 0.05% PBS-tween 20, pH 7.2). Nonspecific binding sites were blocked by dispensing 100 μl of PBS with 5% FCS in each well and incubating the plates for 2h at 4°C. For the inhibition ELISA, an extra blocking step was included where PC-groups on the AsHb were blocked by incubating the coated wells with TEPC-15 antibodies diluted 1:500 in blocking buffer for 2 h. After blocking, plates were washed as before and plasma or antibody samples were added to the wells. Plasma samples were diluted 1:50 in PBST and 100 μl of each sample was tested in duplicate. Plates were incubated for 2h at RT and afterwards washed as previously described. Secondary antibodies (mouse anti-human IgG4 pFc’-HRP (Southern biotech)) were diluted 1:2,000 in blocking buffer and plates incubated for 1h at RT. Finally, plates were washed and 100μl of the o-phenylenediamine dihydrochloride substrate solution (Thermo Fisher Scientific) was added to each well. The substrate reaction was stopped after 10 minutes by adding 50μl of stop solution (4M H2SO4) and optical densities at 490nm were recorded. The cutoff for positivity was calculated as the arithmetic mean OD + 3 times the standard deviation obtained with 10 non-endemic plasma samples from St. Louis, Missouri, USA.
All statistical analyses were performed using GraphPad Prism v6.0 software (La Jolla, CA, USA). Infection prevalence rates at baseline and subsequent time points were compared with the McNemar test. Infection intensities (EPG) and ELISA OD values for paired samples were compared using the Wilcoxon signed rank test. Correlations between EPG and AsHb ELISA OD values were assessed with the Spearman’s rank correlation test. The statistical significance of differences in ELISA OD values obtained with different antigens (AsHb, dAsHb and rAsHb) was assessed with the Wilcoxon signed rank test.
The Ethics Committee of the University of Indonesia, Jakarta approved the sample collection in the Alor Island study as previously described [22]. The Institutional Review Boards at Case Western Reserve University and the Papua New Guinea Medical Research Advisory Committee approved the protocol for sample collection, and all study participants provided informed consent. The Institutional Review Board at Washington University School of Medicine waived the need for an additional review for our use of de-identified human serum samples for this in vitro study.
Genbank: AAA29374.1. Wormbase: GS_08371, ALUE_0001899801, ALUE_0001446901, NECAME_07759 ANCCEY_14143, TCNE_0001552801.
In order to evaluate whether antibodies of infected humans were detecting AsHb, a select number of endemic and non-endemic control plasma samples were first used to evaluate the recognition of AsHb by Western blot and ELISA. Six of 8 endemic plasma samples with proven A. lumbricoides infection (>25 EPG) and 4 of 5 individuals with negative stool examinations had IgG4 antibodies that recognized AsHb (Fig 1). In contrast, none of the non-endemic control plasma samples detected AsHb on Western blot or showed strong reactivity on ELISA. A cut-off for ELISA positivity was determined, based on the OD values of the 10 non-endemic plasma samples that were tested. The cut-off was set at an OD of 0.38.
Specificity of the AsHb IgG4 Western blot was also assessed with plasma samples collected in an area in Papua New Guinea where hookworm infection is nearly universal but ascariasis and trichuriasis are absent. Since 7 of the 12 samples tested had IgG4 antibodies to AsHb (S1 Fig), it appears that some people with hookworm infections develop IgG4 antibodies to AsHb to some degree. A BLASTP search using the AsHb protein sequence obtained from GenBank (AAA29374.1) on Wormbase Parasite revealed two sequences in A. lumbricoides (ALUE_0001899801 and ALUE_0001446901) with >99% amino acid sequence identity. Both sequences form a perfect alignment with AsHb (S2 Fig). Both N. americanus and A. ceylanicum have orthologs to AsHb (NECAME_07759 and ANCCEY_14143 respectively) with sequences that were shorter than the AsHb (88 AA and 97 AA respectively) with sequence identity in the overlapping region of 51.1% and 46.4% and E values of 1e-25 and 2.7e-24 respectively. The ortholog in the canine ascarid Toxocara canis (TCNE_0001552801) has 69.6% amino acid sequence identity and also showed to contain a signal peptide. No AsHb orhtolog (E-value < 1e-05) was present in Trichuris spp. (S1 Table).
This part of the study used ELISA to detect human IgG4 antibodies to AsHb in plasma samples collected before and after MDA. All parasitological and serological data obtained in this study is provided as supplementary information (S2 Table). STH prevalence results before and after MDA are shown in Table 1. Although the prevalence of A. lumbricoides infection decreased significantly from 38.2% to 15.7% after 2 years of MDA (a 58.9% reduction; P < 0.01), the prevalence rate rebounded to 24.4% in 2007 (a 36.1% reduction from baseline; P < 0.01) and to 29.5% in 2009 (a 22.8% reduction from baseline, P = 0.09). In contrast, the average egg output in the community was 81.6% lower in 2009 than in 2002 (128.2 vs. 697.9, P < 0.01). Hookworm and T. trichiura infections were also reduced by MDA (both 1.8% in 2007), however in 2009 hookworm infection rates returned to pre-MDA levels (10.4%) and T. trichiura rates also rebounded (2.3%) [22]. The average community egg outputs for hookworm and whipworm did not change significantly between 2002 and 2009 (5.4 to 6.9 and 3.1 to 1.4 respectively).
A total of 1,066 plasma samples were tested by AsHb ELISA, and these included samples collected prior to any MDA and at intervals following multiple rounds of MDA (Table 1). Employing the earlier determined OD cut-off for the ELISA, 67.6% of the individuals were seropositive at baseline (Table 1, Fig 2 and S3 Fig). After two years of MDA, seroprevalence was significantly reduced to 22.6% in 2004 (P < 0.001) and this was unchanged in 2007 (23.0%). The seropositivity rate in 2009 had further decreased in comparison to 2007 (14.3%, P < 0.05).
In both 2002 and 2009 there was no significant relationship between AsHb ELISA OD values and stool egg counts for any of the STH (S4 Fig). In 2002, 66 of 90 egg positive (73.3%) and 91 of 150 (60.7%) egg negative individuals were seropositive by anti-AsHb ELISA. However, in 2009 only 3 of 20 egg positive (15%) and 4 of 44 (9.1%) egg negative individuals were seropositive.
There was a drastic and significant reduction in both egg prevalence and seroprevalence between 2002 and 2004 in all age groups. Egg prevalence increased again in 2007 and 2009 across all age groups until it almost reached pre-treatment levels whereas seroprevalence did not change after 2004. Similarly, the mean number of eggs excreted by the people in a specific age category was also significantly reduced from 2002 to 2009 over all age categories (Fig 3).
In order to work towards better standardization of the test, we also evaluated whether infected human sera would recognize recombinant AsHb. The AsHb gene was cloned from adult worm cDNA and expressed in E. coli. The protein profile of the purified rAsHb was identical to that of AsHb after Coomassie staining (S5 Fig). Antibodies in a pooled plasma sample from humans with Ascaris infection did not bind to rAsHb by Western blot (Fig 4). To test whether the presence of N-glycan groups on AsHb was important for immune recognition, the native AsHb was deglycosylated with PNGase F to remove any N-linked glycans. A shift in molecular weight was seen in the PNGase F deglycosylated AsHb (dAsHb), indicating the removal of N-linked glycans. This molecular weight shift was not visible after treatment of the rAsHb with PNGase F indicating the absence of PNGase F digestible carbohydrate groups (S5 Fig). Antibodies in pooled plasma from A. lumbricoides infected individuals did not bind to dAsHb by Western blot.
In order to quantify this effect, IgG4 reactivity of 10 plasma samples from Indonesian individuals from 2002 to AsHb, denatured AsHb, dAsHb and rAsHb were evaluated by ELISA (Fig 4). After normalization of the data using the AsHb OD 490 as reference, a significant relative increase in antibody binding (59%, P < 0.05) was seen after denaturing the AsHb. The opposite was true for dAsHb and for rAsHb that were less immunoreactive than the native antigen (-42%, P < 0.05 and -88%, P < 0.01, respectively.
In order to further characterize the glycans present on AsHb that were important for immune-recognition of the antigen, the N-glycan structures were removed from the protein backbone by PNGase F and analyzed by MALDI-TOF-MS. The glycan spectrum of the released N-glycans of AsHb is shown in Fig 4. Major [M-H]- signals were observed at m/z 1176.9, 1380.0 and 1583.2, derived from glycans with the compositions F1N2H3, F1N3H3 and F1N4H3 (F, fucose; N-acetylhexosamine, H, hexose) glycans, respectively, interpreted as α1,6-fucosylated trimannosyl N-glycan core structures substituted with 0–2 GlcNAc residues (indicated in Fig 4). An additional major signal was observed at m/z 1545.1 (F1N3H3+165.1) indicative for the phosphorylcholine (PC) substituted variant of F1N3H3 which is in line with previous observations for N-glycans of A. suum [27]. Further, minor signals at m/z 1748.3 (F1N4H3PC), 1786.3 (F1N5H3) and 1913.4 (F1N4H3PC2) were observed indicating that additional substitutions with HexNAc residues and PC can occur. Incubation of both AsHb and dAsHb with mouse anti-PC monoclonal antibodies (TEPC-15) proved the presence of PC in native AsHb and its absence after deglycosylation with PNGase F (S6 Fig). The recognition of AsHb by human IgG4 antibodies was not significantly reduced when AsHb was pre-incubated with anti-PC antibodies before addition of positive human plasma samples (S6 Fig).
The results of this study suggest that measurement of antibodies to AsHb may be a useful approach for assessing exposure of human populations to A. lumbricoides infection. Egg excretion per person and antibody rates decreased in parallel following MDA while the infection prevalence rate in 2009 was not very different from the baseline rate. The implementation of MDA probably reduced the number of Ascaris eggs that were excreted in the environment. As a result, the ingestion of infective eggs and exposure to migrating stages of the parasite is likely to have decreased in all age categories in the population. This may explain why the antibody rate decreased further between 2007 and 2009 despite suspension of the MDA program after 2007. However, since we do not have quantitative coprological data for the years 2004 and 2007, we do not have precise information on the impact of MDA on infection intensities between 2002 and 2007.
Despite the limited sequence similarity between the AsHb orthologs identified in both hookworm species, our results confirm antigenic cross reactivity between hookworm and A. lumbricoides [28]. Thus, reduced antibody reactivity to AsHb after MDA may have been partly due to the effects of MDA on hookworm prevalence or transmission. However, the significant rebound in hookworm prevalence that occurred between 2007 and 2009 (from 1.8% to 10.4%) was not associated with a rise in antibody rates to AsHb. This study did not investigate antibody responses to AsHb in people infected with T. trichiura. However, the absence of an AsHb homologue in the Trichuris genome and the fact that antibodies in sera from pigs experimentally infected with T. suis had little if any reactivity with AsHb [16] suggest that humans infected with T. trichiura who have not been infected with Ascaris are not likely to have significant antibody responses to AsHb.
We acknowledge that using the AsHb ELISA would be too insensitive to identify or diagnose active A. lumbricoides infection in an individual. However, it is important to notice that the goal of our study was to evaluate the use of this serological test as diagnostic marker for exposure on a community level and not for individual diagnosis of infection. Unlike for A. suum infections in pigs, where experimentally infected pigs can serve as “true gold standard”, there is no such standard available to assess the sensitivity of a serological tool for diagnosis of STH infections in humans [6]. However, since this ELISA would be used to assess transmission intensity on a community level, this is not a major obstacle.
Recombinant antigens are often preferred for serodiagnosis of parasitic infections. One reason for this is that native antigens are sometimes difficult to obtain and purify. However, in the case of AsHb, the antigen is relatively easy to purify from A. suum. In addition, antibodies from infected individuals were only weakly reactive with recombinant AsHb produced in this study or with AsHb after treatment with PNGase F. This finding suggests that IgG4 human antibodies to AsHb are mostly directed against N linked glycan epitopes present on the native antigen. Shared carbohydrate epitopes pose a challenge for developing specific serology tests for helminth infections. Parasites with AsHb orthologs with low amino acid sequence identity may contain the same or similar carbohydrate epitopes that could be responsible for antigenic crossreactivity.
Mass spectrometric analysis of the glycans released by PNGase F treatment of AsHb detected PC-substituted GlcNAc moieties. Blocking the PC-group with TEPC-15 antibodies did not result in a significant blocking of binding of human antibodies to AsHb by ELISA. This result suggests that other glycans motifs may be important antigenic components of AsHb and it is consistent with a previous study that reported that humans do not develop IgG4 subclass antibodies to the PC epitope [29].
The simplicity, affordability and speed of the Kato Katz test has made it the most widely used method for estimating STH infection rates and intensities in large scale control programs [30]. This has important policy implications, because current WHO guidelines for STH preventive chemotherapy are based on infection prevalence rates as assessed by a single Kato-Katz smear with little attention paid to intensity [31]. However, prevalence is not everything. Because of the high degree of aggregation of STH infections, significant reductions in average worm loads may result in small or unnoticeable changes in prevalence rates [32]. Also, infection intensities for ascariasis are often reported according to the broad WHO categories of light (< 5,000 EPG), moderate (5,000–50,000 EPG) or heavy (>50,000 EPG). However, it is unclear whether these categories are appropriate for use in all endemic regions, because of considerable geographic variability in egg production per adult female Ascaris worm [33].
The use of diagnostic tools that are based on the detection of Ascaris eggs or DNA in the stool has significant shortcomings when it comes to the accurate estimation of true prevalence or intensity of the infection. For one thing, only a small fraction of the total number of parasite larvae that migrate through the body ever develop into adult worms in the intestine. For worms that reach the intestine, egg output per worm can vary widely because of different parasite sex ratios, the age distribution of the adult worms, and host immunity [7]. Thus the absence of A. lumbricoides eggs in the stool does not necessarily prove there has been no recent infection or exposure with larval stages. Animal studies have shown that larval stages contribute significantly to morbidity caused by Ascaris infections [34, 35], and this should be considered as part of the health impact of ascariasis in humans as well.
Our results show that IgG4 antibody responses to AsHb do not correlate with A. lumbricoides egg output. This is consistent with results from other studies that have compared anti-Ascaris antibody responses to adult worm counts or EPG in stool [12, 36]. Furthermore, several studies have shown that anti-Ascaris antibody responses are rather dependent on exposure and infection intensity as opposed to being protective or predictive of future levels of infection [36–39].
A significant percentage of people in Alor Island who had Ascaris eggs in their stool lacked IgG4 antibodies to AsHb, and this situation became more common later in the trial when community egg loads were reduced. It is interesting that a number people with more than 500 EPG were seronegative in 2009. While there are several possible explanations for this, we favor the idea that antibody responses to AsHb are stimulated much more by new infections when larvae are migrating through tissues and than by the presence of low numbers of adult worms in the intestinal lumen. When the infection pressure has been reduced to low levels, exposure to new infections with migrating larvae may not be sufficient to induce or maintain IgG4 antibodies to AsHb. Furthermore, experimental infections in pigs have already shown that the number of adult worms in the gut seems almost inversely correlated with the number of eggs ingested [40, 41]. Hence, with lower infection intensities, the chance of larvae establishing in the gut and developing into adult worms increases. It would be interesting to elucidate the effect of infection dose on anti-AsHb responses in pigs, and this might help to explain changes in antibody rates in human populations when infection pressure declines following MDA.
Although AsHb serology is not sensitive for detecting active Ascaris infections in individuals, it appears to be a promising new tool for quantifying exposure to Ascaris infection at the community level. That is to say, it may provide a useful measure of egg input and incoming infection in communities even though it is not sensitive for predicting the presence of eggs in the stool in individuals. Serological surveys for antibodies to AsHb and other STH antigens could be an attractive alternative to stool examination for integrated post-MDA surveillance programs for lymphatic filariasis (LF) and STH, because post-MDA surveys for LF collect finger prick blood samples for use in point of care serology tests [3], and finger prick blood could also be used for STH serology. AsHb serology could also be used as an alternative to Kato-Katz test for mapping the distribution of STH infections, because it should be useful for identifying areas with high transmission rates that have the highest need for intervention.
The crossreactivity with hookworm and possibly Strongyloides stercoralis and Toxocara spp. limits the value of AsHb serology if one is interested in ascariasis alone. Work to develop a more species-specific antibody assay is ongoing. However, hookworm and Ascaris crossreactivity may not be a major flaw for the practical use of AsHb serology, since the same drugs are used to treat both of these infections. We believe that this study has provided a useful proof of principle for the value of antibody serology as an epidemiological tool for assessing STH transmission pressure in populations and for monitoring the impact of STH intervention on transmission pressure. We have shown that AsHb antibody rates correlate well with egg output per person in populations, and they are also likely to correlate well with recent egg input in individuals. However, additional research is needed on this topic. First, it will be necessary to confirm the findings from this study with samples from other areas with high rates of ascariasis. Second, it would also be interesting to evaluate changes in serology over time in different subpopulations or to use serology to try and pinpoint when children are first exposed to STH.
|
10.1371/journal.pgen.1001264 | Genome Sequencing and Comparative Transcriptomics of the Model Entomopathogenic Fungi Metarhizium anisopliae and M. acridum | Metarhizium spp. are being used as environmentally friendly alternatives to chemical insecticides, as model systems for studying insect-fungus interactions, and as a resource of genes for biotechnology. We present a comparative analysis of the genome sequences of the broad-spectrum insect pathogen Metarhizium anisopliae and the acridid-specific M. acridum. Whole-genome analyses indicate that the genome structures of these two species are highly syntenic and suggest that the genus Metarhizium evolved from plant endophytes or pathogens. Both M. anisopliae and M. acridum have a strikingly larger proportion of genes encoding secreted proteins than other fungi, while ∼30% of these have no functionally characterized homologs, suggesting hitherto unsuspected interactions between fungal pathogens and insects. The analysis of transposase genes provided evidence of repeat-induced point mutations occurring in M. acridum but not in M. anisopliae. With the help of pathogen-host interaction gene database, ∼16% of Metarhizium genes were identified that are similar to experimentally verified genes involved in pathogenicity in other fungi, particularly plant pathogens. However, relative to M. acridum, M. anisopliae has evolved with many expanded gene families of proteases, chitinases, cytochrome P450s, polyketide synthases, and nonribosomal peptide synthetases for cuticle-degradation, detoxification, and toxin biosynthesis that may facilitate its ability to adapt to heterogenous environments. Transcriptional analysis of both fungi during early infection processes provided further insights into the genes and pathways involved in infectivity and specificity. Of particular note, M. acridum transcribed distinct G-protein coupled receptors on cuticles from locusts (the natural hosts) and cockroaches, whereas M. anisopliae transcribed the same receptor on both hosts. This study will facilitate the identification of virulence genes and the development of improved biocontrol strains with customized properties.
| Aside from playing a crucial role in natural ecosystems, entomopathogenic fungi are being developed as environmentally friendly alternatives for the control of insect pests. We conducted the first genomic study of two of the best characterized entomopathogens, Metarhizium anisopliae and M. acridum. M. anisopliae is a ubiquitous pathogen of >200 insect species and a plant growth promoting colonizer of rhizospheres. M. acridum is a specific pathogen of locusts. Important findings of this study included: 1) Both M. anisopliae and M. acridum have a very large number of genes encoding secreted proteins, and many of these play roles in fungus-insect interactions. 2) M. anisopliae has more genes than M. acridum, which may be associated with adaptation to multiple insect hosts. 3) Unlike M. acridum, the M. anisopliae genome contains many more transposase genes and shows no evidence of repeat-induced point mutations. The lack of repeat-induced mutations may have allowed the lineage-specific gene duplications that have contributed to its adaptability. 4) High-throughput transcriptomics identified the strategies by which these fungi overcome their insect hosts and achieve specificity. These genome sequences will provide the basis for a comprehensive understanding of fungal–plant–insect interactions and will contribute to our understanding of fungal evolution and ecology.
| Most fungi with sequenced genomes are plants pathogens or saprophytes. However, there are also thousands of entomopathogenic fungal species that play a crucial role in controlling insect populations. The genus Metarhizium includes the best studied entomopathogenic fungi at the molecular and biochemical level. They have a world-wide distribution from the arctic to the tropics and colonize an impressive array of environments including forests, savannahs, swamps, coastal zones and deserts [1]. Metarhizium species are amongst the most abundant fungi isolated from soils with titers reaching 106 conidia per gram in grasslands [2]. The genus contains M. anisopliae, which has a broad host range, as well as specialists, such as the locust-specific pathogen M. acridum. These two species in particular have emerged as excellent model organisms to explore a broad array of questions in ecology and evolution, host preference and host switching, and to investigate the mechanisms of speciation. In addition, both M. anisopliae and M. acridum have been at the forefront of efforts to develop biocontrol alternatives to chemical insecticides in agricultural and disease-vector control programs, and many commercial products are on the market or under development [2]–[4].
Our knowledge of the ecological impact of M. anisopliae and its potential as a biocontrol agent has recently been enhanced by the discovery that it colonizes plant roots where it may simultaneously act as a biofertilizer and biopesticide to boost plant growth [5]. Consistent with its broad lifestyle options, M. anisopliae exhibits an extremely versatile metabolism, enabling growth under various environmental conditions, with sparse nutrients and in the presence of compounds lethal to other fungi [6]. As the asexual stages (anamorphs) of medicinally valued Cordyceps spp. [7], Metarhizium spp. are prolific producers of enzymes and diverse secondary metabolites with activities against insects, fungi, bacteria, viruses and cancer cells [6], [8], [9]. In addition, the enzymes from Metarhizium spp. are frequently exploited as industrial catalysts [10], [11]. M. anisopliae has also been used in studies on the immune systems of invertebrate model hosts to provide insights into emerging human pathogens [12], and it is a developing model for studies on aging [13], [14].
In contrast to the versatile M. anisopliae, the specialist M. acridum is specific for certain locusts and grasshoppers [15]. However, like M. anisopliae, it is a producer of diverse cell types (e.g., conidia, hyphae, appressoria, unicellular blastospores, and multi-cellular hyphal bodies) that facilitate the infection of target insects via adhesion and penetration of the host cuticle, proliferation within tissues and the haemolymph, and eventual eruption through the host cadaver (Figure 1). M. acridum is mass produced and used on a large scale for locust control [16], whereas few other biological control agents have been such a commercial success because of poor efficacy compared to chemicals [17].
Although recent advances have identified the functions of several pathogenicity genes [18]–[22] and technical developments improved the virulence of M. anisopliae [23], [24], the need to understand these fungi and expand their biotechnological potential requires sequenced genomes of M. anisopliae and M. acridum. Sequencing two related species that have evolved very different lifestyles will increase their utility as models, and provide insights into the evolution of pathogenicity. Such sequences will also allow for more rapid identification of genes encoding biologically active molecules and genes responsible for interactions between fungi, plants and insects. These findings could be further translated into the development of improved strains with customized properties that could potentially function as comprehensive plant symbionts to improve plant establishment and sustainable agriculture, particularly on marginal lands.
The genomes were each shotgun sequenced to ∼100× coverage. The M. anisopliae genome (strain ARSEF 23) was assembled into 176 scaffolds (>1 kb; N50, 2.0 Mb) containing 1,271 contigs with a total size of 39.0 Mb (loci tagged as MAA). The M. acridum genome (strain CQMa 102) was assembled into 241 scaffolds (>1 kb; N50, 329.5 kb) containing 1,609 contigs with a genome size of 38.0 Mb (loci tagged as MAC) (Table 1). These assemblies closely correspond to the genome sizes of other Ascomycetes (Table S1). By mapping >6,000 unique expressed sequenced tagged sequences to the scaffolds, each genome was estimated to be >98% complete. M. anisopliae and M. acridum were predicted to have 10,582 and 9,849 protein coding genes, respectively, which is similar to the coding capacity of other Ascomycetes (Table S1). We examined homology relationships between M. anisopliae and M. acridum, and a set of eight other ascomycete genomes (Figure 2A). The results indicated that ∼90% of the genes in both Metarhizium genomes have homologs (E≤1×10−5) in other Ascomycetes. In addition, M. anisopliae has 398 (3.8%) genes with matches restricted to M. acridum (Metarhizium-restricted genes) and 263 (2.5%) orphan sequences. M. acridum has 219 (2.2%) orphan sequences (Figure 2A). Further analysis of the M. anisopliae orphans showed that 21.3% had matches in bacteria, 3.4% in animals and 3.8% in viruses. Similarly, 13.3%, 5.5% and 2.7% of the M. acridum orphans had matches in bacteria, animals and viruses, respectively, consistent with possible horizontal gene transfer events.
The proportion of genes encoding secreted proteins is remarkably large, being 17.6% (1,865 proteins) in M. anisopliae and 15.1% (1,490 proteins) in M. acridum as compared to 7–10% in plant pathogens [25] and ∼5% in N. crassa [26] or A. nidulans [27]. As expected, many of the secreted proteins are in families which could have roles in colonization of insect tissues, such as proteases (Table S2). However, 32.2% of M. anisopliae and 28.7% of M. acridum secreted proteins had no conserved domains or functionally characterized homologs. Of these, ∼22% were Metarhizium-restricted genes and ∼4% were orphan genes in either genome.
Pairwise comparison indicated that the two Metarhizium genome structures have large areas of synteny (Figure 2B, Figure S1A). The lineage specific regions of M. anisopliae and M. acridum contain high densities of transposases, species-specific genes, genes encoding proteins with unknown functions and pseudogenes (Figure S1B). Similar lineage-specific regions were found in Fusarium spp. [28]. Ninety nine percent of the M. anisopliae genome comprises non-repetitive sequences, and the orthologs shared with the M. acridum genome display an average 89.8% amino acid identity. The two Metarhizium species are therefore more closely related than the three Aspergillus species A. nidulans, A. fumigatus and A. oryzae which share only 68% average sequence identity [29]. A phylogenomic analysis revealed that M. anisopliae and M. acridum lineages diverged about 33–43 million years (MY) ago and are most closely related to the mutualistic plant endophyte Epichloe festucae (divergence time 88–114 MY) and to the wheat head blight fungus Fusarium graminearum (divergence time 144–187 MY) (Figure 2C).
The specialist M. acridum harbors more repetitive elements than M. anisopliae but the latter has many more transposases (Table S2). Most of these are DNA transposases (97/148 in MAA; 12/20 in MAC), with subclasses hAT (45/97) and Helitron (26/97) being particularly abundant in M. anisopliae. The Copia (17) and LINE (26) retrotransposons are also abundant in the genome of M. anisopliae, while M. acridum has only three LINE elements and does not contain Copia (Figure 3A). Transcriptome analysis (see below) showed that most (>65%) of the transposase genes were transcribed by the Metarhizium hyphae during the infection process (Table S3).
The number of putative transposases in the M. acridum genome is lower by at least a factor of five than in most Ascomycetes, including M. anisopliae (Table S2). This could be explained by repeat induced point mutations (RIP) introducing CpG to TpA transitions in duplicated sequences during the sexual cycle [30]. This mutational bias is observed in M. acridum (RIP index, 2.17) but not in M. anisopliae (RIP index, 1.09) (Figure 3B). Consistent with Neurospora crassa which has efficient RIP [31], the genome of M. acridum contained twice as many duplicated pseudogenes (254 versus129) as did that of M. anisopliae. The M. anisopliae genome contains more processed and fragmented pseudogenes caused by mobile elements (234 versus 186), consistent with transposons making a greater contribution to genetic instability in M. anisopliae (Table S4). The production of stable biocontrol agents for commercialization might therefore benefit from disabling transposable elements.
An InterproScan analysis identified 2,710 protein families (containing 7,178 proteins) in M. anisopliae and 2,658 families (containing 6,615 proteins) in M. acridum. A stochastic birth and death model [32] showed that relative to M. acridum, 42 families including transporters, transcription factors, cytochrome P450s, proteases and lipases were expanded and three families (protein kinase, aminotransferase and transpeptidase) were contracted in M. anisopliae (Table S5). This resulted in M. anisopliae having more genes in most functional categories except for those involved in signal transduction (Figure 4, Table 2).
To find potential virulence-associated genes, a whole genome blast analysis was conducted against the pathogen-host interaction (PHI) gene database, a collection of experimentally verified pathogenicity, virulence and effector genes from fungi, oomycetes and bacteria [33]. We identified 1,828 putative PHI genes in M. anisopliae (17.3% of its genes, belonging to 383 protein families) and 1,629 putative PHI genes in M. acridum (16.5%, 371 families), of which 1,331 genes were orthologous. Although there are no entries from entomopathogenic fungi in the PHI-base, we proceeded on the assumption that the proof of pathogenicity/virulence of a gene in one fungus also suggests a pathogenicity/virulence function in other fungi [34]. In accordance with this assumption, the search of the PHI database yielded several already characterized M. anisopliae pathogenicity determinants, including subtilisins (see below) and hydrophobins (small cell wall proteins) that have pleiotropic functions in M. anisopliae including attachment of spores to hydrophobic surfaces [35]. The class 2 (MAA_01182 and MAC_09507) and class 1 (MAA_10298 and MAC_04376) hydrophobins had significant similarity with PHI sequences from plant pathogenic fungi. The previously characterized adhesin, MAD1 (MAA_03775) required for specific binding to insect host surfaces [20], resembled EAP1 (PHI acc: 517) from the human pathogen Candida albicans. However, the adhesin MAD2 (MAA_03807) required for binding to plant surfaces [20], had no significantly similar sequence in the PHI database. Orthologs to both MAD1 (MAC_00987) and MAD2 (MAC_00953) were found in the M. acridum genome.
Using the PHI-base content with a focus on ascomycetes, Sexton and Howlett found many parallels in the infection mechanisms used by plant and animal pathogens [36]. To determine how many plant pathogen PHI genes are also found in Metarhizium, we screened the F. graminearum and M. oryzae genomes against the PHI-base and identified 2,053 genes (in 398 families) and 1,713 genes (in 427 families), respectively, representing about 16% of gene contents in these two fungi (Table S6). Approximately, 70% of these genes are orthologous to PHI sequences in M. anisopliae and M. acridum. Fewer Metarhizium orthologs were found in animal pathogenic fungi such as C. albicans, which could be explained by Metarhizium being more closely related to plant pathogens (Figure 2C) as well as the animal pathogens lacking appressoria (infection structures) during host penetration [4].
Insect pathogens such as Metarhizium spp. need to penetrate the protein-chitin rich insect cuticle and solubilize host tissues for nutrition. Therefore, they would be expected to secrete large numbers of degradative enzymes. Indeed, M. anisopliae has more genes encoding secreted proteases (132) than other sequenced fungi (Table S2). The trypsin family has the highest relative expansion among the proteases with 32 genes in M. anisopliae, almost twice as many as M. acridum and 6 to 10 times as many as the other taxa evaluated (Figure 5A, Table S2). A chymotrypsin (MAA_07484) that might have been imported from bacteria through horizontal gene transfer [37] and two trypsins that were recently duplicated in M. anisopliae (MAA_05135 and MAA_05136) are missing from the M. acridum genome (Table S7). Subtilisins (55 in MAA and 43 in MAC; 7 to 31 in other fungi) (Figure 5B, Table S8) and aspartyl proteases (33 in MAA and 25 in MAC; 9 to 21 in other fungi) (Table S9) are also expanded in M. anisopliae due to lineage-specific duplications (Figure S1C). Most of the Metarhizium subtilisins (48 in MAA and 37 in MAC) and aspartyl proteases (27 in MAA and 23 in MAC) had significant matches in the PHI-base. Subtilisins assist in the infection processes of M. anisopliae by degrading host cuticles, providing nutrition and disabling antimicrobial peptides [38]. The importance of Metarhizium aspartyl proteinases has not been demonstrated but they resemble the aspartyl proteases that assist the human pathogen C. albicans by degrading cell surface molecules [39].
Many plant pathogens need glycoside hydrolases, pectate lyases and cutinases to degrade the plant cuticle (waxy layer) and cell wall. The number of glycoside hydrolases (GH) possessed by M. anisopliae (156) and M. acridum (140) is close to the average for plant pathogenic fungi (150) (Table S10). However, only ∼20% of the Metarhizium GH genes (36 in MAA and 29 in MAC) were similar to PHI catalogued sequences as compared to 44% (70 genes) in F. graminearum and 29% (57 genes) in M. oryzae (Table S6). The plant pathogens in particular have additional GH3 cellulases while Metarhizium spp. lack the GH11 family of xylanases. GH3 and GH11 family genes are catalogued in the PHI-base. Overall, fewer genes were associated with plant utilization in Metarhizium than in plant pathogens. This included fewer putative cutinases (2 in Metarhizium spp. 8 to 18 in plant pathogens) and pectate lyases (7 in Metarhizium spp.; 9 to 25 in plant pathogens). However, the GH16 family of xyloglucan/xyloglucosyl transferases involved in decomposition of plant cell walls is well represented in the Metarhizium genomes (18 in MAA and 16 in MAC; 6 to 16 in plant pathogens) (Table S10). More predictably, GH18 chitinases involved in the digestion of insect cuticle chitin [40], are over represented in Metarhizium (30 in MAA and 21 in MAC; 5 to 14 in plant pathogens) (Figure 5C, Table S6). The few chitinases included in the PHI database are involved in fungal developmental processes, as chitin is not a substrate found in animal and plant hosts.
The number of genes for secreted lipases (12 in MAA, 5 in MAC) is well above the average found in other fungi, and 9 M. anisopliae and 5 M. acridum lipases showed significant similarity to genes in the PHI-base, as compared to 3 lipases each in F. graminearum and M. oryzae (Table S6). The role of individual Metarhizium lipases in pathogenicity has not been demonstrated, although a lipase activity inhibitor blocks infection processes in M. anisopliae [41]. Lipases MAA_03127 and MAC_09232 showed best-hit relationships with an extracellular lipase FGL1 (PHI acc: 432) that is a virulence factor in F. graminearum [42].
Metarhizium genomes encode a large number of transporters (484 in MAA and 441 in MAC) (Table S11). Most transporters belong to the major facilitator superfamily (MFS) (269 in MAA; 236 in MAC) but the ATP-binding cassette (ABC) is also well represented (56 in MAA; 51 in MAC) (Table S11). Most of the ABC transporters (52/56 in MAA and 46/51 in MAC) and many of the MFS transporters (124/269 in MAA and 96/236 in MAC) were similar to genes catalogued in the PHI database (Table S6). The ABC transporters are usually implicated in defending the pathogen from host-produced secondary metabolites, whereas MFS proteins are typically involved in the transport of a wide range of substrates and may function as nutrient sensors [43]. Interestingly, both Metarhizium species have more amino acid and peptide transporters than do other fungi (60 in MAA and 57 in MAC; 29 to 38 in other fungi), consistent with their being able to access a range of protein degradation products from insect sources. Homologs of these genes are absent from the PHI database. The only Metarhizium transporter with an experimentally determined function is the sucrose and galactoside transporter MRT (belonging to the MFS superfamily), which is required by M. anisopliae for rhizosphere competence but not for virulence [44]. There are 6 MRT homologs in M. anisopliae and 5 in M. acridum but 12 in F. graminearum and 26 in M. oryzae, suggesting these genes could be generally important for establishing plant-fungus relationships.
Additional evidence about lifestyle could be found in the relatively large number of genes involved in detoxification in both Metarhizium genomes (Table 2, Table S2) as these potentially contribute to interactions with insect hosts (Table S6). However, families of dehydrogenases, acyl-CoA N-acetyltransferases, monooxygenases and cytochrome P450s (CYP) were preferentially expanded in M. anisopliae relative to M. acridum (Table 2, Table S5). One third of the dehydrogenases (92/271 in MAA and 80/236 in MAC) were putative PHI genes (Table S6). M. anisopliae was particularly enriched in zinc-containing alcohol dehydrogenases (17 in MAA; 7 in MAC) required for the biosynthesis of mannitol, a crucial factor for stress tolerance and virulence in the animal pathogen Cryptococcus neoformans [45]. The monooxygenases in particular might be involved in rapid elimination of insect polyphenolics by ortho-hydroxylation of phenols to catechols [46].
The genome of M. anisopliae encodes 123 highly divergent CYP genes verses 100 CYPs in M. acridum (Figure 5D, Table S12). Ninety of the M. anisopliae CYPs and 69 of the M. acridum CYPs are similar to sequences in the PHI-base (Table S6). CYP genes are involved in oxygenation steps during alkane assimilation and the biosynthesis of secondary metabolites as well as with detoxification [47]. M. anisopliae efficiently metabolizes the alkanes that are a major component of the surface layer of the insect cuticle (epicuticle) [48]. Although the CYP52 subfamily is particularly important for alkane oxidation [49], M. anisopliae has only a single CYP52 (MAA_06634) compared to four in M. acridum (Table S12). However, MAA_06634 and its ortholog in M. acridum (MAC_09267) were highly expressed (see below) by M. anisopliae and M. acridum when infecting either cockroach or locust cuticles (Figure S5A). The other CYP genes up-regulated on cuticles were mostly involved in detoxification. M. anisopliae and M. acridum are predicted to contain four and two CYP504s, respectively. CYP504s are used by fungi to degrade phenylacetate [50], an antimicrobial compound found in plants and insects [51]. The subfamily CYP53 is also represented in the PHI database as it is responsible for detoxification of benzoate and its derivatives [52]. M. anisopliae and M. acridum have two and one CYP53 genes, respectively.
The subfamily CYP5081 involved in biosynthesis of helvolic acid, an antibiotic toxin [53], consists of four closely localized CYP loci (PHI genes) in M. anisopliae (MAA_06585, MAA_06586, MAA_06589 and MAA_06593) that are absent in M. acridum. All four CYP5081 genes were expressed by M. anisopliae infecting cuticles (Figure 5D). Both M. anisopliae and M. acridum have three CYP genes putatively encoding lipid dioxygenases (CYP6001: MAA_04954 and MAC_00208; CYP6002: MAC_05834; CYP6003: MAA_03481 and MAC_00918; CYP6004: MAA_0003) and two lipoxygenases (MAA_06278 and MAA_01260; MAC_01254 and MAC_9416). Oxylipins, the end products of these genes, allow Aspergillus nidulans to colonize plant seeds [54], and seeds are also a habitat for M. anisopliae [55], implying that a similar strategy is employed by Metarhizium to establish plant-fungus relationships.
M. anisopliae is a prolific producer of secondary metabolites including insecticidal destruxins [56], but with the exception of the serinocyclins [57] and NG-391 [58], the genes involved in their biosynthesis are unknown. However, diagnostic genes for secondary metabolite production include those encoding polyketides and non-ribosomal peptides (the most prominent classes of fungal secondary metabolites), as well as those responsible for modifications of the core moiety (a peptide or polyketide) such as genes encoding dehydrogenases, methyltransferases, acetyl transferases, prenyltransferases, oxireductases and CYPs [36]. Consistent with expressed sequence tag studies [59], M. anisopliae appears to possess a much greater potential for the production of secondary metabolites than M. acridum or most other fungi (Tables S2 and S13). The M. anisopliae genome encodes 14 putative non-ribosomal peptide synthases (NRPS), 24 polyketide synthases (PKS) and 5 NRPS-PKS hybrid genes, which is more than M. acridum (13 NRPS genes, 13 PKS genes and 1 NRPS-PKS hybrid) and the average in other Ascomycetes (7 NRPS, 12 PKS genes and 1 NRPS-PKS) (Table S13). NRPSs and PKSs are strongly associated with pathogenicity in many plant pathogenic fungi and are well represented in the PHI database. As in other fungi, Metarhizium NRPS and PKS genes were often clustered together with genes that modify their products. One cluster suggests that Metarhizium might produce prenylated alkaloids (Figure S2). M. anisopliae possesses putative NRPS-like antibiotic synthetases (MAA_08272) consistent with defending the cadaver against microbial competitors. It also possesses a putative bassianolide synthetase (MAA_07513), a virulence factor of the insect pathogen Beauveria bassiana [60]. The NRPS-like proteins MAA_07148 and MAC_06316 are most similar to ACE1, a PKS/NRPS hybrid that confers avirulence to M. grisea during rice infection [61]. M. anisopliae NRPS MAA_00969 is similar (43% identity) to HTS1, the key enzyme responsible for the biosynthesis of the host-selective HC-toxin that confers the specificity of Cochliobolus carbonum to maize [62]. Sixteen out of 24 PKS and 5/14 NRPS genes in M. anisopliae are species specific versus 4/13 PKS and 5/13 NRPS in M. acridum, suggesting lineage specific expansion of these families in both Metarhizium species. However, it is reassuring for present and future commercialization of these fungi that we found no orthologs of genes for the biosynthesis of the human mycotoxins gliotoxin and aflatoxin.
To recognize and adapt to invertebrate environments such as the insect cuticle, hemolymph and cadaver, Metarhizium spp. need to rapidly respond to changes in nutrient availability, osmolarity and the host immune system [18], [63]. In Magnaporthe, the Pth11-like G-protein coupled receptor (GPCR) is a PHI gene (PHI-base acc: 404) because it mediates cell responses to inductive cues [64]. M. anisopliae and M. acridum have 54 and 40 putative PTH11-like GPCRs, respectively compared to an average of 32 in other fungi (Table S2, Table S14). The Metarhizium sequences could be grouped into six subfamilies (Figure S3). G protein alpha subunits have been extensively studied in fungi and many are required for pathogenicity because they transduce extracellular signals leading to infection-specific development [65]. Distinct roles for three G protein alpha subunit genes have been revealed in M. grisea, A. nidulans and N. crassa. A fourth G-alpha protein has been identified in the plant pathogens Stagonospora nodorum (SNOG_06158) [66], Ustilago maydis (UM05385) [67], and the saprophyte A. oryzae (BAE63877) [68]. Each of the Metarhizium genomes also contain four G-alpha genes. The genes MAA_03488 and MAC_04984 show best hits (>30% similarity) with SNOG_06158, UM05383 and BAE63877, suggesting they may be orthologous. SNOG_06158 is the most highly up-regulated S. nodorum G-alpha gene in planta [66]. Likewise, MAA_03488 and MAC_04984 are the most highly expressed G-alpha genes during infection of either cockroach or locust cuticles (see below, Table S20).
The chief mechanism used by bacteria for sensing their environment is based on two conserved proteins: a sensor histidine kinase (HK) and an effector response regulator (RR) that functions as a molecular switch controlling diverse activities. In fungi, two component pathways mediate environmental stress responses and hyphal development [69]. M. anisopliae and M. acridum have 10 and 9 histidine kinases, respectively compared to 3 to 20 in other fungi (Table S2).
To regulate cell function, M. acridum has 192 protein kinases as compared to 161 in M. anisopliae which is still above the average (143) found in other fungi (Tables S5 and S15). Much of the M. acridum expansion involves cyclin dependent and cell division control kinases, suggesting that M. acridum has a particularly complex signal transduction cascade controlling cell division. As signal transduction is a critical part of fungal development and infection processes, and accordingly most of the kinases had orthologs in the PHI database (124/161 in MAA and 137/192 in MAC). The high frequency of pseudogenes among kinases (M. acridum, 1:6; M. anisopliae, 1:8), compared to transporters (M. anisopliae, 1:82; M. acridum, 1:33) and other gene families suggests that protein kinases have a particularly high rate of turnover (Table S16). Differentially lost genes tend to function in accessory roles so these kinases might have had redundant functions in signal transduction that changed rapidly under strong selective constraints.
Following signaling transduction, physiological responses are regulated by activation of different transcription factors. M. anisopliae has 510 putative transcription factors compared to 439 in M. acridum, the difference being largely due to M. anisopliae having more C2H2 zinc finger and Zn2/Cys6 transcription factors (Tables S5 and S17). These families are also expanded in some Aspergilli, where the characterized examples are involved in regulating diverse aspects of primary and secondary metabolism, including protein and polysaccharide degradation [70]. The cAMP response element-binding (CREB) protein, a basic leucine zipper transcription factor (bZIP), is a major downstream transcription factor for cAMP/PKA pathways in mammals [71]. CREB has not been characterized in fungi; however, our transcriptome data shows that a putative bZIP transcription factor (MAA_02048 or MAC_02758) is highly expressed by each Metarhizium species coincident with up-regulation of protein kinase A (see below). The physiological role(s) of MAA_02048 are currently under investigation.
Insect bioassays confirmed that M. acridum killed locusts but not cockroaches, while M. anisopliae killed both insects (Figure S4). In order to identify the putative signal transduction and metabolic pathways involved in formation of infection structures, we used RNA-Seq to compare transcriptional responses of M. anisopliae and M. acridum to infection of the optically clear hind wings of adult locusts and cockroaches, respectively. A time period of 24 hours was chosen to focus on the crucial processes involved in prepenetration growth e.g., adhesion to epicuticle, germination and production of appressoria [72].
After sequencing >2.5 million tags for each treatment, it was calculated that >82% of predicted M. anisopliae genes and >88% of predicted M. acridum genes were expressed during pre-penetration growth. This included more than 80% of the M. anisopliae and M. acridum genes with sequences similar to those in the PHI database (Table S18). Germination and growth by M. anisopliae and M. acridum on either insect triggered high level expression of genes associated with translation (e.g., ribosomal proteins) and post-translational modifications (e.g., heat shock proteins) (Figure S5, Table S19). However, otherwise, the two fungi differed greatly from each other in their transcriptional responses to each cuticle, and to a lesser extent the two cuticles elicited different responses from each fungus (Figure 6, Figure S6). The orthologs of many differentially expressed genes are involved in appressorial formation and function in plant pathogens (Table S19), including Cas1 from Glomerella cingulata and Mas1 from M. oryzae [73]. Three of these genes were among the five most highly expressed M. acridum genes on locust cuticle. Their expression levels were ∼2-fold lower on cockroach cuticle, similar to a previously characterized cuticle binding adhesin, Mad1 [20]. This is also consistent with a previous study which showed that M. acridum up-regulated (∼3-fold) a single Mas1-like gene (MAC_03649) in the extracts of locust cuticular lipids but this gene was down-regulated in extracts from beetles (∼4-fold) or cicadas (∼2-fold) [72].
Formation of appressoria would be expected to involve significant modifications of the germ tube cell wall. Between 6 to 10% of the genes highly expressed by M. anisopliae and M. acridum on host cuticles encoded cell wall proteins. However, cell wall remodeling may be a greater feature of post penetration development because a microarray study showed that ∼20% of insect hemolymph-induced genes were involved in cell wall formation [74]. Evidently, different subsets of genes are required before and after penetration of the cuticle. Suppression-subtractive hybridization identified 200 genes expressed by M. acridum in the hemolymph of locusts [75], and only eight of these genes involved in translation were among the 100 genes that were most highly expressed by pre-penetration germlings.
About 60% of the transcripts expressed by M. anisopliae in liquid cultures containing insect cuticles encoded secreted products, including many proteases [76], as compared to ∼20% of the transcripts in pre-penetration germlings (Table S19), indicating that growth in culture does not mimic the environment experienced on the insect surfaces. Despite the lineage-specific expansion of protease gene families in Metarhizium spp., only a few proteases were abundantly expressed by either species on insect epicuticles. Two trypsins were highly expressed by M. anisopliae on cuticle surfaces, but similar to most subtilisins, they were not expressed by M. acridum. Early expression of proteases triggered by nitrogen starvation may allow M. anisopliae to sample the cuticle, resulting in further induction of proteases that could digest the proteinaceous procuticular layer [76]. Consistent with this hypothesis, both Metarhizium species expressed several genes involved in recognition of nitrogen starvation signals, including MAA_03429 and MAC_02501, which resemble the STM1-like GPCR responsible for triggering adaptation to nitrogen starvation in fission yeast Schizosaccharomyces pombe [77] (Table S20).
The profile of dehydrogenases produced on insect cuticles was used to highlight metabolic pathways that participate in pre-penetration growth. The expression profile of dehydrogenases produced on locust and cockroach cuticles was highly correlated (r = 0.96) in M. anisopliae, but much less so in M. acridum (r = 0.69). The most abundant dehydrogenase transcripts expressed by M. anisopliae on both cuticles included enzymes involved in glycolysis, the citric acid cycle and the oxidative branch of the pentose phosphate pathway. Genes involved in metabolizing intracellular lipids, proteins and amino acids were also highly expressed, showing that lipids are an important nutrient reserve, and that there is a high turnover of proteins during the formation of appressoria as previously suggested for M. oryzae [78].
Similar to previous observations [21], M. acridum germlings only produce appressoria on locust cuticle, and these visually resemble the appressoria produced by M. anisopliae on both insect cuticles (Figure 7). Consistent with early host recognition events being key to establishing specificity, M. acridum but not M. anisopliae transcribed different Pth11-like GPCR genes on locust and cockroach cuticles (Figure S3A). The up-regulation of G-protein alpha subunit, phosphatidyl inositol-specific phospholipase C, protein kinase C, Ca/calmodulin-dependent kinase and extracellular signal-regulated protein kinases indicate that the mitogen-activated protein kinase pathway was strongly activated by M. anisopliae during infection of both insects and by M. acridum infecting locust cuticle. Unexpectedly, M. acridum expressed adenylate cyclase and protein kinase A at higher levels on cockroach cuticle even though appressoria formation was not induced (Figure 7, Table S20). Most of the up-regulated signal-tranduction genes were similar to known PHI genes that regulate infection processes in other fungi (Table S6). Overall, our results suggest that M. anisopliae and M. acridum are able to differentiate diverse host-related stimuli on locust and cockroach cuticles using distinct or shared signaling pathways involving PTH11-like GPCRs, calcium-dependant pathways and MAP kinases that are probably under subtle and sophisticated cross-pathway controls.
Recent improvements in next generation sequencing technology and bioinformatics now allows the de novo assembly of high quality eukaryotic genomes [79], [80]. We used such an approach to provide the first draft sequences of insect fungal pathogens M. anisopliae and M. acridum. Metarhizium species are the best studied insect pathogenic fungi and thus serve as an excellent starting point for gaining a broad perspective of issues in insect pathology. Sequencing two related species that evolved very different life styles provides a powerful method to derive lists of candidate genes controlling pathogenicity, host specificity and alternative saprophytic life styles. By using the experimentally verified pathogen-host interaction (PHI) gene reference database [33], we found that >16% of the genes encoded by each genome have significant similarities with genes involved in pathogenicity in other fungi, oomycetes or bacteria. Our study also highlighted secreted proteins which are markedly more numerous in Metarhizium spp. than in plant pathogens and non-pathogens, pointing to a greater complexity and subtlety in the interactions between insect pathogens and their environments. High resolution RNA-Seq transcriptomic analyses found that the two Metarhizium spp. have highly complicated finely-tuned molecular mechanisms for regulating cell differentiations in response to different insect hosts. These were the first large scale transcriptome studies done with insect pathogenic fungi grown under simulated insect parasitism rather than in liquid cultures.
Whole genome analyses indicated that Metarhizium spp. are closer to plant endophytes and plant pathogens than they are to animal pathogens like A. fumigatus and C. albicans. The finding suggested that Metarhizum may have evolved from fungi adapted to grow on plants even though they now infect insects. This inference is supported by the consistent existence of genes for plant degrading enzymes within Metarhizium genomes (Table S2). In contrast, fungal pathogens of humans that are seldom recovered from soil, such as Coccidioides, exhibit few of these enzymes or none [81]. Even necrophytes such as Trichoderma reesei lack many families of plant cell wall degrading enzymes [82], and the existence of such families in Metarhizium spp. implies that these species are able to utilize living plant tissues. Potentially, these enzymes could also facilitate colonization of root surfaces but this must remain speculative because the genetic basis for rhizosphere competence is largely obscure in fungi [5], [83]. Our identification of the full repertoire of Metarhizium genes will help to identify genes responsible for life on the plant root.
M. acridum contains fewer transposase genes than M. anisopliae which might be due to differences in repeat-induced point mutation (RIP). Both M. anisopliae and M. acridum have orthologs (MAA_03836 and MAC_00922) of the N. crassa RIP defective gene (E≤10−80), the only gene known to be required for RIP [84]. The retention of this gene suggests that M. anisopliae might have undergone RIP at some stage in its evolution, even though its genome currently shows no bias towards C:G to T:A mutations. Creating new genes through duplication is almost impossible when RIP is very efficient [31], so the apparent loss of RIP in M. anisopliae may have been a compromise for the massive expansion of some gene families, though at the cost to M. anisopliae of increased transposition. M. acridum has a strong RIP bias, but RIP is only functional when meiosis is frequent. Cordyceps taii has been described as the sexual type (teleomorph) of Metarhizium taii [7], [85] but the sexual stages of M. acridum (and M. anisopliae) are unknown. However, both Metarhizium species have a complement of apparently functional genes whose orthologs in N. crassa and A. nidulans are known to be required for meiosis and sexual development (Table S21). These include putative α-mating type genes and genes with similarity to a high mobility group (HMG) mating type gene, suggesting that they may have the potential to be either self (homothallic) or non-self (heterothallic) fertile under favorable conditions [26]. More studies are required to understand the importance of the RIP mechanisms in the evolution of Metarhizium genomes and to determine the frequency of meiosis. Discovering whether M. anisopliae and M. acridum undergo sexual reproduction also has important implications for understanding the evolution of new strains of these pathogens.
Alternatively to an undiscovered sexual stage, the conservation of sex genes in an asexual species could be due to a recent loss of sexuality, pleiotropy or parasexual recombination following heterokaryon formation [86]. The well known parasexual cycle that occurs in some fungi including Metarhizium provides another mechanism for hybridization [87]. As with sexual hybridization there are numerous barriers between vegetative fusions of different fungal species with the major one being vegetative incompatability, which results from heterokaryon incompatability proteins that block exchange of DNA [88]. M. acridum has fewer (25 genes) heterokaryon incompatibility proteins than M. anisopliae (35 genes), which suggests that M. acridum may be less reproductively isolated than M. anisopliae. However, it is likely that M. acridum with its more specialized lifestyle and narrow environmental range encounters fewer genetically distinct individuals than the more opportunistic M. anisopliae (Table S2).
The evolutionary transition of Metarhizium spp. to insect pathogenicity must have involved adaptations to insect-based nutrition, as indicated by the large number of proteases, lipases and chitinases that can digest insect cuticles and the host body (Table S6). Except for the lipid outer epicuticle, most of the barriers and nutritional resources in insects are proteinaceous, and Metarhizium has a full set of proteases including many different subtilisins, trypsins, chymotrypsins, metalloproteases, aspartyl proteases, cysteine proteases and exo-acting peptidases. The chymotrypsins are M. anisopliae specific, and may have been acquired by a horizontal gene transfer event [37]. Otherwise, the ∼2–10-fold expanded repertoire of various families of secreted proteases has been the result of preservation by natural selection of duplicated genes. These may have facilitated the adaptation to heterogenous environments. Thus, the abundance of aspartyl proteases and carboxypeptidases (active at low pH) and subtilisins and trypsins (active at high pH), reflects the ability of M. anisopliae to grow in media with a wide range of pH values [89]. The ability to produce large quantities of secreted proteases will obviously assist in the rapid degradation of insect host barriers, but the diversity of different proteases might also have been selected because insects frequently exploit anti-fungal protease inhibitors [38].
With the exception of the trypsins, most of the proteases with orthologs in the PHI-base (Table S6) are reported to have a major function in degrading host barriers. Fungal trypsins are regarded as markers of pathogenicity as they are almost exclusively found in pathogens of plants, animals or fungi [90]. M. anisopliae has more trypsins than any other sequenced fungus, including M. acridum, which indicates a recent evolution of this gene family by gene duplication in M. anisopliae (Figure 1C). We also infer that differential gene loss has occurred due to the existence of six trypsin pseudogenes in M. anisopliae (Table S16). At least two active trypsins are expressed during insect infection [91], but the role of these trypsins in disease has not been demonstrated. The only sequences similar (E<1×10−10) to Metarhizium trypsins in the PHI database are from the oomycete plant pathogen Phytophthora sojae (PHI acc.: 652 and 653). Plants produce diverse glucanases to degrade pathogen cell walls, and the P. sojae trypsins quench this by degrading the glucanases [92]. It is feasible that a similar strategy could occur in insect-fungus interactions since the β-glucan recognition proteins, β-1,3-glucanases and β-1,4-glucanases involved in insect immune responses are similar to the anti-fungal glucanases produced by plants [93].
To date ∼20 Metarhizium genes that contribute to infectious capacity have been described [4]. These have provided important new insights into the novel mechanisms by which pathogens evade host immunity by masking cell wall components with a collagen [18], differentially attach to insects or plants with different adhesins [20] and regulate intracellular lipid stores with a perilipin [21]. Some of these genes, like the collagen MCL1 (MAA_01665), seem to be specifically associated with pathogenicity in M. anisopliae, showing that analysis of orphan (species-specific) genes will be crucial for a full understanding of pathogenicity. Other genes and gene families are generally associated with pathogenicity and can be predicted with the help of the PHI database. The >370 families of genes categorized as containing PHI genes in Metarhizium therefore represent good leads for dissecting the molecular genetics of pathogenicity. Many families like the crotonases involved in fatty acid metabolism [94], the PacC transcription factor that mediates the environmental pH signal, and the suppressers of defense responses such as catalases and superoxide dismutases have been well documented as virulence factors in diverse pathogens of plants and animals [36]. It would be surprising if they were not involved in Metarhizium infection processes.
Other sequences identified from comparisons with the PHI database may be less generic in their impact on pathogenesis. As well as secreted proteins, the interaction between a pathogen and its host is to a large extent orchestrated by the proteins that are localized to the cell wall or cell membrane, and these categories are well represented in the PHI database. Plant and animal pathogens frequently have a subset of extracellular membrane proteins containing an eight-cysteine domain referred to as CFEM. In plant pathogens, CFEM-containing proteins function as cell-surface receptors or signal transducers, or as adhesion molecules in host-pathogen interactions [64]. Deletion of CFEM-containing proteins produces a cascade of pleiotropic effects in C. albicans, most effecting cell-surface-related properties including the ability to form biofilms [95]. The genomic sequences reveal that Metarhizium species also have CFEM-containing proteins (MAA_03310, MAA_04981 and MAA_07591 in M. anisopliae; MAC_09015 and MAC_09359 in M. acridum), and functional analysis is underway to investigate the role they play in M. anisopliae development and pathogenesis. There are many other putative PHI protein families that need to be verified as virulence or pathogenicity determinants in Metarhizium. For example, CheY-like domain proteins are response regulators in some bacterial two-component signaling systems [96], but their roles in fungi remain to be determined. Metarhizium spp. have an average number of histidine kinases compared to other filamentous fungi, and yet M. anisopliae, M. acridum, F. graminearum and M. oryzae have 4, 3, 2 and 0 CheY-like proteins, respectively (Table S6), indicating that M. anisopliae is comparatively well supplied with putative effector proteins that promote responses to stimuli. Much more unexpectedly, M. anisopliae has 6 (M. acridum has 1) homologs of heat-labile enterotoxins that play important roles in bacterial pathogenesis [97]. The HMG proteins involved in fungal sexuality are also required for fungal pathogenicity [98]. Both Metarhizium species have four HMG proteins that are predicted to be PHI genes (Table S6).
M. anisopliae produces large numbers of proteins and secondary metabolites that might be dedicated to host interaction and countering insect defenses [6]. The identity and molecular functions of most secondary metabolite encoding genes remain to be determined in Metarhizium, and it will be intriguing to investigate which of their products are required for pathogenicity and or host specificity. However, with respect to the number of PKS and NRPS genes, M. anisopliae appears to possess a greater potential for the production of secondary metabolites than M. acridum and other sequenced Ascomycetes. M. anisopliae kills hosts quickly via toxins and grows saprophytically in the cadaver. In contrast, M. acridum causes a systemic infection of host tissues before the host dies which suggests limited production of toxins, or none [99]. The presence of NRPS MAA_00969 in M. anisopliae is remarkable as almost all similar genes encoding host selective toxins were found in the Dothideomycetes [100]. It is unlikely that MAA_00969 and HTST1 (encoding the HC-toxin) evolved independently, and one possibility is that MAA_00969 was acquired by an interspecific horizontal gene transfer event. There is no evidence to date that M. anisopliae has a relationship with any plant species that would require a specific toxin, and there are no reports of host-specific toxins in fungal pathogenesis of animals or insects.
Specialization in host range in various Metarhizium lineages is associated with a reduction in the range of molecules the fungi can utilize for nutrition or are able to detoxify [101]. Consistent with this is the deficit of dehydrogenases (DHGs) in M. acridum relative to M. anisopliae or saprophytic fungi (Table 2). M. anisopliae also has more cytochrome P450s (CYPs), which are used by fungi to detoxify diverse substrates [102]. Thus, the additional CYPs and DHGs encoded by M. anisopliae may enable it to detoxify the toxic repertoires of multiple insect hosts, as compared to M. acridum that infects only locusts. CYPS and DHGs also contribute to production of different secondary metabolites by oxidation (CYPs) and dehydroxylation (DHGs) of the backbone structures produced by the PKSs and NRPSs [103]. M. anisopliae's PKS and NRPS clusters contain 18 CYPs and 21 DHGs, while M. acridum's PKS and NRPS clusters contain 3 CYPs and 12 DHGs. The insecticidal destruxin A–F subclasses produced by M. anisopliae have the same backbone structure, but more than 30 different analogues [104]. These analogs presumably derive principally from the action of CYPs or DHGs, but the molecular mechanisms have not been determined.
Comparative global transcriptional studies of M. anisopliae and M. acridum provided a broad-based analysis of gene expression during early colonization processes, particularly in terms of the genes involved in host recognition, metabolic pathways and pathogen differentiation (Figure 7, Figure S6). About 20% of the genes most highly expressed by both Metarhizium species are putative PHI genes (Table S19). In spite of the abundance of protease genes in the Metarhizium genomes only a few proteases, mostly the trypins were expressed in the early stages of infection. As mentioned above, the trypsins could possibly serve as suppressors of host defenses. Studies in a range of plant pathogens suggest that early infection is characterized by the catabolism of internal lipid stores and that polymerized substrates are used after the readily available substrates are exhausted [65], [66]. The transcriptome of M. anisopliae shows that it also uses internal lipid stores early in infection, which is consistent with a previous study [21]. Proteases and chitinases are secreted later at very high levels to digest the protein-chitin procuticles [23]. The occurrence of a stress condition during the early phase of the interaction with the insect host was indicated by the massive up-regulation of heat shock proteins (HSPs). MAA_04726 and MAC_01954 are similar (E<1×10−160) to an HSP90 from C. albicans that is a crucial virulence factor governing cell drug resistance and morphogenetic transition [105]. The other highly expressed Metarhizium HSPs (e.g., HSP30s and HSP70s) are considered to be part of a general defense response and did not resemble sequences in the PHI database.
In spite of differences in infection procedures, we were able to identify some concordance between up-regulated Metarhizium genes and metabolic networks up-regulated by M. oryzae [78] and the mycoparasite Trichoderma harzianum [106]. In particular, during early host colonization, they all up-regulated pathways associated with translation, post-translational modification, and amino acid and lipid metabolism. Metarhizium spp. also resemble M. oryzae and T. harzianum in that pathogenicity is associated with nitrogen deprivation and related stresses, indicating that at least some of the physiological conditions on insects, plants and fungal hosts might be similar. For example, the S. pombe STM1 gene links environmental nitrogen with cell differentiation [77]. The up-regulation of similar STM1-like receptors by the three pathogenic fungi could be a common mechanism for linking low levels of nitrogen on the host surfaces with differentiation of infection structures.
In spite of their different host ranges, developmental processes within M. anisopliae and M. acridum are very similar, e.g. formation of appressoria and blastospores. However, comparatively analyzing their host-invading transcriptomes suggested that recognition might be determined in part by regulatory controls that exclusively limit expression of genes for pathogenicity-related developmental processes to individual hosts. Functional characterization should elucidate whether the expansion in M. anisopliae of several families of signal receptors and response elements is indicative of functional redundancy and/or reflective of a need for fine-tuned sensing of the host environments. The differentially regulated Pth11 GPCR genes are clear early candidates for further functional analysis to confirm their role as regulators of pathogenicity, and to investigate how their function varies between strains with different host ranges. Such studies could define the core set of host-specific transcripts and identify targets for effectively altering host range.
In conclusion, we have identified significant differences in gene contents and transcriptional regulations between M. acridum and M. anisopliae, that have led to the latter having a wider biochemical repertoire available for infecting multiple hosts. The genomic sequences will facilitate identifying candidate genes for manipulation to increase the benefits of applying Metarhizium not just as an insecticide but also potentially as a biofertilizer. The range of exploitable fungal virulence genes is enormous as besides the putative PHI genes, other virulence factors such as the systems for evading host immunity are of particular interests.
M. anisopliae strain ARSEF 23 has been studied in the laboratory for more than 40 years [107]. It is a generalist insect pathogen that successfully infects locusts, caterpillars, flies, crickets and beetles, amongst others, and is classified as a Group A strain (good germination in many media and production of appressoria against a hard hydrophobic surface in yeast extract medium) [101]. M. acridum CQMa 102 can only infect acridids and is being mass produced for large-scale locust control in China [16]. It is classified as a Group D strain (little or no germination in yeast extract or glucose media). A recent taxonomic revision assigns M. anisopliae ARSEF 23 to a new species, viz., M. robertsii [108].
The genomes of M. anisopliae and M. acridum were sequenced with the next generation sequencing technology Illumina. DNA libraries with 200 bp, 2 kb and 6 kb inserts were constructed and sequenced with the Illumina Genome Analyzer sequencing technique at the Beijing Genomics Institute at Shenzhen with protocols described previously for the giant panda genome [80]. The genome sequences were assembled using SOAPdenovo [109]. For syntenic relationship analysis, the scaffolds of both genomes were oriented by MEGABLAST for dot plotting and a pair-wise comparison with an Argo Genome Browser [110].
Annotations of the genomic sequences of M. anisopliae and M. acridum were performed with Augustus [111], specifically trained with >6000 unique sequenced Metarhizium ESTs, and the annotated information of F. graminearum was incorporated as a reference. An ab initio predictor, GeneMark [112] was additionally used for ORF prediction with both Metarhizium genomes. Thorough manual checks were conducted on parallel comparisons of the results from two prediction methods. All questionable ORFs were individually subjected to Blast searches against the NCBI curated refseq_protein database and the individual prediction with the best hit was selected for each ORF. Pseudogene identification was conducted with the pipeline of PseudoPipe [113]. Transfer RNAs (tRNAs) were predicted with tRNAscan-SE [114] and ARAGORN [115]. Secreted proteins were identified by SignaIP 3.0 analysis (http://www.cbs.dtu.dk/services/SignalP/).
Ortholog conservation in fungi was characterized with Inparanoid 7.0 [116]. In total, 946 orthologous proteins were acquired and aligned with Clustal X 2.0 [117]. A maximum likelihood phylogenomic tree was created using the concatenated amino acid sequences with the program TREE-PUZZLE using the Dayhoff model [118]. The divergence time between species was estimated with the Langley-Fitch method with r8s [119] by calibrating against the reassessed origin of the Ascomycota at 500–650 million years ago [120].
Whole genome protein families were classified by InterproScan analysis (http://www.ebi.ac.uk/interpro/) in combination with the Treefam methodology that defines a protein family as a group of genes descended from a common ancestor [121]. To identify potential pathogenicity and virulence genes, whole genome blast searches were conducted against protein sequences in the pathogen-host interaction database (version 3.2, http://www.phi-base.org/) (E<1×10−5). The families of proteases were additionally classified by Blastp against the MEROPS peptidase database (http://merops.sanger.ac.uk/). Transporters were classified based on the Transport Classification Database (http://www.tcdb.org/tcdb/). The cytochrome P450s were named according to Dr. Nelson's P450 database (http://drnelson.utmem.edu/CytochromeP450.html). G-protein coupled receptors, protein kinases, transcription factors and GH families were classified by Blastp against GPCRDB (http://www.gpcr.org/7tm/), KinBase (http://kinase.com/), Fungal Transcription Factor Database (http://ftfd.snu.ac.kr/) and CAZy database (http://www.cazy.org/), respectively. All Metarhizium genes with significant hits (E value ≤ 10−5) to GPCRDB sequences and that contained 7 transmemebrane helices (analyzed with http://www.cbs.dtu.dk/services/TMHMM/) were included as putative GPCRs. To analyze fungal secondary metabolite pathways, the genome annotation data from both species were coordinated and analyzed with the program SMURF (http://www.jcvi.org/smurf/index.php). The evolution of protein family size variation (expansion or contraction) was analyzed using CAFE [32].
Genome repetitive elements were analyzed by Blast against the RepeatMasker library (open 3.2.8) (http://www.repeatmasker.org/) and with the Tandem Repeat Finder [122]. RIP index was determined with the software RIPCAL by reference against the non-repetitive control families [30]. The transposons/retrotransposons encoding transposases/retrotransposases were classified by Blastp analysis against the Repbase (http://girinst.org/).
The hind wings from locusts (Locusta migratoria) and cockroaches (Periplaneta americana) were collected and surface sterilized in 37% H2O2 (5 min), washed in sterile water twice and dipped in conidial suspensions (2×107 spores per ml) of M. anisopliae ARSEF 23 or M. acridum CQMa 102 for 20 seconds. The inoculated wings were placed on 1% water agar and incubated at 25°C for 24 hrs. The wings with fungal cells were homogenized in liquid nitrogen and the total RNA was extracted with a Qiagen RNeasy mini kit plus on-column treatment with DNase I. Messenger RNA was purified from 6 µg total RNA. After reverse transcription into double strand cDNA for tag preparation according to the massively parallel signature sequencing protocol [123], it was sequenced with an Illumina technique. We omitted tags from further analysis if only one copy was detected or it could be mapped to different transcripts [124]. Other tags were mapped to the genome or annotated genes by allowing if they possessed no more than one nucleotide mismatch. The abundance of each tag was converted to transcripts per million for quantitative comparison between samples. We used the test of false discovery rate (FDR≤0.001) to estimate the level of differential gene expression by each species under different induction conditions [125].
Metarhizium anisoliae and M. acridum were tested for their ability to kill adult locusts Locusta migratoria and cockroaches Periplaneta americana. For these experiments, conidia from each species were applied topically by immersion of cold-immobilized insects into aqueous suspensions of 5×108 spores per ml. Each treatment was replicated three times with 15 insects per replicate and the experiments were repeated twice. Mortality was recorded every 12 hours and the lethal time values for 50% mortality (LT50) were estimated [18].
The Whole Genome Shotgun projects have been deposited at DDBJ/EMBL/GenBank under the accession ANDI00000000 for Metarhizium acridum and ANDJ00000000 for Metarhizium anisopliae, respectively. The RNA-seq data have been deposited at NCBI GEO repository with accession numbers GSM612996, GSM612997, GSM612998 and GSM612999 for the samples of M. anisopliae infection of locust, M. anisopliae infection of cockroach, M. acridum infection of locust and M. acridum infection of cockroach, respectively.
|
10.1371/journal.pntd.0005429 | The relationship between entomological indicators of Aedes aegypti abundance and dengue virus infection | Routine entomological monitoring data are used to quantify the abundance of Ae. aegypti. The public health utility of these indicators is based on the assumption that greater mosquito abundance increases the risk of human DENV transmission, and therefore reducing exposure to the vector decreases incidence of infection. Entomological survey data from two longitudinal cohort studies in Iquitos, Peru, linked with 8,153 paired serological samples taken approximately six months apart were analyzed. Indicators of Ae. aegypti density were calculated from cross-sectional and longitudinal entomological data collected over a 12-month period for larval, pupal and adult Ae. aegypti. Log binomial models were used to estimate risk ratios (RR) to measure the association between Ae. aegypti abundance and the six-month risk of DENV seroconversion. RRs estimated using cross-sectional entomological data were compared to RRs estimated using longitudinal data. Higher cross-sectional Ae. aegypti densities were not associated with an increased risk of DENV seroconversion. Use of longitudinal entomological data resulted in RRs ranging from 1.01 (95% CI: 1.01, 1.02) to 1.30 (95% CI: 1.17, 1.46) for adult stage density estimates and RRs ranging from 1.21 (95% CI: 1.07, 1.37) to 1.75 (95% CI: 1.23, 2.5) for categorical immature indices. Ae. aegypti densities calculated from longitudinal entomological data were associated with DENV seroconversion, whereas those measured cross-sectionally were not. Ae. aegypti indicators calculated from cross-sectional surveillance, as is common practice, have limited public health utility in detecting areas or populations at high risk of DENV infection.
| In this study, we compared measures of entomological risk collected through routine household entomological monitoring by estimating an association with human DENV infection. Longitudinal entomological and human serology data from Iquitos, Peru, were used to test associations between Ae. aegypti indices and the 6-month risk of DENV seroconversion. Our analysis found no association between cross-sectional measures of Ae. aegypti abundance and the risk of DENV seroconversion. Longitudinal measures of Ae. aegypti were better proxies for DENV risk, primarily among adult stage mosquito indicators. DENV transmission is complex and time-varying; the relationship between vector density and risk is not static nor adequately characterized through periodic entomological surveillance. While entomological monitoring will continue to serve a role in the evaluation of vector control interventions (e.g., comparing pre- and post-intervention abundance), our analysis challenges the validity of most Ae. aegypti indicators as adequate proxies for true DENV exposure risk.
| Dengue virus (DENV), which is transmitted by the bite of female Aedes aegypti mosquitoes, causes more human morbidity and mortality than any other arthropod-borne virus [1]. Since the 1950s, dengue has spread via the globalization of trade and travel, rapid urbanization and expansion of vector habitats [2]. The four serotypes (DENV1, DENV2, DENV3 and DENV4) occur throughout the tropics and infect approximately 390 million persons per year [1]. Until effective DENV vaccines become broadly commercially available, vector control will remain the primary prevention strategy in most dengue endemic settings [3] and even as vaccines become accessible vector control will be needed to supplement vaccine efforts [4], as well as control of other arboviruses also vectored by Ae. aegypti.
The World Health Organization recommends monitoring vector abundance for the targeting and evaluation of vector control interventions [5]. Ae. aegypti monitoring was first employed in yellow fever control programs in the first half of the 20th century [6, 7]. Since then, over two dozen indicators have been proposed to quantify abundance of Ae. aegypti. Entomological monitoring data are typically collected from households sampled from neighborhoods or blocks on a routine or ad hoc basis [8]. The frequency of entomological data collection also varies by setting, and WHO guidelines recommend implementation occur at a frequency from “weeks to months” [5]. As such, entomological monitoring surveys impose cross-sectional measurement of the highly dynamic Ae. aegypti population. Monitoring indicators vary by mosquito life stage (adults, larvae and/or pupae), availability of larval development sites (infestation indices), and process of collection (fixed trap or human-based surveys such as adult aspirator collections, household inspection for larvae) [9]. The public health utility of these indicators is based on the assumption that greater mosquito abundance increases the risk of DENV transmission, and therefore reducing exposure to the vector decreases incidence of infection. Further, by identifying “hot spots” of Ae. aegypti infestation, targeted vector control would be an efficient use of limited intervention resources [10].
To date, studies have not shown a consistent association between various indices and infection or disease outcomes [7]. This may be due to several limitations inherent to the large-scale measurement of Ae. aegypti densities. First, there is no established threshold of Ae. aegypti density associated with an increased risk of human DENV infection [11]. Second, entomological survey techniques may not capture the fine spatial and temporal variability in an urban setting due to the constraints dictated by household-based monitoring, and the fact that indices are calculated from cross-sectional prevalence measures, not derived from continuous monitoring. Third, while adequate sampling of immature and adult populations requires consideration of vector dynamics [12] and spatial relationships [13], the data do not capture the daily productivity of individual larval development sites or the activity of individual mosquitoes over their lifespan [8, 14]. Finally, previous attempts to quantify the association between vector abundance and dengue outcomes may also have been biased due to measurement error caused by operational constraints and collection procedures [9], and methodological issues, such as restricting the analysis outcomes to infected people who sought treatment or small sample size [7].
Ae. aegypti densities may also fail to describe risk of DENV infection due to the complexity of transmission. The probability of transmission is dependent on human movement to introduce DENV into mosquito populations and the presence of susceptible individuals that mosquitoes infect to perpetuate new rounds of transmission [15]. Because Ae. aegypti are daytime-biting mosquitoes that are highly adapted to the human urban environment [16], their frequent biting contact with susceptible human hosts is mediated by social and economic [17] factors that govern human movement through times and spaces where they encounter mosquitoes [18]. While high concentrations of Ae. aegypti within or around a household present an opportunity for clustered DENV transmission, it ignores transmission occurring in other places [19, 20].
To help predict risk and direct public health interventions, there is substantial interest in an improved understanding of the utility of Ae. aegypti monitoring measures in terms of an association with DENV infection, according to mosquito life stage and spatial scale of measurement. We aimed to systematically examine measures of entomological risk collected through routine household surveillance with human DENV infection using longitudinal entomological and human serology data to test associations between Ae. aegypti indices and the 6-month risk of DENV seroconversion.
Written informed consent (and assent for children 8–17 years of age) was obtained for all individuals providing serological data. Written informed consent was provided by parents or guardians for children under 18 years of age. Written consent (1999–2003) or oral consent (2008–2010) was obtained from an adult head of household for entomological surveys as approved by the institutional review boards. Oral consent for entomological surveys was documented upon obtaining access to the household and heads of households were provided information sheets describing the data collection procedures. Data collection procedures were approved by the University of California, Davis (Protocols 2002–10788 and 2007–15244), Instituto Nacional de Salud, and Naval Medical Research Center Institutional Review Boards (Protocols NMRCD.2001.0008 and NMRCD2007.0007). This ancillary analysis was approved by the Institutional Review Board at the University of North Carolina at Chapel Hill (Study # 14–3151).
The analytical cohort was constructed using entomological and serological data collected between 1999–2003 and 2008–2010 from two longitudinal cohort studies implemented in Iquitos, Peru. Iquitos, the largest city in the Peruvian Amazon, has a population of approximately 350,000 [21]. DENV1 is presumed to have been introduced in 1990–1991 [22], followed by DENV2 in 1995 [23], DENV3 in 2001 [24], and DENV4 in 2008 [25]. Seasonal epidemic levels of DENV transmission occurred throughout this period [21, 25]. From 1999–2003, study activities were implemented in four city districts: Maynas, Punchana, Belen and San Juan. During the period 2008–2010, data were collected from two neighborhoods: Maynas and Tupac Amaru (located within the Maynas and Punchana districts).
Procedures for entomological data collection were previously described [13, 21]. Briefly, once households were enrolled, two-person study teams collected entomological data following a circuit to survey neighboring households on the same and/or neighboring block (with block defined as a group of households that shared a common perimeter defined by city streets) within an approximately two-week period. The entire study area required approximately four months to complete data collection, upon which entomological surveys resumed following the same schedule.
Adult Ae. aegypti were collected using CDC backpack aspirators (1999–2009) [26] or Prokopack aspirators (2009–2010) [27] in both the exterior and interior of the participating household by passing the vacuum tube over common Ae. aegypti resting sites, outside walls, vegetation, and the entrance of potential larval habitats. Pupae and larvae were collected via enumeration of all wet containers or other larval development sites that contained water upon inspection. During surveys, all observed pupae and a sample of larvae were collected in small plastic Whirlpack bags; larval density was estimated as one of four levels (0, 1–10, 11–100, >100). All adult, larval and pupal samples were transported to and examined at the study laboratory, counted and identified to species and sex. Pupal data were recorded as observed counts. The total number of adult male and female Ae. aegypti mosquitoes collected in the interior and exterior of the dwelling were recorded. Household demographic data were collected for variables including enumeration of household residents by age and sex, household water source, sanitation facility, presence of electricity, type of building material, roof structure, and any reported use of insecticide or larvacide.
The indicators were classified by scale (household or block) and life stage (adult, pupal and/or larval). Household-level indicators were calculated using the observed survey data. To construct block-level indicators, all household survey data were first aggregated by block using a unique block identification number and circuit schedule. Indicators were then calculated using the aggregated block-level Ae. aegypti data. Block-level measures were then linked back to individual households by matching on block identifier and date of collection. The household-level indicators and their definitions are summarized in Table 1 and block-level indicators are summarized in Table 2.
Since the distribution of Ae. aegypti counts across all life stages is narrow in most settings, we dichotomized the continuous indicators to determine if categorical characterization of mosquito abundance would reflect a better fit to the data. To test categorical (dichotomous) versions of continuous indicators, a preliminary analysis was conducted to identify cut-off values by estimating the sensitivity and specificity of the mosquito density in terms of DENV infection at different levels (data not presented). There is no consensus in the literature as to what categorical values of mosquito density measures correlate with DENV infection, therefore we used the following systematic approach to select categories and then test for an association. This approach was used to allow the distribution of the continuous indicator value to inform categorization without data mining for an association. To choose a categorical cutpoint, the sensitivity of the mosquito density indicator to identify a DENV seroconversion was calculated for increments of five (e.g., a Breteau Index of 0, 5, 10, etc.), with the exception of the Potential Container Index, which was estimated for increments of two. The cutpoint was chosen as >0 if the sensitivity was less than 50% at that value. If cutpoints greater than zero had a sensitivity >50%, then the cutpoint (not zero) with the highest sensitivity was chosen for evaluation. Once a categorical variable was defined, it was then tested for an association with risk of DENV seroconversion. Categorical classification of continuous indicators tested is listed in Tables 1 and 2. Data on eggs or exact larval counts were not collected in the parent study; therefore, indices relying on this information could not be tested.
Cross-sectional entomological indicators were calculated using vector data from a single entomological survey observation. Longitudinal household-level indicators were calculated as an average of entomological data observed within the 12 months preceding the start of the seroconversion interval (up to three survey visits collected approximately every four months). If a paired sample interval began before any entomological data collection, the cross-sectional measure of mosquito density was used. For block measures, indicators were calculated by averaging block-level densities calculated from surveys conducted within 12 months from the start of the seroconversion interval.
In the parent study, members of households selected for entomological monitoring were asked to provide blood samples every six months [21, 33]. Samples were collected at the participant’s home, stored in ice and transported to the study laboratory within four hours of collection. Sera were tested at two (1999–2003) and four (2008–2010) serum dilutions plaque reduction neutralization test (PRNT) [34] at the United States Naval Medical Research Unit No. 6 laboratory in Lima, Peru. To identify seroconversion to DENV, a serum sample was considered positive for DENV if a dilution neutralized 70% of the test virus (PRNT70) [21, 33].
The primary outcome of interest in this analysis was seroconversion to any circulating DENV serotypes as determined by PRNT70. The longitudinal serological samples used in this analysis were previously reviewed to determine seroconversion [33]. In brief, to minimize misclassification of serological data, the full serological profile of subjects was reviewed as follows: if the increase in titer that reduced DENV plaques between a negative sample and a subsequent sample was at least 20% and all subsequent samples were positive, the subject was determined to have seroconverted. However, if subsequent PRNT results were not consistent with respect to seroconversion (e.g., negative-positive-negative), the subject was classified as not having seroconverted. For this study, serological results for all paired samples were classified as a binary outcome (any seroconversion versus no seroconversion).
Fig 1 illustrates the construction of the analysis cohort. Serological data were reviewed to identify paired sample observations taken approximately six months apart that could be linked to household entomological data. To account for operational constraints around serology collection, the at-risk interval was defined as 140 to 220 days. Each paired sample interval for which a subject was susceptible to any of the circulating DENV serotypes (DENV1 and DENV2: all study years; DENV3: 2001–2010; DENV4, 2008–2010) was included in the risk set.
For household-level indicators, entomological data were matched by the date nearest the end of (but within) each paired serological sample interval. For block-level indicators, datasets were constructed by restricting to serological observations from blocks in which at least five households were surveyed, using the month and year of block data collection to anchor in time block-level densities to serology. Finally, longitudinal densities were calculated by averaging entomological data collected in the 12 months preceding the start of the seroconversion interval. Fig 2 illustrates how cross-sectional and longitudinal measures of vector abundance were calculated and linked to the 6-month seroconversion paired sample interval.
The association between each Ae. aegypti indicator and the 6-month risk of DENV seroconversion was estimated using a log binomial generalized estimating equation (GEE) [35], separately for each household-level and block-level indicator, and for both the cross-sectional and longitudinal scenarios. The log link with a binomial distribution allowed for estimation of risk ratio point estimates by exponentiating the beta coefficient for the indicator variable and calculation of 95% confidence intervals (CI) [36]. For models using household-level densities, the GEE accounted for clustering due to repeated individual measures and dependence due to household membership using an exchangeable correlation structure; models for block-level densities accounted for repeated observations from individuals and block level membership. A priori, we chose dengue transmission season, participant age and sex as confounding variables for use in all adjusted analyses of household-level indicators and season, participant age (dichotomized at 18 years), and any reported use of larvacide by the head of household for adjustment of all block-level indicators. Dengue transmission season was defined as by the start of the seroconversion interval (May-August (reference group), September-December, January-April). All analyses were conducted in SAS/STAT software, version 9.4 of the SAS system for Windows (SAS Institute, Cary, NC).
Sensitivity analyses were conducted to account for possible bias resulting from construction of the dataset. The objective of the sensitivity analysis was to determine if the adjusted risk ratios were sensitive to decisions made to construct the analytical dataset. To implement these analyses, the same method as described in the main analysis was employed. First, different inclusion criteria for serological observations was used to test more restrictive or relaxed scenarios, as well as stratification by study years. Second, sensitivity analyses included alternate strategies for linking serology to entomology. Third, vector densities were calculated from entomological data 6 months prior to serology compared to 12 months prior to serology. Finally, the analysis was stratified by aspirator type used during data collection.
Written informed consent (and assent for children 8–17 years of age) was obtained for all individuals providing serological data. Written informed consent was provided by parents or guardians for children under 18 years of age. Written consent (1999–2003) or oral consent (2008–2010) was obtained from an adult head of household for entomological surveys as approved by the institutional review boards. Oral consent for entomological surveys was documented upon obtaining access to the household and heads of households were provided information sheets describing the data collection procedures. Data collection procedures were approved by the University of California, Davis (Protocols 2002–10788 and 2007–15244), Instituto Nacional de Salud, and Naval Medical Research Center Institutional Review Boards (Protocols NMRCD.2001.0008 and NMRCD2007.0007). This ancillary analysis was approved by the Institutional Review Board at the University of North Carolina at Chapel Hill (Study # 14–3151).
In total, 13,526 households contributed 90,330 entomological observations and 25,755 paired serological samples (from 6,775 individuals). A total of 20,176 serological observations could be linked to entomological data. Fig 1 details the analytical sample size. For the cross-sectional household-level analysis, 4,089 household entomological observations (from 1,377 unique households) were linked to 8,153 paired blood samples (from 3,824 individuals). For the longitudinal household-level analysis, 15,548 entomological observations from those 1,377 households were used to calculate average densities and matched to the 8,153 serological observations. The same set of serological and entomological observations were included in the block-level analyses, with the exception of 579 serological observations for which a block density could not be obtained (<5 households per block-visit were surveyed). A total of 7,574 serological paired samples (from 3,644 individuals) were used in the cross-sectional and longitudinal block-level analyses.
The mean age of individuals at first paired sample was 20.9 years (standard deviation: 16.3, range 2–96) and 57.7% of subjects were female. At first study visit, most households contributing any serological data reported access to electricity (99.7%), piped sanitation (77.1%), and potable water (75.2%), had open or partially open household roof structure (93.0%), and were constructed from either mud and/or wood (49.2%) or concrete and/or brick (50.8%). Only 28.2% of households reported using Abate (larvacide) at enrollment. There were a total of 1,191 seroconversions (14.6%) in the analysis of household level indicators and 1,129 seroconversions (14.9%) in the analysis of block-level indicators. Tables 3 and 4 present the distribution of entomological indicators.
The adjusted RR point estimates and 95% CI are presented in Table 5. The household-level point estimates ranged from 0.75 (95% CI: 0.48, 1.34) to 1.05 (95% CI: 0.91, 1.21), suggesting no difference in the 6-month risk of DENV seroconversion based on Ae. aegypti density. At the block level, six indicators showed significant protective effects, which could be the result of higher background immunity, correlation with factors related to lower DENV risk, or chance. Compared to the adjusted RR estimates, crude risk ratio point estimates were similar for the household-level indicators and were slightly larger for block-level indicators (S1 Table).
Using the average of densities measured in the 12 months prior to the paired sample, the RR point estimate shifted above the null for categorical measures of adult density, adult female mosquitoes, and presence of adult mosquitoes indoors (any adults as well as only females), ranging from 1.25 (95% CI: 1.12, 1.39) to 1.30 (95% CI: 1.17, 1.46). This suggests that the observation of an adult female mosquito during a household survey performed during the 12 month period prior to collection of paired sera is associated with an approximately 25% increased risk in acquisition of DENV infection compared to the risk among individuals residing in households where no adult female was observed at any survey during the 12 months preceding the paired sera. In addition, four immature stage indicators suggested an elevated risk of DENV infection: any pupae observed; the Single Larval Method (categorical); Container Index (categorical) and the Stegomyia Index (categorical).
Analysis of block-level indicators that incorporated repeated measures demonstrated a similar trend in which all measures calculated based on adult mosquito data shifted in comparison to the cross-sectional analysis: the Adult Premise Index (RR: 1.01; 95% CI: 1.01, 1.02 when continuous and RR: 1.24; 95% CI: 1.01, 1.48 as categorical) and the Adult Density Index (RR: 1.24; 95% CI: 1.02, 1.50 as continuous and RR: 1.72; 95% CI: 1.22, 2.43 as categorical). The Pupa Index (categorical) and the Infested Receptacle Index (categorical) were the only immature stage block-level indicators to demonstrate any association with DENV infection. Figs 3 and 4 compare the risk ratios calculated for cross-sectional to longitudinal densities for both household and block-level indicators.
A number of sensitivity analyses were performed to determine if the construction of the analytical cohort introduced bias. Ae. aegypti densities calculated from 6 months prior to the start of a seroconversion interval followed a similar pattern as results presented in Table 5 (S2 Table). Sensitivity analyses in which the inclusion of seroconversion events was relaxed and restricted did not alter interpretation of the main findings (S1–S3 Figs, S3–S5 Tables). Future comparison of relaxed serological inclusion criteria were compared with the 6 and 12 month longitudinal entomological measures of Ae. aegypti (S4–S6 Figs). When analyzed separately, use of different aspirators over the course of data collection did not result in substantially different results for adult stage measures (S6 Table).
The principal finding of this analysis is that a higher household level Ae. aegypti density calculated from cross-sectional entomological data was not associated with an increase in the risk of DENV infection. Compared to cross-sectional measures, the average Ae. aegypti density in the past 12 months resulted in more plausible effect estimates, especially for adult indices which monitor the life stage relevant to DENV transmission. Entomological evidence suggests that Ae. aegypti populations in Iquitos are highly variable in time and space [37] and the indices obtained from trimestral surveys are unlikely to capture all of the fine-scale temporal variation that occurred. The lack of an association between cross-sectional measures of larval, pupal and adult stage indicators of Ae. aegypti abundance suggests that measures of entomological risk calculated from periodic household surveys are not sufficient proxies for the 6-month risk of DENV infection. The lack of association may be the result of cross-sectional entomological survey procedures in which adult data are measured over a short period of time, resulting in lower or higher densities being attributed to the entire risk period [38]. By comparing measures of Ae. aegypti density calculated from cross-sectional data to an average density, we are able to explore the potential for non-differential measurement error of mosquito abundance to bias the association between Ae. aegypti monitoring indicators and DENV infection towards the null. This may be due to the large proportion of households with low levels of infestation being misclassified as having no Ae. aegypti present when relying on a single measurement.
Immature stage indicators were not associated with risk of DENV infection, with the exception of a few categorical indicators calculated from longitudinal data. This could be due to high larval mortality, the short lifespan of larvae and pupae, and brief time interval of data collection, resulting in immature population measures that do not always correlate in space and time with the biologically relevant adult measures [39]. For block-level indicators, aggregating household data could skew calculation of the indicator if the distribution of larval and pupal counts was concentrated in only a few households. Block-level indicators such as the Breteau Index and the House Index, which classify containers or households as “infested” if any larvae or pupae are observed, may not capture the contribution of container productivity. The pupae per person and pupae per hectare measures are sensitive to bias from inaccuracies in population or area data, as well as sampling error as the pupal life stage is ephemeral [40]. We also compared the number of infested containers at enrollment to the number observed in subsequent study visits to confirm that long-term participation in the study was not associated with improved household container management (data not presented). Prior studies also show that the spatial distribution of these measures in Iquitos varied considerably over time [37, 39].
The major strengths of this analysis include the use of DENV infection (not disease) as an outcome, examination of longitudinal data, and its generalizability to similar settings in which routine, periodic entomological surveillance is conducted. While dengue disease is relevant from a public health perspective and easier to quantify, DENV infection, measured as seroconversion, is more important in terms of understanding patterns of transmission from mosquitoes to humans. Most prior studies of entomological indicators and dengue outcomes [7] used symptomatic disease as the outcome. Symptomatic cases represent the small fraction of all infections that were severe enough to seek medical evaluation, thus introducing selection bias. This analysis also benefitted from longitudinal serological data, which enabled exclusion of paired sample observations once an individual was determined to no longer be at risk of infection by circulating serotypes.
Most prior studies used cross-sectional entomological data to test for an association with dengue outcomes. Longitudinal entomological monitoring allowed the use of multiple (1 to 3 per household) mosquito measures per household. This may overcome some of the measurement error of entomological assessment and account for the temporal variability associated with entomological data collection, in which a household with lower levels of abundance could be misclassified as “unexposed” to Ae. aegypti. Our results comparing the RRs estimated from cross-sectional to longitudinal entomological measures of Ae. aegypti abundance suggest the possibility that in any single entomological survey, a household with low-levels of Ae. aegypti infestation may be misclassified as having no infestation, at least for adult stage measures of abundance, which would bias the RR downwards.
The objective of this analysis was to evaluate the utility of periodic entomological monitoring as a proxy for DENV infection risk as it is typically implemented in the control setting. Under this monitoring framework, our findings are likely generalizable to similar dengue-endemic settings as the timing of serological and entomological collection employed are representative of the routine periodic monitoring used in dengue control programs. Our data have an added advantage given they were generated as part of a research study, and were subjected to rigorous monitoring of field collection procedures.
Our results should be interpreted in light of several limitations. First, a large proportion (9,739 of 20,176) of serological data failed to meet the 6-month inclusion criteria, which could have resulted in bias due to their exclusion. Results from sensitivity analyses to include paired samples taken more than six-months apart did not qualitatively change our findings (S1–S3 Figs). Second, the entomological and serological monitoring data relevant for DENV transmission did not perfectly coincide temporally, possibly leading to bias due to time of measurement. In sensitivity analyses, results were not sensitive to different approaches to link entomology and serology (S1–S6 Figs). Our dataset contains more entomological monitoring data than would be available in most control settings. Even with a detailed longitudinal dataset of domestic vector density, we did not observe informative associations with DENV risk. Therefore, our study reveals the inherent limitations of using Aedes survey methods. The design of any entomological surveillance system should consider operational feasibility given the investment needed to generate sufficient data to describe temporal and spatial variability in vector density. While ovitraps were not used in the parent studies, we expect ovitrap-based sampling of Ae. aegypti to still be subject to the same limitations that apply to monitoring pupal, larval and adult mosquitoes. Alternatives that need to be evaluated include fixed trap methods, such as ovitraps or adult trapping methods that can monitor larger areas continuously overtime, better capturing temporal differences. Monitoring traps still presents operational challenges. Methods using fixed traps usually sample fewer houses than are possible with household-based survey methods [41]. Novel technologies that capture house-to-house temporal differences are yet to be developed.
Second, even though the use of averages is not the most sophisticated method to incorporate temporal lags, it is implementable in basic statistical software and may be of utility to dengue program managers. Furthermore, in the control setting, it is highly unlikely that DENV infection outcomes would be well-resolved in time with entomological surveillance data as DENV infection cannot be monitored in real-time. Our study reports the relative risk of DENV seroconversion in a six-month period, an outcome similar in length to periods of increased dengue activity that occur seasonally in many endemic settings. While the temporal resolution between entomological and serological data collection in our study was not well resolved, our results provide quantitative evidence to challenge the use of periodic Ae. aegypti surveillance to generate suitable surrogates for DENV risk.
Third, while PRNT70 is the most specific serological test for dengue infection, results from this assay may be biased due to cross-reactions from antibodies directed against multiple serotypes present in a single sample or with closely related viruses. The algorithm used to classify seroconversions was conservative, possibly underestimating the number of seroconversions, but this bias is likely non-differential with respect to mosquito density. We also acknowledge the possibility that block or neighborhood-level susceptibility to DENV may affect the performance of Ae. aegypti indicators, but it was not possible to address it in the analysis without full enumeration and serological testing of the entire study population, not just sampled individuals. Nevertheless, in the endemic setting, an assumption that herd immunity exists would only further undermine the utility of entomological monitoring endpoints to serve as correlates of infection. To ensure individuals in our analysis were susceptible to DENV infection, we reviewed longitudinal serological profiles to exclude those who were likely no longer at risk of the circulating serotype. We also tested a household-level variable estimating the proportion of susceptible individuals but this had no impact on the overall results (data not presented). While our results may be generalizable to other areas with endemic transmission, this analysis should be repeated in a setting with a largely susceptible population to determine if household-based entomological surveillance is associated with DENV risk in such locations. Nevertheless, the majority of infections in our dataset were DENV3 and DENV4, which were novel at the time of introduction in the community, so herd immunity may not have played a significant role for a large subset of our data.
In this analysis, continuous indicators were tested as linear terms to maintain consistency with their definitions in the literature. It is possible that log-transformation or inclusion of polynomial terms could improve model fit, but such manipulation would reduce interpretability. For continuous indicators, the RRs measure the relative risk for a one-unit change in the indicator value; these measures are likely not informative for targeting interventions. From a public health perspective, categorical indicators are more useful to trigger vector control activities. In Iquitos, levels of infestation were heavily dispersed and binary classification (any v. none) was most informative.
Finally, it is possible that vector control efforts could have reduced the vector population, making it difficult to detect an association between Ae. aegypti density and DENV risk. Over the period included in this analysis, there were large scale vector control interventions from October 2002-Jan 2003 and others in late 2003 [42]. Nevertheless, the majority of our data included periods where there was not extensive vector control. We did control for household larvacide use as a covariate in the analysis of block-level indicators to account for the possibility that some households implemented some form of vector control.
Our results provide the first quantitative evidence of the limited utility of Ae. aegypti monitoring indicators as proxy measures of DENV infection. DENV transmission is complex and time-varying; the relationship between vector density and risk is not static nor adequately characterized through periodic entomological surveillance. None of the RRs presented in this analysis represent a causal relationship between household or block-level mosquito density and true exposure to DENV. It is logistically impossible to monitor human-vector contact to establish where and when mosquito-human interaction and infection occurs. Therefore, Ae. aegypti indicators serve as surrogates of true exposure, which will always remain unmeasured. Although adult measures that incorporated longitudinal data demonstrated an association with DENV seroconversion in our study, it is possible that some unmeasured variable associated with social network patterns, housing quality and day-time human movement further modifies dengue risk.
Entomological monitoring indicators were not designed to account for the complexity of human-vector interaction, particularly given the role human movement may extend the boundaries of contact; it is likely that a substantial proportion of transmission occurs outside the home [18]. Technological advances in mosquito monitoring may eventually enable dengue control programs to quantify fluctuations in mosquito populations with greater precision across time and space. Nevertheless, DENV infection is difficulty to identify in real-time, especially given that most infections are inapparent in endemic settings. Without information on where and when individuals are infected, even detailed data of domestic vector density will require aggregation or categorization in order to attribute mosquito density to an interval-defined outcome (such as the six-month seroconversion window, as in this analysis) as DENV infection is measured at a coarse temporal interval.
Globally, the incidence of dengue has continued to intensify and expand despite significant investments in vector control [1, 43]. While vector control remains the only prevention strategy available to reduce DENV transmission in most settings, the persistence of DENV suggests transmission dynamics require a more complex understanding of human-vector interaction. Entomological monitoring will continue to serve a role in the evaluation of vector control interventions as it will be necessary to compare entomological measures of risk pre- and post-intervention as indicators of impact. Our analysis challenges the validity of most Ae. aegypti indicators as adequate proxies for true DENV exposure risk, and challenges the assumption that domestic vector data correlate with DENV transmission. In dengue-endemic settings such as Iquitos, single cross-sectional measures of adult mosquito density and the immature stage indicators commonly used by dengue control programs, such as the Breteau Index and Container Index, will likely fail to predict risk of DENV infection. Measuring adult mosquito density over multiple occasions may be the best option, but is difficult to implement. Our findings should be considered in the development and revision of enhanced DENV surveillance guidelines. Dengue control programs weighing the operational feasibility and cost of entomological monitoring against the limited utility of these indicators may wish to seek alternative monitoring frameworks that incorporate human dengue-related outcomes, such as passive case detection, and where possible, sero-surveys and active case detection.
|
10.1371/journal.pgen.1005625 | Curly Encodes Dual Oxidase, Which Acts with Heme Peroxidase Curly Su to Shape the Adult Drosophila Wing | Curly, described almost a century ago, is one of the most frequently used markers in Drosophila genetics. Despite this the molecular identity of Curly has remained obscure. Here we show that Curly mutations arise in the gene dual oxidase (duox), which encodes a reactive oxygen species (ROS) generating NADPH oxidase. Using Curly mutations and RNA interference (RNAi), we demonstrate that Duox autonomously stabilizes the wing on the last day of pupal development. Through genetic suppression studies, we identify a novel heme peroxidase, Curly Su (Cysu) that acts with Duox to form the wing. Ultrastructural analysis suggests that Duox and Cysu are required in the wing to bond and adhere the dorsal and ventral cuticle surfaces during its maturation. In Drosophila, Duox is best known for its role in the killing of pathogens by generating bactericidal ROS. Our work adds to a growing number of studies suggesting that Duox’s primary function is more structural, helping to form extracellular and cuticle structures in conjunction with peroxidases.
| Fruit fly geneticists rely on a handful of dominant mutations that modify adult morphology in a way that is easy to spot, like changing the shape of the fly’s wings, eyes or bristles. One of the first such mutants identified in the early days of fly genetics and to this day likely the most widely used mutation, is Curly, which causes an upward curvature in the adult wings. Despite its importance as a marker, the genetic cause of Curly has remained unknown. Here, we reveal that Curly mutations occur in the gene duox, which encodes a ROS-generating enzyme. ROS once thought to be merely harmful by-products of metabolism, can also have beneficial purposes. Here we provide evidence that Duox generates ROS to help form and stabilize the wings of fruit flies. Furthermore, we identify a second enzyme, Cysu, which uses the ROS generated by Duox to crosslink proteins in the wing, thereby stabilizing and shaping its structure. Duox occurs in numerous organisms, including humans and fulfills a number of other functions, in particular in immunity and pathogen defense. With this new knowledge, Curly mutations will provide an excellent tool to study and understand the roles Duox plays in a variety of biological contexts.
| Over 90 years ago, Lenore Ward first described a dominant mutation, Curly, that causes the wings of Drosophila melanogaster to bend upwards [1]. Since then, Curly has become a ubiquitous second chromosomal marker used by Drosophila geneticists on a daily basis to follow and track mutations. Despite its widespread use, how Curly mutations dominantly alter wing curvature has remained obscure. Waddington first proposed that Curly causes an unequal contraction of the dorsal and ventral wing surfaces during the drying period shortly after flies emerge from their pupal cases [2, 3]. Others have subsequently demonstrated that comparable alterations in wing curvature can be caused by differential growth of the dorsal and ventral epithelia [4]. Irrespective of the mechanism, that similar wing phenotypes have been described for D. pseudoobscura and D. montium mutants [5] suggests the underlying cause of curly wing formation is evolutionarily conserved among Drosophilids [2]. The major factor limiting our understanding of Curly’s function in wing morphogenesis, however, is the fact that its molecular identity has remained unknown.
In this manuscript we uncover the long unknown molecular nature of Curly. We show that mutations in the gene duox cause the Curly wing phenotype. Duox is a member of a highly conserved group of transmembrane proteins collectively referred to as NADPH oxidases. These enzymes function to transfer electrons across biological membranes to generate ROS by transferring electrons from NADPH to oxygen through flavin adenine dinucleotide (FAD) and heme cofactors [6]. Several biological functions have been described for Duox. Perhaps the best studied of these in Drosophila is its role in host defense where it is thought to generate ROS to kill pathogens [7]. However, Duox also plays an important role in providing ROS, specifically hydrogen peroxide, for heme peroxidases to catalyze the formation of covalent bonds between biomolecules. In mammals, Duox generates hydrogen peroxide for thyroid peroxidase to catalyze the iodination and crosslinking of tyrosine residues in the formation of thyroid hormones [8, 9]. Duox is also expressed in tissues other than the thyroid, such as the gastrointestinal tract, where its function is less clear [6]. In insects, worms and sea urchins, Duox participates in the formation of extracellular structures through the crosslinking of tyrosine residues [10–12]. Indeed, instead of its function in generating bactericidal ROS, the tyrosine crosslinking activity of Duox may be the primary ancestral function, as it appears to be conserved across phyla.
Here we show that specific mutations in the NADPH binding-domain encoding region of duox cause a Curly wing phenotype. Using Curly, we demonstrate that duox is required during the last day of pupal development to stabilize the wing. Furthermore, through suppression experiments, we identify a novel heme peroxidase, Curly Su (Cysu), that works with duox to adhere the dorsal surface of the wing to the ventral one. Uncovering the molecular identity of Curly not only provides an entry point for the functional understanding of this prominent wing mutant phenotype, but also will allow for the discovery of novel duox interacting genes and regulators through unbiased genetic screens. Only through these approaches can we hope to understand the precise molecular function of Duox in the myriad biological processes in which it is involved.
In the course of genetically following a loss-of-function mutation in the gene duox, duoxKG07745, using the standard Drosophila genetic tool Curly of Oster (CyO) balancer, we noticed that we were unable to recover progeny containing both duoxKG07745 and CyO. Since a) duox and Curly mutations fail to complement one another and b) Curly roughly maps to 23A4-23B2 [13], the chromosomal region containing duox, we wondered whether Curly might be an allele of duox.
Balancer chromosomes contain numerous inversions and chromosomal aberrations. To exclude the possibility that our inability to recover progeny was due to some other lesion in the 23A4-23B2 region of CyO, we crossed duox KG07745 to various Curly mutations not associated with CyO [1, 13, 14]. Consistent with our earlier results, all the Curly alleles tested failed to complement duoxKG07745, suggesting that Curly mutations indeed reside within the duox locus (Fig 1A). To provide conclusive evidence, however, that duox and Curly are one and the same, we expressed duox ubiquitously in a Curly mutant background. Ubiquitous expression of duox restored viability, allowing recovery of homozygous Curly mutants (Fig 1A). These results strongly suggest that the Curly phenotype is due to mutations in the duox gene.
To precisely determine how Curly mutations alter duox’s nucleotide sequence, we sequenced duox in two Curly mutants: CyK, which was previously generated using ethyl methanesulfonate [14]; and Cy1, the original spontaneous mutation identified by Ward [1]. Remarkably, we found that the same nucleotide was mutated in both CyK and Cy1 (Fig 1B). This single nucleotide mutation resulted in the conversion of a conserved glycine (number 1505) in the NADPH binding domain of Duox to a serine in CyK and to a cysteine in Cy1 (Fig 1C, red). Among the residues within the NADPH binding pocket of Duox, glycine 1505 is extraordinarily well conserved. It is present in all known NADPH oxidases from yeast to humans (Fig 1D), suggesting that it is functionally important. Therefore a conversion of a conserved glycine to a polar amino acid specifically in the NADPH binding pocket of Duox causes Curly. Taken together, our results demonstrate, after more than 90 years since its discovery, that the Curly phenotype is due to mutations within the duox gene.
Mutations in the NADPH binding encoding region of duox cause a Curly phenotype, but how exactly do they influence Duox’s function? From complementation experiments, it is clear that Curly mutations reduce Duox’s normal function as they are not viable in combination with either a loss-of-function duox mutation or a deficiency uncovering the duox locus (Fig 1A). However, Curly mutants also act dominantly, causing the wings of Curly flies to bend upwards (hence the name) in contrast to the straight wings of wild-type or duox heterozygous flies. One possible explanation is that Curly mutations act in opposition to Duox’s normal activity as dominant negatives or antimorphs. This, however, is unlikely because ubiquitous overexpression of wild-type Duox in a Curly mutant background failed to restore normal wing shape (S1 Fig). Another possibility is that Curly mutations increase the normal function of Duox either by increasing expression or constitutively activating Duox. However, this too seems implausible because removing wild-type Duox in Curly mutants worsened (causing lethality) rather than ameliorated the viability phenotype (Fig 1A). Instead, the most likely explanation is that Curly mutations are neomorphic, causing a dominant gain-of-function in Duox that is different from its normal function.
To further investigate the potential neomorphic nature of Curly mutations, we tested whether Curly mutations require NADPH, the substrate of Duox, to generate a wing phenotype. In the first instance, we attempted to alter the endogenous levels of NADPH by reducing the amount of niacinamide, a precursor of NADPH, in the diet of Curly flies. Interestingly, we found that a reduction of dietary niacinamide caused a dose-dependent decrease in the expressivity of the wing phenotype in Curly mutants (Fig 2A). To test this genetically, we next knocked down CG6145, which encodes a NAD+ kinase that phosphorylates NAD+ to generate NADP+ [15]. Specific knockdown of CG6145 in the wing using apterous-gal4 (apGal4) strongly suppressed the Curly wing phenotype (Fig 2B). Together these results suggest that Curly mutants require NADP+ and/or NADPH in order to cause changes in wing curvature. Therefore, not only do Curly mutations reduce Duox’s normal function, they also endow it with a new function that requires sufficient substrate to dominantly alter wing shape.
Having determined that Curly is most likely a neomophoric allele of duox, we decided to use Curly mutations to explore duox’s function in vivo in the wing. To do this, we generated a mutant form of duox, henceforth referred to as duoxCyK, in which glycine 1505 was mutated to a serine, as in the CyK mutant. In order to express duoxCyK conditionally we fused it to an Upstream Activating Sequence (UAS) element so that we could control its expression in a time-dependent and tissue-specific manner using the Gal4 system [16]. To demonstrate that this transgene was functional and able to recreate the Curly phenotype, we expressed it ubiquitously throughout the fly using tubpGal4 [17]. Indeed, ubiquitous expression of duoxCyK, but not wild-type duox, resulted in upward bent wings resembling those of CyK mutants (Fig 2C and 2D). This not only demonstrates that the duoxCyK transgene is functional, but also provides further evidence that Curly is an allele of duox.
Duox could be required autonomously within the wing for its formation, or instead act non-autonomously in other tissues, as is the case for curled, a mutation that causes similar changes in wing morphology [2]. To determine where duox is required, we first expressed duoxCyK or duox RNAi in the wing using apGal4. Expression of duoxCyK, but not wild-type duox, caused the upward wing curvature, whereas knockdown of duox in the wing caused a slight downward curvature or cupping of the wing (Fig 3A). This demonstrates that duox is required autonomously in the wing for its formation.
Changes in wing curvature can be caused by differential growth between the dorsal and ventral wing surfaces, or could be caused by changes in cuticle structure [3, 4]. If duoxCyK were differentially influencing growth, we would expect it to be required early in pupal development, but not later, when proliferation and growth of the ventral and dorsal wing surfaces is largely complete [18]. To determine when in development duoxCyK is acting, we conditionally expressed it at various developmental stages using a heat shock-inducible driver. Expression of duoxCyK, but not duox, on the last day of pupal development, but not before then, resulted in upturned wings (Fig 3B). This suggests that Duox does not influence wing growth and instead perhaps plays a role in the formation of the wing cuticle. Taken together, our results demonstrate that duox acts autonomously in the wing to stabilize the cuticle during the last day of pupal development concurrent with the formation of the wing cuticle.
Hydrogen peroxide generated by NADPH oxidases is often used by peroxidases, most frequently heme peroxidases, to crosslink proteins and kill pathogens [6]. Heme peroxidases are essential for NADPH oxidase-dependent crosslinking reactions, but largely dispensable to their other functions [6]. To test whether Duox acts with a heme peroxidase in the wing to crosslink proteins and stabilize it, we individually knocked down all known D. melanogaster heme peroxidases (Fig 4A) [19] using RNAi in the wings of duoxCyK flies. If duoxCyK acts alone in the wing and does not participate in crosslinking reactions, then knockdown of heme peroxidases should not affect wing curvature. If, however, duoxCyK requires a heme peroxidase to function, then knockdown should suppress the Curly phenotype. Consistent with this, knockdown of one peroxidase, CG5873, fully suppressed the Curly phenotype in a manner resembling duox knockdowns, suggesting that Duox functions together with CG5873 to crosslink molecules to form the wing (Fig 4A). As CG5873 is a suppressor of Curly we have named it Curly Su, or cysu for short.
If cysu acts with duox to form the wing then one would expect it to: a) be expressed in the wing on the last day of development–the period in which Duox functions to form the wing; and b) have a similar phenotype to Duox when silenced in the wing. To test this first prediction, we generated a strain expressing endogenously mCherry-tagged Cysu. Consistent with Cysu functioning in the wing, we observed expression of mCherry-tagged Cysu in the wing on the last day of development using confocal microscopy (Fig 4B). To test our second prediction, we knocked down Cysu in the wing using RNAi. Cysu knockdown caused a slight downturning and cupping of the wing resembling Duox knockdown wings, further suggesting that Cysu and Duox function together to shape the wing (Fig 4C). Interestingly, defects in scutellum and notum formation were also observed in Duox and Cysu knockdowns, suggesting that they might play a broader role in cuticle formation (Fig 4D). Consistent with this, Cysu was expressed in the thorax of wild-type flies, but not Cysu knockdowns (Fig 4E). In summary, Duox and the heme peroxidase Cysu act together to stabilize the wing during development.
The Drosophila adult wing is made up of two cuticle panels that are synthesized and secreted in a step-wise fashion by the dorsal and ventral ectodermal epithelial cells toward the end of pupal development [20, 21]. Upon eclosion, the dorsal and ventral cuticular surfaces of each wing expand and within an hour or so become bonded and adherent to one another [18]. To determine how altering Duox influences wing cuticle formation, we imaged wild-type and duox mutant wings using transmission electron microscopy (Fig 5). In wild-type wings, the ventral and dorsal cuticles are closely apposed and tightly bonded (Fig 5A). By contrast, in Duox knockdown wings the ventral and dorsal cuticles rarely directly contact one another, and instead are separated by a gap filled with disordered, electron poor material (Fig 5B). Interestingly, Curly mutant wings on the other hand appeared much more similar to wild-type wings with occasional abnormal bunching of the dorsal cuticle (Fig 5C). Whether this pinching and bunching reduces the surface area of the dorsal wing causing the wing to curve upward remains unknown. However, ultrastructure experiments clearly demonstrate a role for Duox in the adhesion and bonding of the two cuticle wing surfaces during wing formation.
Here we have shown that the Curly mutation arises in the NADPH-binding pocket encoding region of duox. Using Curly mutations and duox RNAi, we show that Duox is required within the wing to maintain its shape beginning on the last day of pupal development. Results from our genetic studies suggest Duox does this by supplying hydrogen peroxide to the heme peroxidase Cysu to facilitate the bonding of the two wing cuticle surfaces, likely by physically crosslinking them, during wing formation.
In all Curly mutants sequenced, a glycine residue, 1505, in the NADPH-binding pocket of Duox is mutated. This glycine is present in all NADPH oxidases from microbial eukaryotes to humans, and more broadly in oxidoreductase and ferric reductase NAD-binding domains (PFAM PF00175 and PF08030, respectively). Though mutagenesis studies have not been conducted on this residue itself, it sits beside an equally conserved cysteine residue, which has been studied in detail because mutations in it cause chronic granulomatous disease in humans [22]. This cysteine residue does not appear to be important for NADPH oxidase assembly or binding NADPH [22, 23]. Instead it is thought to be required for orienting bound NADPH for efficient electron transfer (via hydride) to FAD, and eventually oxygen [22]. Given glycine 1505’s proximity, it is possible that mutations in it similarly affect the transfer of electrons from NADPH to FAD. Consistent with this is the observation that Curly mutants are neither homozygous viable nor viable over a deficiency, suggesting that mutation of glycine 1505 causes a reduction in Duox’s normal function.
Although Curly mutations reduce Duox’s normal function, they also endow it with a new function. Precisely what this new function is remains obscure, however it likely requires a source of electrons because altering the NADPH/NADP+ by removing niacinamide from the food or knocking down NAD+ kinase suppressed the wing phenotype. It is known that the expressivity of the Curly wing phenotype can be suppressed by larval crowding and/or starving larvae [24]. Given this, it is possible that reduced uptake of niacinamide is a cause of the decreased expressivity of the Curly wing phenotype in starved larvae. Riboflavin shortage during the larval stage has also been suggested to be a cause of this suppression [2, 25]. Since riboflavin is a precursor of FAD, a co-factor also necessary for Duox’s function, it too may suppress the wing phenotype by reducing endogenous FAD and in turn reducing Duox’s activity. Regardless, Curly mutations are likely neomorphic and their sensitivity to environmental factors is likely mediated by changes in substrate availability.
Duox is required autonomously for wing stabilization. Results from this study and another strongly support this assertion [10]. Expression of duoxCyK or knockdown of duox on the last day prior to eclosion, but not earlier, caused defects in wing morphogenesis. This suggests that Duox and Curly do not influence growth or proliferation of the wing epithelia because these processes are complete by this time [18]. Instead, ultrastructural analysis suggests that Duox plays an important role in forming the cuticle of the wing. In duox knockdowns, frequent gaps between the two wing cuticle surfaces were observed, in contrast to the wild-type wings. Defects in adhesion of the two cuticle surfaces were also apparent in Curly mutants. Unlike wings from duox knockdowns, however, the cuticle surfaces in Curly wings were most often tightly apposed with occasional bunching of the dorsal surface. It is possible that in the Curly mutants this aberrant pinching of the dorsal surface decreases its area relative to the ventral surface causing the wing to bend, as first intimated by Waddington 75 years ago [3]. However, we do not know whether this is the cause of the curling or just a consequence of it.
Duox is known to be involved in the formation of extracellular matrices and cuticles [10–12]. Typically, it does this by supplying hydrogen peroxide to heme peroxidases, which use the hydrogen peroxide to perform crosslinking reactions. Consistent with Duox playing a role in crosslinking the cuticle we found that the heme peroxidase Cysu was essential for Duox function in the wing. Duox is unusual among NADPH oxidases in that it contains its own peroxidase homology domain, which in Caenorhabditis elegans and D. melanogaster has been proposed to fulfill the function of heme peroxidases, thereby obviating their need [7, 26, 27]. However, given that the peroxidase homology domain of Drosophila Duox lacks many amino acid residues, including the proximal and distal histidines, essential for efficient peroxidase function it is unclear how well it functions in this capacity [27]. Indeed, our results suggest that in D. melanogaster, Duox requires the heme peroxidase Cysu not only for stabilizing the wing cuticle, but also in the formation of the notum and scutellum. These findings point to a more general role for Duox and Cysu in cuticle formation.
In Drosophila, Duox has been intensely studied in the context of host defense and gut immunity. In the gut, Duox is thought to generate ROS to kill pathogens; flies that have reduced Duox activity have increased susceptibility to infection [7, 26]. Upon infection ROS generated by Duox kill pathogens, and possibly signal intestinal epithelial cells to proliferate and renew [7, 28]. Our results, as well as others, demonstrate that Duox is also critical in the formation of cuticle structures and extracellular matrices [10–12]. It is possible that Duox performs a similar function in the Drosophila intestine, perhaps by forming extracellular barriers or structures to protect against infection. Indeed, Duox in conjunction with heme peroxidases has been shown to form such barriers in guts of ticks and mosquitos [29, 30]. It would therefore be interesting to explore whether Duox and possibly Cysu are also involved in forming barriers to protect against infection in the Drosophila intestine.
Duox is an important protein that has a number of diverse functions, which we are only beginning to understand. Curly mutations provide an excellent opportunity to further explore Duox’s functions by identifying unknown interactors and regulators through unbiased genetic suppressor screens. The identification of Cysu through such an approach demonstrates its feasibility and utility. Such approaches will not only tell us about Duox’s function in the wing, but also about its role in immunity and beyond.
Flies were propagated in polystyrene vials (28.5 mm diameter) containing cornmeal molasses yeast medium at 25°C. For most crosses 5 virgin females were mated with 3 to 5 males. Fly strains used are shown in Table 1.
duox and duoxCyK were conditionally expressed using a gal4 driver downstream of the heat-shock (HS) protein 70 promoter (HS-gal4). At various times during development Drosophila were incubated at 37°C for 2 hours to induce expression.
To verify the location of P{SUPor-P}DuoxKG07745 we performed inverse PCR. Consistent with the FlyBase report (FBal0226250) the 3’ flanking sequence was at genomic position 2L:2,826,884. However, unexpectedly, we found the 5’ flanking sequence to be at position 2L:2,755,447. This suggests that P{SUPor-P}DuoxKG07745 contains a deletion that perturbes 17 genes from Bacc to duox.
The duox open reading frame was amplified from cDNA and cloned into the pVALIUM22 vector [31] between XbaI and EcoRI restriction sites using standard methods. The CyK mutation was generated by site-directed mutagenesis using a QuikChange site-directed mutagenesis kit. Constructs were integrated into the attP2 site on the third chromosome using phiC31 integrase by BestGene.
mCherry-tagged Cysu expressing flies were made by BestGene by injecting mimic construct #1315 into Mi{MIC}CG5873MI11428 (BDSC 56608) [32].
Genomic DNA was crudely isolated by homogenizing one to two flies in 0.2 mg/ml Proteinase K (Roche MC00079), 10 mM Tris pH 8.0, 1 mM EDTA and 25 mM NaCl and incubating for 25 min at 55°C. Proteinase K was subsequently inactivated by boiling the samples for 5 min. duox was then PCR amplified from genomic DNA and sequenced by Genewiz.
Flies were raised on a defined medium with various concentration of niacinamide from embryo to adult. A solution was prepared as described in Table 2 and the pH was adjusted to 7.0 with NaOH. 20 mg/ml agar yeast culture grade (Sunrise Science Products 1910) was dissolved in the solution by heating before adding 0.4 mg/ml cholesterol (Sigma C3045) and niacinamide (Sigma N0636).
Fluorescent images were acquired with a 10X/NA objective on a Zeiss LSM 780 confocal microscope. All other images were obtained with a Zeiss SteREO Discovery.V8 microscope.
For transmission electron microscopy, whole flies were immersed in 95% ethanol briefly to get rid of any air bubbles, decapitated and immersed into fixative containing 4% glutaraldehyde in 0.1M PIPES buffer, pH 7.2 at room temperature for 2 hours, and then overnight at 4°C. Flies were next embedded in 1% agar and post-fixed with 2% osmium tetroxide with 1.5% potassium ferricyanide in 0.1M PIPES buffer for 1 hour then en block stained with 1% uranyl acetate in ddH2O at 4°C overnight. Samples were dehydrated with ethanol at room temperature before incubation with propylene oxide and embedment in Spurr resin (Electron Microscopy Sciences, Hatfield, PA). 500nm semi-thin sections were stained with 0.1% toluidine blue to evaluate the area of interest. 60nm ultrathin sections were cut, mounted on formvar coated slotted copper grids and stained with uranyl acetate and lead citrate by standard methods. Stained grids were examined under Philips CM-12 electron microscope (FEI; Eindhoven, The Netherlands) and photographed with a Gatan (4k x2.7k) digital camera (Gatan, Inc., Pleasanton, CA).
|
10.1371/journal.pcbi.1004394 | Identification of Ohnolog Genes Originating from Whole Genome Duplication in Early Vertebrates, Based on Synteny Comparison across Multiple Genomes | Whole genome duplications (WGD) have now been firmly established in all major eukaryotic kingdoms. In particular, all vertebrates descend from two rounds of WGDs, that occurred in their jawless ancestor some 500 MY ago. Paralogs retained from WGD, also coined ‘ohnologs’ after Susumu Ohno, have been shown to be typically associated with development, signaling and gene regulation. Ohnologs, which amount to about 20 to 35% of genes in the human genome, have also been shown to be prone to dominant deleterious mutations and frequently implicated in cancer and genetic diseases. Hence, identifying ohnologs is central to better understand the evolution of vertebrates and their susceptibility to genetic diseases. Early computational analyses to identify vertebrate ohnologs relied on content-based synteny comparisons between the human genome and a single invertebrate outgroup genome or within the human genome itself. These approaches are thus limited by lineage specific rearrangements in individual genomes. We report, in this study, the identification of vertebrate ohnologs based on the quantitative assessment and integration of synteny conservation between six amniote vertebrates and six invertebrate outgroups. Such a synteny comparison across multiple genomes is shown to enhance the statistical power of ohnolog identification in vertebrates compared to earlier approaches, by overcoming lineage specific genome rearrangements. Ohnolog gene families can be browsed and downloaded for three statistical confidence levels or recompiled for specific, user-defined, significance criteria at http://ohnologs.curie.fr/. In the light of the importance of WGD on the genetic makeup of vertebrates, our analysis provides a useful resource for researchers interested in gaining further insights on vertebrate evolution and genetic diseases.
| Duplication of existing genes with subsequent divergence of duplicated copies has long been recognized as the primary source of genomic innovation. Gene duplication is thus at the root of the evolution and complexification of living organisms. However, gene duplicates have been retained differently depending on the genomic scale of their duplication and their implication in genetic diseases. The scale of genomic duplication spans from small scale segmental duplication to whole genome duplication (WGD), which corresponds to a dramatic doubling event of a species genome. In particular, all vertebrates, including human, descend from two rounds of WGDs, which occurred in their jawless ancestor some 500 MY ago. Interestingly, WGD gene duplicates, also called ‘ohnologs’, have be shown to be more frequently implicated in genetic diseases in human. Hence, identifying ohnologs appears central to better understand the evolution of vertebrates and their susceptibility to genetic diseases. In this study, we present a computational approach to predict ohnologs in six vertebrate genomes, including human, based on the comparison of their local gene content (i.e. synteny) with the genomes of six invertebrate outgroups. We show that such synteny comparisons across multiple genomes enhance the statistical power of ohnolog identification compared to earlier approaches.
| Gene duplication and their subsequent divergence is the primary source of new genes in eukaryotes. The importance of evolution by gene duplication is exemplified by a large number of paralogous genes in most eukaryotic genomes. In addition to duplication of single genes or genomic segments, duplications of the entire genome have now been firmly established in all major eukaryotic kingdoms. Multiple lineages including unicellular yeast and paramecium, as well as many plants and animals are known to descend from polyploid ancestors, often through multiple rounds of genome duplications [1]. In vertebrates, whole genome duplications (WGD) were first hypothesized by Susumu Ohno [2] (the 2R-hypothesis), after whom WGD duplicated genes are now referred to as “ohnologs”.
Interestingly, duplicated genes originating from whole genome duplication have been preferentially retained in different functional categories as compared to duplicated genes originating from small scale duplication [3–6]. In particular, many ohnologs have been retained in gene families involved in development, signaling and gene regulation [3, 7–10], and led to the emergence of novel cell types in vertebrates, such as the neural crest, the midbrain/hindbrain organizer and neurogenic placodes [11]. In addition, ohnologs are frequently associated with diseases such as cancer [3, 5, 6, 12–14], and are particularly prone to dominant deleterious mutations [5, 6] as rationalized from a population genetics perspective [5, 15]. These observations suggest that the identification of ohnologs with high statistical confidence has important implications to better understand the developmental complexity of vertebrates as well as their enhanced susceptibility to dominant deleterious mutations and associated diseases.
However, the identification of ohnologs in vertebrate genomes is not straightforward [16]. During the millions of years of evolution following WGD, sister regions created by WGD are redistributed across the paleopolyploid genome by chromosomal rearrangements and degenerate by the loss of the majority of ohnologs (Fig 1). In principle, these degenerated WGD duplicated regions sharing a few ohnolog pairs can be identified in the paleopolyploid genome by comparing its genome-wide synteny either with itself (Fig 1I) or with outgroup genomes diverged before the WGD event (Fig 1J and 1K). Yet, the two rounds of WGD at the onset of vertebrates are among the oldest known genome duplications and the conservation of gene order (or micro-synteny) between extant vertebrate and invertebrate outgroup genomes is limited [17]. This makes WGD detection methods based on micro-synteny conservation [18–23] difficult to apply to WGD from early vertebrates. Other methods, not-based on synteny, such as Ks-based methods [24, 25] and more recent phylogenetic methods [26, 27], cannot be easily applied to the 500 MY-old WGD in vertebrates either, due to the saturation effect of the synonymous mutation rates Ks [28] and the difficulty in distinguishing between the two rounds of WGD in the phylogeny of early vertebrates [17, 29].
As an alternative, a number of studies have proposed to identify ohnologs in the human genome by relaxing strict gene-order criteria and searching, instead, for content-based synteny [30] between the human genome and a single invertebrate outgroup genome [17, 31] or within the human genome itself [3, 4, 32]. Using content-based synteny criteria, however, increases the odds of old duplicates being incorrectly identified as ohnologs, if no quantitative assessment of the statistical confidence of ohnolog pair candidates is performed. In addition, performing synteny comparison with a single outgroup may lead to omission of many ‘true’ ohnolog pairs, whose orthologs have moved to different non-syntenic regions in the extant outgroup genome (Fig 1).
In this study, we have extended these latter approaches to six amniote vertebrates (human, mouse, rat, pig, dog and chicken) by investigating the conservation of content-based gene synteny relative to six invertebrate outgroup genomes (lancelet, two seasquirts, sea urchin, fly and worm, S1 Fig). We also analyzed the synteny conservation from the regions created by 2R-WGD within each of the vertebrates, and then integrated the synteny information from both self and outgroup comparisons. The integration of synteny information across multiple genomes enables to identify ohnologs that are no longer in significant synteny in a particular vertebrate genome, as long as their ortholog status can be unequivocally established with proper ohnologs in other vertebrates. We present below the general principles of our multiple genome comparison approach to identify 2R ohnologs and provide a quantitative assessment of the statistical confidence of each ohnolog pairs by comparison with the expected spurious synteny obtained with shuffled genomes. We show that the synteny comparison across multiple genomes enhances the statistical power of ohnolog identification in vertebrates compared to earlier approaches. The resulting ohnolog pairs and families are accessible at http://ohnologs.curie.fr/ for three statistical confidence levels and can also be recompiled for specific, user-defined, significance criteria.
We implemented content-based synteny comparisons between each amniote vertebrate and multiple invertebrate outgroup genomes. Initial ohnolog candidates were identified, in each vertebrate genome, using a window-based approach to detect putative synteny blocks between each vertebrate and the six outgroup genomes (outgroup comparison, Fig 1J), extending earlier similar approaches [17, 30, 31]. Additional synteny block candidates were also identified by comparing each vertebrate genome to itself (self comparison, Fig 1I) [3, 32] and ohnolog pair candidates were further restricted to paralogous pairs duplicated at the base of vertebrates according to Ensembl compara [33–35] (see S1 Text, Supplementary Materials and Methods). S1 Fig lists the numbers of human ohnolog pair candidates identified by each invertebrate outgroup and human-human synteny comparison, before applying any filtering on the statistical support of candidate synteny blocks. We identified a total of 15,107 such putative ohnolog pair candidates, including 11,428 identified with at least one outgroup and 15,054 identified by self comparison alone.
To narrow down this initial list of ohnolog candidates, we developed a quantitative approach to assess the statistical confidence of each ohnolog pair candidate. This quantitative approach and corresponding ‘q-score’, ranging from 0 to 1, estimates the probability that each ohnolog pair is simply identified by chance. Hence, lower q-scores imply more statistically significant ohnolog pairs (see S1 Text). Finally, we integrated q-scores for outgroup-comparison and self-comparison from all vertebrates, and filtered the ohnolog pairs based on the resulting combined q-scores. A flowchart summarizing our algorithmic approach is depicted in Fig 2. The pipeline of the approach is outlined below with methodological details described in Supplementary Materials and Methods (S1 Text).
The strict, intermediate and relaxed criteria lead to three sets of ohnolog pairs in the human genome with decreasing statistical confidence levels: 2,695 ohnolog pairs with very high confidence, 4,827 with high confidence and 8,178 with medium confidence, respectively (Table 1). These predicted ohnolog pairs are also significantly different from ohnolog pairs reported in earlier studies [3, 4], Table 1. In particular, 617 (23%) of the 2,695 strict ohnologs pairs from our analysis are not identified in [3]. For example, the strict ohnolog pairs between the transcription factors SOX11 and SOX12 or between the microtubule-associated proteins MAP2, MAP4 and MAPT are missing in [3]. Conversely, 3,695 (44%) of the 8,383 ohnolog pairs reported in [3] are excluded by the present analysis. More precisely, we found that 1,853 (50%) of these 3,695 ohnolog pairs ruled out by our analysis have not been duplicated at the base of vertebrates according to Ensembl compara, while 813 (22%) discarded ohnolog pairs are not supported by our quantitative multi-genome synteny comparison and the remaining 1,029 (28%) are excluded by both duplication timing and quantitative multi-genome synteny assessment. For example, the 3-oxoacid CoA-transferase genes OXCT1 and OXCT2, previously reported as ohnologs [3], have in fact been duplicated more recently than the 2R-WGD (i.e. in mammals according to Ensembl compara). By contrast, the signaling genes WNT1 and WNT3, also reported as an ohnolog pair [3] are not supported by our quantitative multi-genome synteny criteria and have also been duplicated earlier than the 2R-WGD (i.e. in bilateria or coelomata according to Ensembl compara).
The distribution of our ohnolog pairs with respect to all six outgroups is depicted on a six way Venn diagram in Fig 3 (percentages) and S8 Fig (numbers). Ohnolog pairs range from 1,416 with sea urchin comparison to a maximum of 5,994 using Drosophila melanogaster as outgroup. There are only 3.8% (293) ohnolog pairs identified by all outgroups, while each outgroup combination shaded in green in Fig 3 contributes to more than 2% of the total number of ohnolog pairs. This illustrates that many ohnologs would not be identified using just a single outgroup genome owing to lineage specific rearrangements in the outgroup genomes, limitations of genome assembly/annotation or homology criteria. In particular, while 90% (6,943) ohnolog pairs in human are identified by at least one chordate outgroup genome, 10% (772) ohnolog pairs are only identified by synteny comparison with non-chordate genomes. For example, the homeobox protein ohnolog pair VAX1/VAX2 and the nuclear receptor co-repressor ohnolog pair LCOR/LCORL are only identified by synteny comparison with D. melanogaster and C. elegans.
The final human ohnolog counts for strict, intermediate and relaxed criteria are respectively, 3,544 ohnologs (Strict Criteria); 5,504 ohnologs (Intermediate Criteria) and 7,831 ohnologs (Relaxed Criteria), Table 1. This is also to be contrasted with the results of previous studies that used either content-based synteny comparison with a single outgroup [17, 31] or only self comparison [3, 4, 32] without statistical significance criteria to filter out spurious synteny block conservation. We found that the available sets of human ohnologs from these early studies also present significant differences from our results. For instance, the set of 7,075 ohnolog genes from [3] shows significant differences from ours (S9 Fig), as 14%, 18% and 23% of our human ohnologs for strict, intermediate and relaxed criteria, respectively, have not been identified in [3]. Conversely, 57%, 33% and 15% of this early ohnolog data set are excluded from our strict, intermediate and relaxed human ohnolog sets, respectively (S9 Fig). As discussed above, this is due to inconsistent duplication times, according to Ensembl Compara, and/or limited statistical supports for each confidence criteria.
We then reconstructed ohnolog families from ohnolog pairs using a depth first search algorithm [36] (S1 Text). The resulting ohnolog families also contain paralogs which are small scale duplicates with respect to each other but form ohnolog pairs with a third gene of the family. Accounting for such small scale duplicates, eventually lead to ohnolog families with an expected maximum of four ohnologs retained from the two rounds of WGD in early vertebrates. However, as most genes lose their duplicates after WGD, most ohnolog families are expected to be of size two or three.
We obtained 1,381, 2,024 and 2,642 ohnolog families using strict, intermediate and relaxed criteria, respectively, for the human genome. Most remarkably, for almost all of these families, the size never exceeds four ohnologs, as expected for two rounds of WGD. As depicted in Table 1, all but 7 ohnolog families (99.5%) have a size smaller or equal to four for the strict criteria. Even with the most relaxed criteria, 96.7% of ohnolog families are consistent with a maximum family size of four ohnologs. Furthermore, a sharp decline in the number of families was observed beyond size four, suggesting a limited number of false positive ohnologs incompatible with two rounds of genome duplications. Interestingly, however, many three- or four-ohnolog families could not be identified independently in individual amniote genomes, but only by integrating synteny information from different amniote genomes, such as the four-ohnolog family ERAS/HRAS/KRAS/NRAS (relaxed criteria).
We also applied the same approach to generate ohnolog families from the ohnolog pairs provided by [3] and [4]. 95.1% of ohnolog families from [3] are consistent with two rounds of WGD and only 85.4% of ohnolog families from [4] have sizes up to four ohnologs. Clearly families exceeding four ohnologs must result either from the erroneous concatenation of distinct ohnolog families or include non-ohnolog genes. For instance, the ohnolog status of TRPV5 and TRPV6 [3] from the large family of six ion channels (TRPV1-6) are not supported by our quantitative assessment of self- and outgroup synteny. Conversely, we could also identified previously overlooked ohnologs, through high confidence assessment of self- and outgroup synteny. For instance, the guanine exchange factor RGL2 was found to be part of a four-ohnolog family with strict criteria, RGL1/RGL2/RGL3/RALGDS, RGL4 (with RGL4 a small scale duplicate of RALGDS).
In addition to the human genome, our synteny comparison approach across multiple genomes also identified ohnologs in five other amniote genomes: four mammals (mouse, rat, pig and dog) and one bird (chicken). Starting from ohnolog pairs in each species, the same approach was used to generate ohnolog families. A summary of individual ohnologs, ohnolog pairs and ohnolog families for these genomes is given in S2 Fig for strict, intermediate and relaxed quantitative criteria.
The level of annotation of these genomes is variable and the number of annotated protein coding genes range from 15,310 for chicken to 22,865 for the rat genome (S3 Fig). Using the relaxed criteria, a minimum of 4,282 to a maximum of 9,708 ohnolog pairs could be identified for chicken and rat, respectively. The six way Venn diagram in Fig 4 summarizes the fractions of retention versus lineage specific loss of ohnologs in the analyzed amniote genomes for the relaxed criteria (see S10 Fig for ohnolog numbers). Statistics for the strict criteria are given in S11 Fig. The identification of consensus ohnologs in this context implies that we are able to detect their ohnolog status through self- and outgroup synteny comparison or, alternatively, through orthology with bona fide ohnologs in other amniotes (see S1 Text). Indeed, ohnologs that are no longer in significant synteny in a particular vertebrate genome can still be identified, as long as their ortholog status can be unequivocally established with proper ohnologs in other vertebrates. This enables to circumvent strict synteny conditions in a specific genome.
By contrast to the small fraction of ohnolog genes identified by the six outgroups (i.e. 3.8%, Fig 4), 36.6% of predicted ohnologs are shared by all six amniotes, 53.9% by the five mammals and 74.3% by human, mouse and rat, while only a few other combinations of specific amniotes contribute to more than 2% of all ohnologs (see sectors shaded in red in Fig 4). This illustrates that the ohnologs have been largely conserved in mammals and to a lesser extent across amniotes. Likewise, ohnolog family sizes in each amniote genome consistently follow similar distributions as observed in human (Table 1) with a sharp decline in the number of families beyond the maximum size of four ohnologs (S2 Fig). In fact, the numbers of ohnologs in each family are most often the same in human and other mammals (in particular mouse) with occasional differences, typically missing ohnologs, in chicken which has significantly fewer genes (including ohnologs) than other amniotes considered in this study. For example, chicken has lost a number of adipokine genes [37] such as SERPINE1, which is part of a four-ohnolog family in mammals, SERPINE1/SERPINE2/SERPINE3/SERPINI1|SERPINI2 (where SERPINI1 and SERPINI2 are small scale duplicates). Similarly, all three ohnolog genes in the family of DNA binding Forkhead box protein A, i.e. FOXA1/FOXA2/FOXA3, are missing in the annotated chicken genome. Hence, differences in the shared ohnologs in Fig 4 arise due to lineage specific ohnolog loss or, possibly, due to missing annotations of genes and/or orthologs in these genomes.
We have so far restricted our synteny conservation analysis across multiple genomes to selected amniote genomes. In particular, amphibians and fishes have not been included in the analysis. This is because assembled chromosomal scaffolds of available amphibians (e.g. Xenopus) and non-teleost fishes (e.g. elephant shark and coelacanth) do not contain enough genes to be included in a content-based synteny conservation analysis (e.g. 81% of X. tropicalis genes are on chromosomal scaffolds with fewer than 50 genes). As for teleost fish genomes, they experienced a third more recent (3R) WGD, about 300 MY ago [38] in addition to the two rounds of (2R) WGD common to all vertebrates. This additional 3R WGD implies methodological issues specific to teleost fish genomes, which will be addressed in a forthcoming extension of our computational approach to identify ohnologs through multiple genome synteny comparison.
As outlined in the introduction, ohnologs have been reported to be preferentially retained in functional categories associated with development, signaling and gene regulation in the human genome [3, 7–10]. We performed a Gene Ontology (GO) enrichment analysis on four amniote vertebrates using DAVID [39] and observed the same general trend across these amniote genomes (Fig 5A). This confirms that ohnologs are associated with similar functional categories in different vertebrates.
In addition, ohnologs have also been associated with disease mutations [5, 12–14], in particular with dominant deleterious mutations frequently implicated in cancers and dominant genetic diseases [5, 6, 15]. Fig 5B confirms such cancer and genetic disease associations for all three ohnolog confidence criteria adopted in this study. This is particularly significant for core cancer genes [5, 40] (amounting for just 8.3% of non-ohnologs but up to 21.6–26% of ohnologs, i.e. a 2.6–3.1 fold increase, p = 3.4 × 10−153 Fisher Exact Test) and autosomal dominant diseases (amounting for just 2.1% of non-ohnologs but up to 5.4–5.9% of ohnologs, i.e. a 2.6–2.8 fold increase, p = 3.4 × 10−27 Fisher Exact Test) in agreement with earlier reports [5, 6] and evolutionary models [15]. We also analyzed the enrichment of ohnologs in genes with autoinhibitory protein folds, which are prone to dominant deleterious mutations. To this end, we collected genes with autoinhibitory protein folds either from careful literature curation [5] or based on the annotation of structural domains frequently associated with autoinhibition (i.e. SH3, DH, PH, CH, Drf and Eth domains), identified using Hidden Markov Model (HMM) search [41] against the PFAM database [42] (see Supplementary Methods). We observed that the ohnologs are particularly enriched in genes with autoinhibitory protein folds (amounting for just 1.4% of non-ohnologs but up to 9–12% of ohnologs, i.e. a 6.4–8.6 fold increase, p = 4.4 × 10−150 Fisher Exact Test) [5].
The data of all the ohnolog pairs and families for the six vertebrate genomes is accessible through the ‘Ohnologs’ server at http://ohnologs.curie.fr/. There, users can i) search for a particular gene, ii) browse pre-compiled ohnolog families and ohnolog pairs or iii) generate ohnolog families based on their own, user-defined, quantitative filters. The server is implemented in Perl-CGI and is hosted on a virtual machine at Institut Curie.
On the Search page (S12 Fig), the user can search for a gene of interest in any of the six available vertebrates using either Ensembl Id, gene symbol or any desired keywords. Search by functional categories is also possible using Gene Ontology Id or term. If a keyword search does not match any gene directly, we display all the genes matching that keyword in gene symbol, text description or GO term. A hyperlink from this page directs to the details on its ohnolog families and its possible association with human diseases points to Genecards [43] and Cosmic [44] databases. This page also contain links to details in UniProt and Entrez databases if available. If the gene exists in our analysis, and is an ohnolog, users are directed to the details about ohnolog families for each statistical confidence levels (i.e., strict, intermediate and relaxed criteria), S13 Fig.
Alternatively, users can also generate ohnolog families using our multi genome comparison analysis, for any of the six available vertebrate genomes using an arbitrary, user-defined, quantitative criteria for the outgroup and self comparisons. The default values correspond to the strict criteria. The result pages display all the pre-calculated or custom generated families, which can also be downloaded.
In the light of the importance of ohnologs in the evolution of vertebrates and their enhanced association with diseases, our analysis provides a useful resource to gain further insights on the impact of WGD in extant vertebrates.
|
10.1371/journal.pcbi.1006211 | Accounting for non-stationarity in epidemiology by embedding time-varying parameters in stochastic models | The spread of disease through human populations is complex. The characteristics of disease propagation evolve with time, as a result of a multitude of environmental and anthropic factors, this non-stationarity is a key factor in this huge complexity. In the absence of appropriate external data sources, to correctly describe the disease propagation, we explore a flexible approach, based on stochastic models for the disease dynamics, and on diffusion processes for the parameter dynamics. Using such a diffusion process has the advantage of not requiring a specific mathematical function for the parameter dynamics. Coupled with particle MCMC, this approach allows us to reconstruct the time evolution of some key parameters (average transmission rate for instance). Thus, by capturing the time-varying nature of the different mechanisms involved in disease propagation, the epidemic can be described. Firstly we demonstrate the efficiency of this methodology on a toy model, where the parameters and the observation process are known. Applied then to real datasets, our methodology is able, based solely on simple stochastic models, to reconstruct complex epidemics, such as flu or dengue, over long time periods. Hence we demonstrate that time-varying parameters can improve the accuracy of model performances, and we suggest that our methodology can be used as a first step towards a better understanding of a complex epidemic, in situation where data is limited and/or uncertain.
| As our world becomes more and more globalized, infectious disease poses an ever-increasing threat to human health. The multitude of environmental and behavioral factors, which account for the spread of infectious diseases, are ever-evolving and thus infectious diseases propagation is complex.
In the face of this complexity, mathematical models offer valuable tools to study the dynamics of epidemic diseases. Developing adequate statistical and mathematical tools, that take account of the time-varying nature of the different mechanisms responsible for disease propagation, remains a major challenge. To take this increasingly important aspect into consideration, we propose a flexible methodology that encompasses time-varying aspects of the epidemic. It does this via diffusion process equations for time-varying parameters. Considering the relative paucity of available data, our principal assertion is that it is preferable to use this flexible framework with time-varying parameters, that tracks epidemiological patterns, and updates the key parameters according to data, than to use a more complex model.
| Our world constantly faces the threat of emerging and re-emerging diseases and it has been shown that this has intensified over the past 50 years. This intensification is due, in part, to climate change, urbanization and globalization [1] meaning that infectious diseases remain a constant and unpredictable threat to human health.
Numerous factors contribute to the propagation of an infectious disease. These include increased human connectivity, limited availability of economic resources for adequate intervention, increasing antimicrobial resistance, evolution of the dominant strains and increasing parasite and vector resistance to the most widely used drugs and insecticides, etc. A key factor in this huge complexity is non-stationarity [2], meaning that the characteristics of the dynamical epidemiological processes evolve with time. Thus, the mechanisms of transmission are uncertain, making it difficult to obtain quantitative predictions. One of the classic aspects of non-stationarity, is the seasonality of epidemiological dynamics, linked to environment and climate [3–4] but the environmental variability can shape the disease propagation in unforeseeable ways on small and large spatial scales [5–8]. Intervention and control may also modify the course of an epidemic. A less well-described but equally important cause of non-stationarity is linked to social cycles, e.g. school terms, religious holidays and agricultural cycles [9–13]. Research increasingly focuses on the effect of behavioral change in the presence of epidemiological risk as a source of non-stationarity [14–16]. Societal responses and changing human behavior play an important role in our connected society. Thus, during an epidemic, depending on the availability of information on the disease, people exhibit a variety of behaviors including anxiety and social distancing that might greatly influence the course of an epidemic.
For all of the above reasons, the spread of pathogens through human populations can be complex and hard to predict. In the face of this complexity, mathematical models offer valuable tools to study the dynamics of epidemic diseases, in order to synthesize information to understand observed epidemiological patterns and to test different hypothesis on the underlying key mechanisms [17]. Moreover, mathematical models play a crucial role in infectious disease prevention by assessing the impact of different control measures, e.g. vaccination strategies [18–19].
Nonetheless, there are very few, if indeed any, cases where modelers can access all the necessary information to reliably predict the course of an epidemic. This is particularly the case when we consider the non-stationarity features of epidemics and their transient nature poses a challenging problem for modeling. Further to this, different hypothesis must be formulated. In the case of influenza, for example, some researchers have suggested using a quantitative relationship between climatic variables and the effective transmission rate [20]. Another recent example illustrates non-stationarity in epidemiology. Between November 2010 and February 2011, despite a low level of population susceptibility, an unexpected third wave of infection by the H1N1pdm09 pandemic virus was observed in the United Kingdom. Using a compartmental mathematical model of influenza transmission, this third wave was explained, by a substantial increase in the transmissibility of the H1N1pdm09 virus [21]. It has been proposed that this modification of the transmissibility was caused by the virus evolution with a better adaptation to the human host, or by climatic factors, namely the very cold weather experienced in the United Kingdom at that time, or by a combination of these factors [21].
To tackle the problem of non-stationarity in epidemiology, some approaches use a linear function to reconstruct the effective reproduction number (average number of secondary cases per primary case, Reff). Wallinga and Teunis [22] proposed a generic method that requires only case incidence data and the distribution of the serial interval (the time between the onset of symptoms in a primary case and the onset of symptoms of secondary cases) to estimate Reff over the course of an epidemic. This approach has been improved by numerous authors and applied to real time estimation of Reff [23–25]. Other authors estimated using mathematical models Reff for each season [26,27]. To calculate the time-varying infection rate the reconstructed time series of Reff derived from the notification data can be used [28].
However, the time evolution of Reff by definition depends not only on the time evolution of the epidemiological parameters but also on the number of susceptibles. More complex approaches have therefore been proposed. These approaches use semi-mechanistic models that incorporate the known compartmental structure of disease transmission but do not specify the form of the transmission rate equation that is estimated based on the data. In an early paper, the force of infection is estimated by neuronal network or kernel regression [29]. Now it is more common to use B-spline [30–33]. An alternative approach is to use diffusion models driven by fractional Brownian motion to model time-varying parameter of major epidemiological significance [34–36]. The models developed assign diffusion processes to the time-varying parameters embedded in a state-space framework. With the Kalman filter, the time-evolution of some key parameters (average transmission rate, mean incubation rate, and basic reproduction rate) were estimated during the course of the HIV/AIDS epidemics in the Paris region [34–35]. Dureau et al. [36] generalized this approach using a Bayesian framework with an adjusted adaptive particle Markov chain Monte Carlo algorithm (PMCMC), but only applied to the transmission rate, for short epidemics, with application to the 2009 pandemic flu. Very recently, an algorithm relying on robustly estimating the time-varying infection rate, based on the method of the unknown input observers from control theory, has been proposed [37]. Similarly, an approach for the reconstruction of time-dependent transmission rates, by projecting onto a finite subspace, spanned by Legendre polynomials, has been introduced [38].
In our previous works [34–36], we have introduced an approach for reconstructing the time evolution of some key parameters with just the weak hypothesis according to which they follow a basic stochastic process. The parameter time evolution is estimated solely based on observations of the incidence or the prevalence. Here, we propose to expand this approach to recurrent epidemics over time periods longer than just one season. The underlying idea of this approach is to capture unknown influences by considering time-varying parameters. As with other semi-mechanistic approaches, the key advantage of this approach, for the parameter dynamics, is that it is data-driven, and thus the shape of change does not need to be specified beforehand. We applied our framework both to a toy model, where parameters and observation process are known, and to two real data sets. This allows us to demonstrate that this data-driven approach is very effective for tackling the non-stationarity of recurrent epidemics, even with long time series. It has other benefits too. For instance, with limited access to information, it can capture unknown influences. By so doing, and by analyzing the parameter time evolution, this framework allows a more thorough analysis of the different influences, facilitating their introduction in more complex models with pertinent hypotheses based on observations.
Our approach is based on three main components: an epidemiological model embedded in a state-space framework, a diffusion process for each time-varying parameter and an up-to-date Bayesian inference technique based on adaptive PMCMC.
The main advantage of the state-space framework is the use of two sets of equations, the first set describes the propagation of the disease in the population and the second is for the observation process. This allows for consideration of unknowns and uncertainty both in the epidemiological mechanisms and in the partial observation of the disease:
{x˙(t)=g(t,x(t),θ'(t),u(t))y(t)|x(t)∼f(h(x(t)),θ'(t))
(1)
The first equation is for the epidemiological model, with x(t) representing the state variables (for instance, S(t) the susceptibles, I(t) the infectious and R(t) the removed for the classical SIR model) and θ'(t) the epidemiological parameters. The second is the observational process defined by probabilistic law f and a reporting rate on transformation of some state variable h(x(t)) because we may not be able to directly measure all state variables but just some or a function of them. In these equations, y(t) are partial observations of x(t), u(t) is the process noise describing different form of stochasticity and the observational noise is included in f. In our applications, h(x(t)) will be the cumulative sum of new cases over the observation time step, that is generally the quantity observed by Public Health systems.
Considering the time-varying parameters θ(t) as a subset of θ'(t), we make the assumption that they evolve more or less randomly and do not follow a defined mathematical function. In the absence of prior information the use of diffusion motion allows us to impose few restrictions on the evolution of θ(t). We consider that they follow a continuous diffusion process (a discrete diffusion process was used in [35]):
dθ(t)=σdB(t)ordlog(θ(t))=σdB(t)
(2)
where σ is the volatility of the Brownian process (dB(t)) and will be estimated during the fitting process. The use of a Brownian process can be viewed as a weak hypothesis for the imposed motion of θ(t) and the volatility σ being a regularized factor. Intuitively, the higher the values of σ the larger the changes in θ(t). The logarithm transformation avoids negative values which have no biological meaning. When prior knowledge on θ(t) is available this Brownian process can be modified to account for a drift in (2) (see [36]).
For the time-varying parameter, we focus on the parameter of the force of infection classically defined as:
λ(t)=β(t).S(t).I(t)N
(3)
with β(t) the transmission rate usually defined by a sinusoidal function. The control or the behavior modification can also be taken into account:
λ(t)=β(t).(S(t)εS(t)).(I(t)εI(t))N
(4)
εi(t) describe the clustering of the population [39,40] but can also describe a reduction in the population due to voluntary avoidance behavior or social distancing. However due to the absence of structural identifiability properties [41, 42] it should be very difficult to estimate simultaneously both β(t) and εi(t).
For model estimation we use Bayesian methods, coupling particle filter and MCMC for partially observed stochastic non-linear systems [36,43] (see Methods). The implementation provided in SSM software [36] is used.
We start our demonstration by showing that it is possible to reconstruct both the trajectory of a SIRS model (SIRS stands for Susceptibles, Infectious, Removed and Susceptibles again) and that of the sinusoidal transmission rate. In this example, the trajectory of each variable has been simulated with a model for which all the parameters were known. Moreover we also knew the observation process that has generated the data, a Poisson law for the incidence with an observation rate equal to 1. Fig 1 displays the reconstructed trajectories of both the incidence and the transmission rate highlighting the potential of the method. The parameter estimations are in perfect agreement with the values used to generate the observations and the estimation process has correctly converged (S1 and S2 Figs). This clearly demonstrates the feasibility of accurately ascertaining the time evolution of the transmission rate and correctly estimating the Reff (see Fig 2). It is worth emphasizing that the SIRS model is a complicated example for different reasons. First, even with a constant transmission rate the SIRS model can generate oscillations (damped oscillations, see [17,44]). Secondly, the model trajectories are not very sensitive, a modification of ± 10% can induce minor modifications of the trajectories that are inside or near the 95% CI of our inferences (Fig 2). Moreover in this example we have used initial conditions outside the attractor of the dynamics to generate transients that appear more realistic for real applications, but are more complex to reconstruct. The robustness of our approach has also been tested: (i) using long time series and initial conditions near the attractor (Fig 3A and S3A Fig); (ii) modifying the number of inferring parameters (S4–S6 Figs), for instance estimating just the volatility parameter (S7 and S8 Figs); (iii) considering the possibility of not using the transformation log in the diffusion process (S9 and S10 Figs) and (iv) using a true β(t) with 2 or 3 periodic components (Fig 3B and 3C and S3B and S3C Fig).
We have also explored the performance of our approach by comparing their inferences to those of the true model. The re-estimation of the true model on its own data is displayed in S11–S13 Figs. Table 1 presents indices of the goodness-of-fit of the true model and models with time-varying β(t) with different number of parameters inferred. As expected, the error on β(t) is smaller when the true equation is used (Table 1). However, regarding the estimated incidence, the true model and our approach give similar results both in terms of mean and variance (Table 1). It could be argued that the price of the flexibility of our approach is a greater variability in some of the trajectory estimations (Table 1). Nevertheless the average dynamics are always estimated correctly.
As misspecification is an important problem (e.g. [45]) we have also compared the performance of our approach to those of a misspecified seasonal SIRS model. We have thus used the example of a sinusoidal β(t) with two periodic components (see Fig 3B) and computed the indices of the goodness-of-fit of the true model with the SIRS model with 1 year sinusoidal β(t) and with our time-varying periodic β(t). The results clearly show that our approach performed better than the misspecified model for the three trajectories analyzed, Incidence, β and Reff (Table 2). Once again the price of the flexibility of our approach is a greater variability in some of the trajectory estimations. However this is preferable to a large error in the median trajectories as occurred in those observed with the misspecified model (Table 2).
Our methodology is also applicable to other more complex or simpler tasks. For instance, it can follow the time evolution of a parameter describing the availability of susceptibles, εS(t) (Fig 4 and S14 and S15 Figs). Fig 4 shows the accurate reconstruction of the trajectory of the incidence and also of the trajectory of εS(t) that shifted at a given time point and decreased slightly thereafter. This highlights once again the potential of our approach as it is never easy to estimate a discontinuous dynamic with a continuous process (2).
In previous works, the dynamics of influenza in Israel have been analyzed using a discrete deterministic SIRS model and weekly data from Israel’s Maccabi health maintenance organization [20,46]. To describe the seasonality of this recurrent epidemic, the authors used a linear model between the transmission rate and local climatic variables, daily temperature and relative humidity [20,46]. We have re-analyzed their dataset (but limited to 1998–2003 due to a modification in the reporting) to reconstruct the time evolution of β(t). Our results (Fig 5 and S16 Fig) clearly show the potential of our method, highlighting that the β(t) fluctuations are more irregular and complex than a simple sinusoidal function.
Our last example is on dengue in Cambodia. Again the idea is to relax the assumption of a sinusoidal β(t) in a SEIR model. Monthly data from the capital Phnom Penh [47], for which the meteorological data is available from the international airport, was used. We can accurately describe the 12 year time series and reconstruct the time evolution of β(t) (Fig 6 and S17 Fig). Our results stress that the β(t) oscillations are more complex than a simple sinusoidal function. Sometimes bi-modality occurs over one season. In general one observes a fast growth of β(t) and a slow decrease. Moreover the amplitude of the β(t) varies from year to year, perhaps depending on the fluctuations in the mosquito population and in the environment. Interestingly the peak in β(t) appears 1 to 2 months before the incidence peak. This delay can be explained by the extrinsic incubation period and might be used in a warning system.
To explain the β(t) oscillations we have explored the potential effects of local and global climatic variables using wavelet decomposition [48] as one of our main underlying hypotheses is non-stationarity. We observed very significant coherency between β(t) and climate for the local climate for the seasonal mode (Fig 7 and S18–S20 Figs) and also for the 2–3 year components with global climatic variable (S21 Fig). Thus, the rhythm of β(t) can be explained perfectly by climatic factors. Nevertheless, again mainly due to large non-stationarity, by using solely one or two climatic variables we are able to correctly describe dengue evolution in the short-term (Fig 7C, red area) but not over a large time period (Fig 7C, blue area). This reflects the complexity of such a disease where the ecology of the vectors, the environment, the climate, the immune status of the human population and its behavior are all involved. This large non-stationarity association between dengue and climatic factors has recently been demonstrated using statistical models (dynamic generalized linear models) and data from a medium-sized city in Colombia [49]. The authors showed that dengue cases correlate with climatic variables (temperature, rainfall, solar radiation and relative humidity) but these correlations change over time, some intervals showing a positive association, while in others the association is negative [49]. The non-stationarity association between dengue and climate may be explained by the fact that a climatic variable has different effects depending on the biological cycle of the pathogen or of the vector. Moreover the effects of one climatic variable can also depend on other climatic variables potentially enhancing the non-stationarity association.
As there remain numerous uncertainties during the course of each epidemic, we are increasingly aware of the importance of developing adequate statistical and mathematical tools. Such tools need to take account of the time-varying nature of the underlying ecological and biological mechanisms as well as social and behavioral influences involved in an epidemic. Because of this, time-varying parameters modeled with a diffusion process, that track epidemiological patterns and update the key parameters according to data appear to be a worthwhile approach. Indeed developing a more complex model would be difficult considering the relative paucity of available data.
We propose a flexible modeling framework that encompasses time-varying aspects of the epidemic. It does this via diffusion process equations for time-varying parameters and also considers uncertainty associated with key parameters and data. This data-driven framework for time-varying parameters has been coupled with simple stochastic models and a robust Bayesian procedure for inference. To test its efficiency, our proposed methodology was first applied to a toy model and then to real epidemiological examples.
Our results clearly demonstrate the potential of our framework. Firstly, our methodology was able to accurately reconstruct both the incidence and the sinusoidal transmission rate of a simple SIRS model just based on noisy observations (Figs 1–4 and S4,S5,S7,S9 and S14 Figs). Based on these reconstructions one can also closely estimate Reff which is one of the key relevant epidemiological parameters. Our results also highlight the flexibility of our developed methodology. It can reconstruct the time evolution of a shifting parameter (εS(t), see Fig 4 and S14 Fig) as well as an oscillating parameter that influences the nonlinear part of the model (β(t), see Figs 1–3 and S4,S5,S7 and S9 Figs). The comparison using goodness-of-fit indices with the inferred true model allows us to highlight the fact that our methodology performs as well for the observed incidence. Its flexibility results in greater variability in some other trajectories mainly β(t) and Reff (Table 1). Moreover, in the absence of knowledge of the true evolution of the transmission rate, our approach appears to capture the dynamic observed more accurately than a misspecified model (Table 2). Secondly, applied to real datasets, our framework is able, based solely on simple stochastic models, to reconstruct complex epidemics such as flu or dengue over long time periods (Figs 5 and 6). In such cases, the reconstruction of the time evolution of the transmission rate clearly stresses that, on real datasets, it is difficult to assimilate the dynamic of this parameter as a simple sinusoidal function. It is more irregular in amplitude and sometimes multi-modal over one season.
Considering the paucity of information available regarding the complexity of the mechanisms involved during an epidemic, describing and fitting a full model for a given transmissible disease is always challenging. Our data-driven methodology can be used as a first step towards a better understanding of a complex epidemic, where data is limited or lacks certainty. Indeed most of the unknowns and uncertainties can be put into time-varying parameters. The potential effects of all these uncertainties can then be explored by analyzing the reconstructed time evolution of the time-varying parameters. See Fig 7 for such preliminary analysis of dengue in Phnom Penh. This allows a more thorough analysis of the influences and the interactions between both the human behavior and complex environmental drivers. In a recent paper [50], the authors reviewed evidence of interactions between seasonal influenza virus and other pathogens (bacteria or virus). They concluded that it is important to incorporate these different coinfecting pathogens in models of seasonal flu in order to get a better estimate of the burden of influenza. Our framework could be an alternative to the development of complex models with all the potential interactions between pathogens and to estimate the strength of the interactions. After reconstructing the time evolution of the transmission rate the statistical association between the coinfecting pathogens and the transmission rate could be tested. This screening may facilitate the construction of more complex models that could incorporate only the most significant coinfecting pathogens in the seasonal flu model.
Our methodology also has other advantages. Taking account of the simplicity of the model used, and the fact that weak hypotheses on the dynamics of the time-varying parameters have been included, our proposed methodology can retrospectively test the impact of interventions. This has previously been done in the case of HIV epidemics [34–35], where it was hypothesized that the reduction in the transmissibility was due to a modification of the sexual behavior in the population and the increase in the seropositive period duration due to the introduction of the first antiviral treatments. Evaluation of interventions has also been done recently in the case of the Ebola epidemic in West Africa [51]. The relative simplicity of our methodology is also suitable for short-term predictions and it can then easily be used to predict an epidemic in real time. Starting with a given estimated state defining the system, the fitting process can be run again each time new data is available and the new posteriors are used for new predictions [36]. This can inform public health decisions and indeed has been done recently to great effect in the case of the Ebola epidemics in West Africa [52].
A major challenge in model fitting is the reliability of data collected and also the non-identifiability of the mechanistic models that always have very rich dynamical behavior. The question of identifiability is too often avoided in epidemiological models applied to a topical Public Health issues. There is, however, considerable literature on this subject (e.g. [41,42,53–55]). Identifiability is not evident even for a simple seasonal SIR model [56]. To solve this problem one can fit a combination of parameters or fix some of them (the population size for instance) [57]. In our applications there is a clear limitation due to practical non-identifiability of reporting rate and initial conditions. To fix these problems we have used informative priors (see Method). Using informative priors or fixing some parameters gives very similar results (compare Fig 1 and S4–S7 Figs). Related to this is the misspecification of models [45]. In our cases, as with other semi-mechanistic models the time-varying parameter methodology captures some of the information in the data but not in the mechanistic part of the model. If the model is misspecified due to lack of precision, it compensates for it and the dynamics of β(t) will drive improvements in the model to make it more complex and realistic (Table 2). If the model is misspecified to the extent that it creates mechanisms that do not exist, the reconstructed β(t) would compensate for these effects but it will be harder to interpret.
In this work we have used simple mechanistic models. The proposed methodology is not limited to simple models. For instance, a two-strain dengue model has also been tested. In this case the main problem was linked to the unavailability of both seroprevalence and incidence for each strain. Indeed, one of the major difficulties with these multi-strain models is the identification of the initial conditions (e.g. [58]). Nevertheless it is worth emphasizing that the Bayesian inference method used in our framework, PMCMC, the approximation of the likelihood is limited for a large number of parameters and/or equations [59]. In such cases testing other methodologies like ABC [60,61] is advisable.
It is always difficult to fit complex models with rich behaviors based on very limited information. In this regard we agree with Metcalf et al. [62] who stressed that nowadays we need seroprevalence studies to quantify the immunological status of the population, because in most cases the magnitude of the outbreak is difficult to evaluate without precise seroprevalence data.
To tackle the uncertainty and the non-stationarity of epidemics, our methodology, although it appears non-standard, makes important progress towards a better understanding of the mechanisms responsible for disease propagation. We believe that, should it form part of the development of the next predictive tools for Public Health, it will make a significant contribution to improving the understanding and control of infectious diseases in our increasingly uncertain world.
Among the various approaches developed to study nonstationary data, wavelet analysis is probably the most efficient. In particular, this method gives us the possibility of investigating and quantifying the evolution in time of the periodic components of a time series (see [69]). Wavelets constitute a family of functions derived from a single function, the ‘‘mother wavelet”, Ψa,τ(t), that can be expressed as the function of two parameters, one for the time position τ, and the other for the scale of the wavelets a, related to the frequency. More explicitly, wavelets are defined as:
Ψa,τ(t)=1aψ(t−τa)
The wavelet transform of a time series x(t) with respect to a chosen mother wavelet is performed as follows:
Wx(a,τ)=1a.∫−∞+∞x(t).Ψ*(t−τa).dt=∫−∞+∞x(t).Ψa,τ*.dt
where * denotes the complex conjugate form. The wavelet transform Wx(a,τ) represents the contribution of the scale a to the signal at different time positions τ. The computation of the wavelet transform is done along the signal x(t) simply by increasing the parameter τ over a range of scales a until all coherent structures within the signal can be identified. Here, as mother wavelet, we have used the Morlet wavelet [69].
With the wavelet approach, we can estimate the repartition of variance at different scale a and different time location τ. This is known as the wavelet power spectrum: Sx(a,τ) = | Wx(a,τ) |2. An important point with the continuous wavelet is that the relationship between the wavelet frequency f0 and the wavelet scale a can be derived analytically. For the Morlet wavelet this relationship is given by:
1f=4πaf0+2+f02
Then when f0 = 2π, the wavelet scale a is inversely related to the frequency, f ≈ 1/a. This greatly simplifies the interpretation of the wavelet analysis and one can replace, on all equations, the scale a by the frequency f or the period 1/f.
To determine the statistical relationship between two time series, wavelet coherence can be computed (e.g. [48,70]):
Rx,y(f,τ)=(|〈Wx,y(f,τ)〉|2|〈Wx(f,τ)〉|2.|〈Wy(f,τ)〉|2)1/2
where the angle brackets around terms indicate smoothing in both time and frequency, Wx(f,τ) is the wavelet transform of series x(t), Wy(f,τ) is the wavelet transform of series y(t), and Wx,y(f,τ) is the cross-wavelet spectrum. The values of wavelet coherence are between 0 < Rx,y(f,τ) < 1. The wavelet coherency is equal to 1 when there is a perfect linear relation at particular time and scale between the two signals, and equal to 0 if x(t) and y(t) are independent.
To complement this, phase analysis can be used to characterise the association between signals (e.g. [48,70]). The phase difference provides information on the sign of the relationship (i.e., in phase or out of phase) and can be computed, for complex mother wavelet, with the wavelet transform Wx(f,τ) as:
ϕx(f,τ)=tan−1Im(Wx(f,τ))Re(Wx(f,τ))
Similarly with the cross-wavelet transform Wx,y(f,τ) the phase difference between the two time series can be computed:
ϕx,y(f,τ)=tan−1Im(Wx,y(f,τ))Re(Wx,y(f,τ))
|
10.1371/journal.ppat.1005468 | NAD+-Glycohydrolase Promotes Intracellular Survival of Group A Streptococcus | A global increase in invasive infections due to group A Streptococcus (S. pyogenes or GAS) has been observed since the 1980s, associated with emergence of a clonal group of strains of the M1T1 serotype. Among other virulence attributes, the M1T1 clone secretes NAD+-glycohydrolase (NADase). When GAS binds to epithelial cells in vitro, NADase is translocated into the cytosol in a process mediated by streptolysin O (SLO), and expression of these two toxins is associated with enhanced GAS intracellular survival. Because SLO is required for NADase translocation, it has been difficult to distinguish pathogenic effects of NADase from those of SLO. To resolve the effects of the two proteins, we made use of anthrax toxin as an alternative means to deliver NADase to host cells, independently of SLO. We developed a novel method for purification of enzymatically active NADase fused to an amino-terminal fragment of anthrax toxin lethal factor (LFn-NADase) that exploits the avid, reversible binding of NADase to its endogenous inhibitor. LFn-NADase was translocated across a synthetic lipid bilayer in vitro in the presence of anthrax toxin protective antigen in a pH-dependent manner. Exposure of human oropharyngeal keratinocytes to LFn-NADase in the presence of protective antigen resulted in cytosolic delivery of NADase activity, inhibition of protein synthesis, and cell death, whereas a similar construct of an enzymatically inactive point mutant had no effect. Anthrax toxin-mediated delivery of NADase in an amount comparable to that observed during in vitro infection with live GAS rescued the defective intracellular survival of NADase-deficient GAS and increased the survival of SLO-deficient GAS. Confocal microscopy demonstrated that delivery of LFn-NADase prevented intracellular trafficking of NADase-deficient GAS to lysosomes. We conclude that NADase mediates cytotoxicity and promotes intracellular survival of GAS in host cells.
| Invasive infections due to group A Streptococcus (S. pyogenes or GAS) have become more frequent since the 1980s due, in part, to the emergence and global spread of closely related strains of the M1T1 serotype. A feature of this clonal group is the production of a secreted enzyme, NAD+-glycohydrolase (NADase), which has been suggested to contribute to GAS virulence by intoxication of host cells. For NADase to exert its toxic effects, it must be translocated into the host cell by a second GAS protein, streptolysin O (SLO). SLO is a pore-forming toxin that damages cell membranes in addition to its role in translocating NADase. In order to distinguish effects of NADase on host cell biology from those of SLO, we used components of anthrax toxin to deliver NADase to human throat epithelial cells, independently of SLO. Introduction of NADase into GAS-infected cells increased the intracellular survival of GAS lacking NADase or SLO, and the increase in bacterial survival correlated with inhibition of intracellular trafficking of GAS to lysosomes that mediate bacterial killing. The results support an important role for NADase in enhancing GAS survival in human epithelial cells, a phenomenon that may contribute to GAS colonization and disease.
| Since the 1980’s, there has been a sustained, worldwide increase in the incidence of severe, invasive infections due to group A Streptococcus (Streptococcus pyogenes or GAS), particularly necrotizing fasciitis and streptococcal toxic shock syndrome [1–3]. The reasons for the emergence of invasive GAS disease are incompletely understood; however, a partial explanation may be the global dissemination of a clonal group of strains of the M1T1 serotype. The invasive M1T1 strains harbor bacteriophage-associated genes encoding such virulence factors as the pyrogenic exotoxin SpeA and the secreted DNase Sda1 (also called SdaD2), both of which have been associated with GAS pathogenicity in model systems. In addition, these strains secrete NAD+-glycohydrolase (NADase), a property that generally was not present among M1 strains isolated prior to 1988 [4–6]. NADase is encoded by nga, which is located in an operon together with ifs, encoding an intracellular inhibitor that dissociates from NADase upon NADase secretion, and slo encoding the cholesterol-dependent cytolysin/hemolysin, streptolysin O (SLO) [4,7–9]. Genomic analyses of multiple M1 isolates from the past century indicate that the invasive M1T1 strain acquired a 36-kb chromosomal region that includes the nga and slo genes prior to emergence of this strain in the 1980s [10–12]. The association of NADase activity with contemporary invasive M1T1 isolates has suggested that production of the enzyme might contribute to virulence.
Physical association of NADase with hemolytic activity in GAS culture supernatants led to early misidentification of NADase and SLO as a single protein, although subsequent studies clearly separated the two [13–15]. A new paradigm for the interaction of NADase and SLO was proposed by Madden et al., who found that NADase could be translocated into the cytosol of epithelial cells in vitro after its secretion from GAS bound to the cell surface [16]. Translocation required the concomitant expression of SLO, which suggested a model in which NADase associates with SLO on the epithelial cell surface and is transferred across the cell membrane in a process dependent on SLO. These and subsequent studies provided evidence that SLO-mediated delivery of NADase augmented the cytotoxic effect of SLO and induced epithelial cell apoptosis [16,17]. NADase-deficient mutants were found to have reduced virulence in mice compared to wild type GAS, supporting a role of the enzyme in pathogenesis of invasive infection [18,19].
The exposure of human oropharyngeal keratinocytes to GAS that produce both SLO and NADase, but not to those producing SLO alone, results in depletion of intracellular NAD+ and ATP. This finding is consistent with the enzymatic function of NADase to hydrolyze cellular NAD+ to nicotinamide and adenosine diphosphoribose and, secondarily, to deplete cellular ATP [20]. In previously published work, we used isogenic mutants deficient in SLO or NADase to study the role of each toxin in enhancing intracellular survival of GAS. These studies revealed that NADase-deficient GAS are more efficiently killed after internalization by keratinocytes compared to SLO+NADase+ GAS [21]. The increased survival of NADase-producing strains is associated with failure of GAS-containing vacuoles to fuse with lysosomes to form an acidic, bactericidal compartment [21]. While these observations have suggested a role for NADase in GAS pathogenesis, prior studies have been limited in their capacity to distinguish effects of NADase from those of SLO, since SLO is itself a cytotoxin and is required to deliver NADase to host cells. The goal of the present investigation was to distinguish effects of NADase from those of SLO during interaction of GAS with human epithelial cells. To this end, we developed a system that delivers NADase to the cytosol of host cells independently of SLO. Utilizing the anthrax toxin platform to deliver enzymatically active or inactive forms of recombinant NADase to cells infected with various GAS strains, we have obtained direct evidence that the catalytic activity of NADase is a critical effector of GAS intracellular trafficking and intracellular survival.
Anthrax toxin, the major virulence factor of Bacillus anthracis, is an A-B type toxin composed of the catalytic moieties, lethal factor (LF) and edema factor (EF), and the receptor binding/pore forming protective antigen (PA; MW 83 kDa). Upon release by the bacteria, PA83 binds to its cellular receptors and is cleaved by cell surface furin to a 63 kDa form (PA63), which then self-assembles to form a heptameric or octameric prepore [22–24]. The prepore binds the enzymatic LF and/or EF moieties to form complexes that are subsequently endocytosed [25,26]. The low pH of the endosome causes the PA prepore to undergo a conformational change into the pore form, which inserts into the endosomal membrane and translocates the catalytic LF and EF moieties into the cytoplasm [27–29].
The intrinsic activity of the anthrax toxin system for intracellular delivery of its catalytic components can be harnessed to translocate heterologous proteins into the cytosol of its target cells. Fusing the non-catalytic N-terminal PA-binding domain of LF (LFn, residues 1 to 263)) [30] to any of a variety of unrelated “cargo” proteins enables them to undergo PA-dependent translocation to the cytosol. Examples include a cytotoxic T lymphocyte epitope from Listeria monocytogenes, the gp120 portion of the HIV-1 envelope protein, and the activity domains of Pseudomonas exotoxin, diphtheria toxin, or shiga toxin [31–35]. In the current study, we fused LFn to NADase or its variants and utilized the anthrax toxin platform to deliver enzymatically active or inactive forms of the enzyme to human oropharyngeal keratinocytes independently of SLO. Results of in vitro infection experiments utilizing this system provide direct evidence that the enzymatic activity of NADase is a critical effector of GAS intracellular trafficking and survival.
Functional analysis of GAS NADase has been complicated by the necessity to co-express its endogenous inhibitor IFS (Immunity Factor for Streptococcal NADase) to prevent toxicity to the cell that produces the active enzyme [8,9]. IFS must be removed for NADase to be enzymatically active. Previously, expression and purification of NADase in E. coli was achieved by directing secretion of recombinant NADase to the periplasmic space, allowing IFS to remain in the cytosol [9,36]. In our hands, the yield was low with this approach, and a portion of IFS remained in the NADase-containing fraction, presumably due to incomplete exclusion of cytosolic proteins in the periplasmic preparation.
In order to produce sufficient quantities of NADase free of IFS, we developed a novel scheme for expression and purification. Because initial experiments indicated low expression levels of NADase and its fusion constructs, the nga and ifs gene sequences were codon-optimized for expression in E. coli. We then exploited the high-affinity binding of IFS to NADase to purify native and variant forms of the enzyme and various fusion constructs using His6-tagged IFS (S1 Fig). In the first step, we purified the NADase-IFS-His6 complex, which bound to a Ni-charged resin. His6-tagged IFS was then released from untagged NADase by denaturing the two proteins with guanidinium chloride. A second round of affinity chromatography was used to separate His6-tagged IFS, which was retained by the Ni column, from untagged NADase in the flow through fraction. NADase was then refolded slowly by removal of guanidinium chloride by dialysis. High protein purity was achieved by Q column purification of proteins after the first Ni column affinity purification, and then again after renaturation of IFS-free NADase constructs. Each of the purified recombinant proteins migrated predominantly as a single band of the expected molecular size on SDS-PAGE (Fig 1A).
In addition to native NADase, two variant forms were expressed and purified, both as individual proteins and as fusions to LFn. Variant 190NADase lacks the N-terminal 190 amino acids required for SLO-dependent translocation of NADase [37]; NADaseG330D harbors a point mutation that almost completely abrogates NAD-glycohydrolase activity [6,38,39]. Since the protocol involved protein denaturation and renaturation, we confirmed that the purified LFn-NADase, LFn-190NADase, and NADase proteins retained similar levels of NAD-glycohydrolase activity (Fig 1B). The Kcat value for LFn-NADase was estimated at 4200 reactions/sec, which compares favorably with published estimates of 3700 and 8000 reactions/sec, determined for purified NADase using a highly sensitive HPLC-based assay [36,38]. LFn-NADaseG330D and NADaseG330D lacked detectable catalytic activity. However, both LFn-NADaseG330D and NADaseG330D were able to compete with NADase for binding of IFS after renaturation (S2 Fig). We also analyzed the secondary structure of purified recombinant NADaseG330D by circular dichroism spectroscopy and found nearly identical results as those for purified recombinant (and enzymatically active) NADase (S2 Fig). Together, these analyses provide evidence that renaturation of the enzymatically inactive variants LFn-NADaseG330D and NADaseG330D restored the native conformations of the purified proteins.
We tested the ability of LFn-NADase and its variants to interact with and translocate across PA pores in planar bilayers in vitro as measured by ion conductance. Occlusion of pores in DPhPC bilayers was monitored for 60 sec following addition of each recombinant protein (final concentration 1 μg/ml) to the cis compartment. All of the constructs tested (LFn, LFn-NADase, LFn-190NADase, LFn-NADaseG330D, and LFn-190NADaseG330D) blocked conductance rapidly (within 20 sec) and almost completely (Fig 2A). Subsequently, translocation was initiated by addition of KOH to the trans compartment to increase the pH to ~7.5. Translocation of free LFn and LFn-190NADase, as measured by return of ion conductance, was rapid (within ~80 sec) and essentially complete (~80–90%) (Fig 2B). LFn-NADase took longer (240 sec) to achieve comparable translocation. LFn-NADaseG330D and LFn-190NADaseG330D constructs were less efficiently translocated, with about ~60% translocation achieved in 240 sec. Interestingly, addition of IFS to the cis compartment (final concentration 6 μg/ml) before addition of KOH to the trans compartment prevented translocation of LFn-NADase (Fig 2B). To test whether the binding of NADase to IFS prevents NADase unfolding, which is necessary for translocation, we used differential scanning fluorimetry to measure the melting temperature of NADase, IFS, and NADase-IFS complex. The melting temperatures were determined to be 43°C, 60°C and 76°C, respectively (S3 Fig). Thus, the tight binding of IFS increases the Tm of NADase by more than 30°C, presumably preventing its unfolding, a required step for the translocation of LFn-NADase across PA pores.
Having determined that LFn-NADase could be translocated through PA pores in an artificial membrane in vitro, we investigated whether PA pores could mediate delivery of LFn-NADase into human oropharyngeal keratinocytes. We reasoned that NAD+-glycohydrolase activity of LFn-NADase would deplete cellular energy stores resulting in inhibition of protein synthesis. Accordingly, NADase and its variants were tested for PA-mediated translocation into OKP7 cells by measuring inhibition of cellular protein synthesis. In the presence of PA, LFn-NADase and LFn-190NADase were efficiently translocated, with half-maximal inhibition of protein synthesis observed at a LFn-NADase concentration of ~1 nM in the cell culture medium (Fig 2C). LFn-190NADase gave an almost identical result, a finding that implies the N-terminal domain of NADase involved in SLO-mediated translocation is dispensable for delivery of the enzyme by the anthrax toxin system. In the absence of PA, no LFn-NADase translocation was observed (S4 Fig), and, as expected, NADase and 190NADase also did not translocate. Translocation of the enzymatically inactive forms of NADase, LFn-NADaseG330D and LFn-190NADaseG330D, did not inhibit cellular protein synthesis, even at concentrations up to 1,000 times that required for inhibition by LFn-NADase (Fig 2C).
A key determinant of translocation by PA is the phenylalanine clamp, a structure formed by the F427 side chains within the lumen of the PA pore [40]. We tested LFn-NADase and LFn-190NADase for cytotoxicity in the presence of PA F427H, a mutant form of PA, which forms pores that lack the ability to mediate translocation. The F427H mutation completely blocked LFn-NADase translocation (Fig 2C), implying PA-mediated translocation is dependent on interaction with the Phe clamp and occurs through the central pore.
It has been suggested that introduction of NADase into host cells exerts cytotoxic effects that are independent of NAD+-glycohydrolase activity of the protein [38]. The anthrax toxin delivery system enabled us to test this hypothesis in the absence of other GAS virulence factors. We found that exposure of OKP7 keratinocytes to LFn-NADase in the presence of PA resulted in rounding, pyknosis, and uptake of propidium iodide indicating loss of cell viability (Fig 3). Treatment with LFn-NADase resulted in 52% cell death as assessed by propidium iodide staining. In addition, treatment with LFn-NADase caused significant cell loss when compared to untreated cells, presumably due to cells becoming non-adherent upon loss of viability. In contrast, identical exposure to enzymatically inactive LFn-NADaseG330D in the presence of PA caused no cytotoxicity compared to untreated cells (1% cell death for each condition). These results provide direct evidence that the cytotoxic effects of NADase are due solely to its enzymatic activity.
Previous studies on the effects of NADase on epithelial cells have utilized model systems in which cells are exposed to live GAS in vitro [17,21]. In order to compare effects of NADase delivered by the anthrax toxin system with those associated with exposure to live GAS, we measured intracellular NAD-glycohydrolase activity under both conditions. Our goal was to determine the concentration of LFn-NADase to be added to OKP7 cells so that the subsequent PA-mediated delivery would result in an intracellular NADase activity comparable to that achieved by exposure to live GAS in prior studies. We found that addition of LFn-NADase to a concentration of 10 nM achieved a level of NADase activity in the cytosol of the keratinocytes that corresponded to approximately 50% of that associated with infection by NADase-producing GAS strain 188 at an MOI of 10 (Fig 4, S5 Fig).
Infection of OKP7 cells with GAS is associated with survival of 10 to 15% of intracellular bacteria, whereas fewer than 1% of GAS deficient in NADase activity survive intracellularly for 24 hours [21]. We reasoned that anthrax toxin-mediated delivery of exogenous NADase might rescue intracellular GAS that did not produce enzymatically active NADase. We found that addition of 10 nM LFn-NADase to OKP7 cells in the presence of 20 nM PA increased the intracellular survival of GAS strain 188 G330D, which expresses enzymatically inactive NADase, by 14-fold, from 0.35% at 24 hours to 5% (Fig 5A). Thus, addition of exogenous NADase restored intracellular GAS survival to an extent roughly commensurate with the amount of NADase activity delivered to the cytosol of the host cell, i.e., approximately 50% of the activity associated with infection of the cell by the parent strain, 188. Addition of 1 nM LFn-NADase had a lesser and not statistically significant effect, increasing survival at 24 h of 188 G330D 2.5-fold to 0.9%. The ability of exogenously delivered NADase to restore intracellular GAS survival was dependent on the catalytic activity of the protein: addition of LFn-NADaseG330D had no effect on the 24-hour survival of GAS within OKP7 cells, even at 100 nM (Fig 5B).
The process of SLO-dependent translocation of NADase across the eukaryotic cell membrane requires a 190-aa domain in the amino terminus of the NADase protein, a part of the molecule that is dispensable for enzymatic activity [37]. As suggested by the protein synthesis inhibition assays (Fig 2C), we found that LFn-190NADase could function in lieu of LFn-NADase to increase the intracellular survival of 188 G330D, albeit slightly less efficiently than LFn-NADase (7-fold increase in survival versus 14-fold for LFn-NADase, Fig 5C). These results imply that the catalytic domain of NADase plays a dominant role in the intracellular survival of GAS. However, the small but reproducible improvement in survival imparted by LFn-NADase compared to LFn-190NADase suggests that the N-terminal translocation domain has an as-yet-unidentified function in enhancing intracellular survival.
The anthrax toxin delivery system allowed us to evaluate the contribution of NADase to GAS intracellular survival in the absence of SLO. Because SLO is ordinarily required to translocate NADase during GAS infection, it has not been possible previously to assess the role of NADase on GAS intracellular survival independently of SLO. To address the discrete contribution of each toxin, we added LFn-NADase and PA during infection of OKP7 cells with 188 SLO- and assessed the effect on intracellular survival (Fig 5D). Delivery of LFn-NADase increased intracellular survival of 188 SLO- by ~8 fold, from 0.25% to 2.0%. Thus, delivery of NADase prolongs the survival of both 188 G330D and 188 SLO- strains, results that imply both SLO and NADase are required for maximum resistance to intracellular killing. The fact that SLO-independent delivery of NADase partially corrects the survival defect of an SLO-deficient strain indicates that the reduced survival of SLO- GAS is due in part to the absence of NADase delivery, but also that SLO possesses NADase-independent activities that contribute to the intracellular survival of GAS. Thus, the synergistic action of SLO and NADase mediates optimal intracellular survival.
Previous studies have implicated SLO and NADase in GAS resistance to killing by epithelial cells. After internalization, SLO-deficient mutants are contained within endosomes or autophagosomes that fuse with lysosomes, an event associated with acidification of the GAS-containing vacuole and efficient bacterial killing [21,41]. The anthrax toxin system allowed us to assess directly the ability of NADase to interfere with fusion of the GAS-containing compartment with lysosomes. We found that delivery of exogenous LFn-NADase to the cytosol of GAS-infected OKP7 cells reduced the co-localization of 188 G330D with the lysosomal marker LAMP-1 (Lysosomal–Associated Membrane Protein 1) by 4-fold, from 41% to 10% at 6 h of infection (P<0.001, Fig 6). Delivery of LFn-NADase also inhibited trafficking of 188 SLO- to a LAMP-1-positive compartment, reducing co-localization with LAMP-1 from 86% to 51% (P<0.05). These findings correlate with the effects of LFn-NADase on the intracellular survival of 188 G330D and 188 SLO- and provide direct evidence that NADase contributes to GAS intracellular survival by interfering with lysosomal fusion to the GAS-containing vacuole.
Because the effect of LFn-NADase on endosomal trafficking was evident within the first few hours of infection, it seemed likely that the survival of intracellular GAS was largely determined during this time period. We found that a delay of only 2 hours in the addition of LFn-NADase to cells infected with 188 G330D largely abrogated the 24-hour survival benefit of LFn-NADase compared with that conferred by addition of the toxin at the time of initial infection (Fig 7). This result is consistent with the finding that, in the absence of NADase expression, GAS are trafficked to a degradative compartment by lysosomal fusion as early as 1 hour after infection (Fig 6).
A role for NADase in the virulence of GAS was suggested by the association of NADase production with M1T1 GAS isolates from invasive infections, beginning in the 1980s. Subsequent studies by Caparon and coworkers established a compelling model for SLO-dependent translocation of NADase into host cells, and intoxication of the cells was shown by our group to result in depletion of cellular NAD+ and ATP [16,20]. Experiments with NADase-deficient mutants supported a role for NADase in synergistic cytotoxicity with SLO, in induction of apoptosis, and in enhancing intracellular survival of GAS internalized by epithelial cells [17,38]. However, these functions of NADase during GAS infection have been inferred almost entirely from comparisons with mutants that lacked NADase or produced an enzymatically defective protein. The requirement of SLO for translocation of NADase has made it difficult to analyze the biological effects of NADase separately from those of SLO, which is required for NADase delivery, but which also has intrinsic cytotoxicity due to its pore-forming activity. An additional level of experimental complexity arises from the tightly bound endogenous inhibitor of NADase, IFS, whose co-expression is required for NADase production, but which must be removed to restore enzymatic activity.
In the current study, the anthrax toxin system provided a tractable platform to deliver enzymatically active, highly purified, IFS-free NADase or variant forms to the cytosol of human oropharyngeal keratinocytes. This system permitted direct investigation of the function of NADase in the cell biology of GAS infection, independent of the effects of SLO. We found that SLO-independent cytosolic delivery of LFn-NADase inhibited protein synthesis in oropharyngeal keratinocytes in a dose-dependent manner (Fig 2C). Nearly identical inhibition was observed upon delivery of LFn-190NADase, which lacks the N-terminal domain of NADase required for SLO-mediated translocation, but preserves the catalytic domain. By contrast, LFn-NADase G330D, an enzymatically inactive variant, had no inhibitory effect, even at high doses. Consistent with these results, sufficient doses of NADase delivered by the anthrax toxin system resulted in cytotoxicity and cell death that was dependent on the catalytic activity of the protein (Fig 3). These results support the view that the intrinsic cytotoxic activity of NADase on eukaryotic cells depends on the enzymatic activity of the toxin. Depletion of cellular NAD+ and ATP is expected to have a broad range of inhibitory effects on cellular functions. It remains possible that the synergistic toxicity of NADase with SLO also involves a second, non-enzymatic mechanism, as suggested by Chandrasekaran et al, although the molecular basis for such an effect has not been determined [38].
Previous studies found that SLO was required for prolonged GAS intracellular survival in keratinocytes [21,41]. Shortly after bacterial internalization, GAS production of SLO results in damage to the endosomal membrane, which exposes the bacteria to the cytosol where they become ubiquitinated. Ubiquitin is a signal for targeting intra-cytosolic bacteria to autophagosome-like compartments [21,42]. Fusion of lysosomes with these compartments leads to their maturation into degradative autolysosomes and efficient bacterial killing. Autophagosomes containing NADase-deficient GAS appear to follow this pathway; however, the step of lysosomal fusion is impaired for autophagosomes containing NADase-producing GAS, and this impairment is associated with enhanced intracellular survival [21]. The anthrax toxin system allowed us to study directly whether NADase prevents lysosomal fusion with GAS-containing vacuoles in infected cells. We found that cytosolic delivery of NADase inhibited the co-localization of GAS 188 G330D (expressing an enzymatically inactive NADase) with the lysosomal marker LAMP-1 (Fig 6). Inhibition of lysosomal fusion was associated with a 14-fold increase in intracellular survival to a level approaching that of the NADase-producing parent strain. Delivery of enzymatically inactive NADase G330D had no effect on GAS intracellular survival, supporting the essential role of enzymatic activity in enhancing intracellular survival. In similar experiments, we tested the effect of NADase delivery on the intracellular survival of the SLO-deficient GAS strain 188 SLO-. Supplying exogenous NADase partially rescued survival of 188 SLO-, a result that implies that SLO contributes to GAS intracellular survival in part through delivery of NADase, but also through function(s), such as pore-formation, independent of NADase translocation. These data are consistent with the observation that a GAS strain producing a non-pore-forming SLO that is competent for NADase translocation (SLO Y2552A) was defective for intracellular survival [21].
Results of these experiments provide the most direct evidence to date on the contribution of NADase to the cell biology of GAS infection. Use of the anthrax toxin delivery system isolated the effects of NADase from those of SLO and defined an unambiguous role for NADase in cytotoxicity for host epithelial cells and in enhancing GAS intracellular survival. Both functions were dependent on cytosolic delivery of NADase and on the enzymatic activity of the toxin to degrade NAD+. Together, these findings provide a plausible molecular basis for the association of NADase expression with GAS virulence.
The OKP7/bmi1/TERT (OKP7) keratinocytes used in this study are immortalized normal human soft palate keratinocytes [43,44]. These cells were a gift of James Rheinwald and were provided through the Harvard Skin Disease Research Center. OKP7 cells were cultured in keratinocyte serum-free medium (KSFM, Gibco/Invitrogen) as described previously [41].
GAS strain 188 and its mutant derivatives were used in this study. GAS strain 188 is an isogenic unencapsulated mutant of the M type 3 necrotizing fasciitis isolate 950771 [45]. Use of an unencapsulated mutant allowed efficient internalization of GAS by human cells in vitro because the hyaluronic acid capsule inhibits GAS internalization. Escherichia coli XL1-Blue was used as a host for molecular cloning (NEB) and was grown in Luria-Bertani (LB) medium (Novagen). GAS was grown in L3 medium as described with two modifications: the final CaCl2 concentration was 0.015% and type 1-S bovine hyaluronidase was omitted [46].
Generation of LFn-NADase-IFS constructs. The LFn-NADase-IFS-encoding construct was created by first PCR-amplifying separately the LFn-encoding sequence [47] and the nga-ifs genes from GAS genomic DNA. These amplicons were then used as templates for overlap PCR to generate the LFn-NADase-IFS-encoding DNA fragment, incorporating a BamHI restriction site inserted between the LFn-encoding sequence and nga-ifs. This product was subsequently cloned into pET43.1a vector (Invitrogen, Grand Island, NY) between the NdeI/XhoI restriction sites such that the in-frame fusion construct generated a His6-tag at the C-terminus of IFS.
Protein expression from the LFn-NADase-IFS-encoding construct, named MRW001, was insufficient for downstream studies. To improve expression, a DNA fragment encoding NADase G330D-IFS (enzymatically inactive NADase and IFS) was codon-optimized for expression in E. coli and synthesized by GENEWIZ, Inc (South Plainfield, NJ 07080). Codon-optimized nga-ifs was generated by OuikChange site-directed mutagenesis (Agilent Technologies) of the codon-optimized NADase G330D-IFS-encoding construct. These two constructs served as templates for PCR to generate DNA constructs encoding NADase, NADase G330D, 190NADase (aa 190–451), and 190NADase G330D using appropriate primers. Each of these PCR products was cloned between the BamHI/XhoI restriction sites in MRW001, in place of nga-ifs (S1 Fig).
Generation of NADase constructs. Codon-optimized DNA fragments encoding NADase, NADase G330D, and 190NADase were amplified by PCR using appropriate primers and cloned into the NdeI/XhoI restriction sites of pET43.1a to incorporate a C-terminal His6-tag on the IFS protein.
Generation of IFS and LFn constructs. In the first step, DNA fragments encoding an N-terminally His6-tagged Sumo protein, codon-optimized IFS, and LFn were amplified by PCR in separate reactions [47,48]. These PCR products served as template for overlap PCR to generate the His6-Sumo-IFS and His6-Sumo-LFn-encoding constructs, which were subsequently cloned into the NdeI/XhoI restriction sites in the pET43.1a vector.
Recombinant proteins used in this study are described in Table 1.
LFn-NADase, LFn-190NADase, NADase and their variants were expressed in BL21(DE3) cells (Invitrogen) using IPTG induction. Proteins were initially purified using Ni-charged metal affinity chromatography. Each partially purified protein preparation was loaded onto a High Performance Q column (GE) in buffer A (20 mM Tris, pH 7.5), washed with buffer A, and eluted with a gradient of 0 to 1 M NaCl in the same buffer. The proteins were then denatured in 6 M guanidinium chloride, pH 8.0, and the His-tagged IFS was removed from untagged LFn-NADase proteins by Ni-charged metal affinity chromatography. The IFS-free proteins were renatured by dialysis into buffer A containing 350 mM NaCl and 5 mM DTT. The renatured proteins were subsequently dialyzed in buffer A containing 5 mM DTT. Finally, the proteins were subjected to another round of Q column purification. Protein solutions were filter sterilized and stored at -80°C.
LFn and IFS fused with N-terminally His6-tagged Sumo protein were overexpressed using IPTG in BL21(DE3) cells (Invitrogen). The proteins were initially purified using Ni-charged metal affinity chromatography. Sumo was removed by cleavage with Sumo protease, and the reaction was monitored by SDS-PAGE. N-terminally His6-tagged Sumo and Sumo protease were removed from the now untagged protein of interest using Ni-charged metal affinity chromatography. DTT (5 mM) was added to the final protein eluate for NADase constructs.
Recombinant wild type PA and PA F427H were overexpressed in the periplasm of E. coli BL21 (DE3), purified by anion-exchange chromatography, and converted to the prepore form of PA using a protocol published elsewhere [50].
NADase and NADaseG330D were dialyzed against 10 mM sodium phosphate, 0.5 mM DTT, pH 8.0, and introduced at a concentration of 3.55 μM (determined by A280 measurements) into a stoppered 0.1 cm quartz cuvette. Equal concentration of the two proteins was confirmed by SDS-PAGE and Coomassie staining. CD spectra were measured in a JASCO J-815 Spectropolarimeter at 20°C from 185–260 nm in 0.5 nm steps with a 1 nm bandwidth. Five scans were averaged and smoothed, a background buffer-only spectrum was subtracted, and the data for the two protein species were plotted and overlayed to assess similarity.
Differential scanning fluorimetry was used to calculate the melting temperature of NADase, IFS, or NADase-IFS complex. A 10 μM solution of each protein was prepared in PBS containing 5X SYPRO Orange (Sigma), and the solution was dispensed in wells of a 96-well PCR plate. The plate was subjected to a temperature scan from 10 to 93°C at a rate of 1°C min−1 in an ABI Prism 7300 real time PCR instrument (Applied Biosystems/Invitrogen) using an excitation wavelength of 492 nm; fluorescence emission was recorded at 610 nm. Fluorescence emission of SYPRO Orange in aqueous solution increases upon binding to hydrophobic regions of proteins exposed by temperature-induced protein unfolding. The peak of the curve of the first derivative of the measured fluorescence intensity, plotted as a function of temperature, represents the melting temperature of the protein.
NADase activity of the recombinant proteins was determined as described (Bricker et al., 2002). Briefly, two-fold serial dilutions of NADase, LFn-NADase, or LFn-190NADase were incubated with 0.67 mM NAD+ for a period of 1 h at 37°C. The reaction was then terminated by the addition of 2 M NaOH and the fluorescence of uncleaved NAD+ was allowed to develop for 1 h, at which point the plates were read in a fluorimeter with excitation/emission wavelengths of 355nm/560nm. Samples without NADase served as controls. The results were expressed as fraction of total NAD+ that was cleaved at a given NADase concentration.
Thirty-five nM NADase was added to 17.5 nM, 35 nM and 70 nM of LFn-NADase G330D, LFn-190NADase G330D and NADase G330D in a 96-well plate. Seventy nM IFS, sufficient to completely inhibit enzymatic activity of 35 nM NADase, was then added to the wells. To this mixture, 0.67 mM NAD+ was added and the reaction incubated for a period of 1 h at 37°C. The reaction was then terminated by the addition of 2 M NaOH and the fluorescence of uncleaved NAD+ was allowed to develop for 1 h at which point the plates were read in a fluorimeter with excitation/emission wavelengths of 355nm/560nm. Samples without NADase served as controls. The results were expressed as percentage inhibition of NADase activity. Complete cleavage of NAD was labeled as 0% inhibition of NADase activity and no cleavage was labeled as 100% inhibition of NADase activity.
OKP7 cells were grown in 6-well dishes at 37°C in 5% CO2 to approximately 70% confluence (~2x105 cells/well). Cells were washed and incubated in KSFM containing GAS at a multiplicity of infection (MOI) of 10 unless otherwise indicated or supplemented with 20 nM PA and LFn-NADase at 10−8, 10−9, or 10−10 M for 2 h. A control lacking PA protein was also included. Fifteen min prior to harvesting cells, clindamycin (10 μg/ml) was added to prevent NADase production by GAS during sample processing. For intracellular NADase measurements, cells were washed, trypsinized, and permeabilized by incubation in PBS containing saponin (0.005% w/v) and protease inhibitors for 20 min at 37°C. Cells were removed by centrifugation for 2 min at maximum speed on a bench-top centrifuge and the supernatant containing cytosolic material was passed through a 0.2 μm filter. This filtrate, the cytosolic fraction, was kept on ice until NADase measurement. NADase activity was determined as previously described [17]. Experiments were performed three times. Intracellular activity was represented as the percentage NAD+ substrate depletion.
Planar phospholipid bilayer experiments were performed in a Warner Instruments Planar Lipid Bilayer Workstation (BC 525D, Hamden, CT). Planar bilayers were formed by painting a 35 mM solution of 1,2-diphytanoyl-sn-glycerol-3-phosphocholine (DPhPC) in n-decane (Avanti Polar Lipids, Alabaster, AL) on a 200 μm aperture of a Delrin cup in a Lucite chamber. The aperture separated two compartments, each containing one ml of 100 mM KCl, 1 mM ethylenediaminetetraacetic acid (EDTA), and 10 mM each of sodium oxalate, potassium phosphate, and 2-(N-morpholino)ethanesulfonic acid (MES), pH 5.5. Both compartments were stirred continuously.
Upon formation of a bilayer membrane, up to 5 μg PA prepore (25 pM) was added to the cis compartment in the presence of a constant voltage of +20 mV with respect to the trans compartment. After incorporation of PA pores as monitored by conductance across the membrane, the cis compartment was perfused to remove any free PA. Once the current had stabilized, 1 μg of LFn-NADase or a variant was added to the cis compartment, and interaction with PA channels was monitored by the decrease in conductance. After occlusion of PA pores had reached a steady state, excess LFn-NADase was removed by perfusion of the cis chamber. KOH was then added to the trans compartment to raise the pH of the buffer to 7.5. An increase in conductance indicated that the pH gradient between the cis and trans compartment had triggered the translocation of LFn-NADase across the PA pore into the trans compartment.
OKP7 cells were plated in a 96-well plate at a density of 104 cells/well approximately 40 h prior to the protein synthesis inhibition assay. PA (20 nM) and LFn-NADase diluted in KSFM were added to the plates. The plates were then incubated at 37°C for 24 h, after which toxin-containing medium was removed and was replaced with L-Leucine-deficient F-12 medium supplemented with L-[4, 5-3H] Leucine (Perkin Elmer). The plates were incubated for 1 h at 37°C. Next, the plates were washed with ice-cold PBS, liquid scintillation cocktail was added, and incorporation of radioactivity in the cells was measured in a scintillation counter. Results were normalized and expressed as a fraction of the radioactivity incorporated in OKP7 cells that were not treated with toxin.
OKP7 cells were infected at an MOI of 10 with GAS that had been grown to exponential phase (A600nm~0.25) and washed twice in KSFM. When appropriate, PA (20 nM) and LFn-NADase (1 nM or 0.1 nM) were added to the cells at the time of infection. Infected cell monolayers were treated with 20 μg/ml penicillin G and 200 μg/ml gentamicin for 45 minutes beginning 1 h 15 min post-infection. At 2 h post-infection, viable intracellular bacteria were quantified as described previously [51]. To determine intracellular survival at later time points, infected monolayers were washed at 2 h post-infection and fresh medium containing penicillin G (1 μg/ml), but not PA or LFn-NADase, was added. Infected monolayers were incubated for 4 h or 24 h post-infection, at which times the total intracellular CFU were determined as above.
OKP7 cells were cultured on coverslips in 24-well plates. Cells were infected with GAS as described above except that antibiotics were omitted to prevent cellular uptake of non-viable bacteria. Instead, extracellular bacteria were removed by extensively washing the cells with PBS at 2 h post infection, after which the cells were incubated in fresh KSFM. Infected cells were processed 1 h, 3 h, or 6 h post-infection. At each of these time points, monolayers were washed three times with PBS and extracellular GAS were stained with Alexa Fluor 660-conjugated anti-GAS IgG at 4°C for 15 min in the dark. Excess unbound antibody was removed by washing with PBS. Subsequently, cells were fixed and permeabilized by incubation in ice-cold methanol at −20°C for 5 min. Cells were then washed three times with PBS and incubated at room temperature for 1 h with mouse anti-LAMP-1 IgG. After three washes in PBS, cells were incubated for 1 h with goat anti-mouse Alexa Fluor 568-conjugated IgG and with Alexa Fluor 488-conjugated anti-GAS IgG at room temperature in the dark for 1 h. Slides were mounted using Prolong Gold (Molecular Probes) and stored at room temperature in the dark for 16–24 h prior to imaging. Confocal microscopy was performed at the Harvard Digestive Diseases Center core facility as previously described [51]. Images were acquired and analyzed using Slidebook 5 and Slidebook 6 (Intelligent Imaging Innovations, Denver, CO). For quantification, co-localization of intracellular bacteria with LAMP-1 marker was determined from three independent experiments. Images were evaluated by an observer who was blind to the experimental conditions. At least 100 intracellular bacteria were scored for each experiment.
OKP7 cells were cultured on coverslips in 24-well plates and grown to 40–50% confluence. Cells were then incubated with medium containing 20 nM PA and either 100 nM LFn-NADase or 100 nM LFn-NADase G330D for a period of 48 h. Cells that were not treated with any toxin served as a negative control. Cells were then washed with PBS and incubated with 500 μl of PBS containing 1 μg/ml propidium iodide for a period of 30 min at room temperature in the dark. Cells were then visualized under a Nikon Eclipse TS100 fluorescence microscope with a standard TRITC filter set (Ex 535/50, Em 610/75, DM 565) and images were acquired. For easy visualization, images showing dead cells stained with propidium iodide were colored red and merged with bright-field images showing the total number of cells (ImageJ software).
Significance of differences between experimental groups was assessed by Student’s t-test. P values of less than 0.05 were considered statistically significant.
|
10.1371/journal.pgen.1000550 | The Genomics of Speciation in Drosophila: Diversity, Divergence, and Introgression Estimated Using Low-Coverage Genome Sequencing | In nature, closely related species may hybridize while still retaining their distinctive identities. Chromosomal regions that experience reduced recombination in hybrids, such as within inversions, have been hypothesized to contribute to the maintenance of species integrity. Here, we examine genomic sequences from closely related fruit fly taxa of the Drosophila pseudoobscura subgroup to reconstruct their evolutionary histories and past patterns of genic exchange. Partial genomic assemblies were generated from two subspecies of Drosophila pseudoobscura (D. ps.) and an outgroup species, D. miranda. These new assemblies were compared to available assemblies of D. ps. pseudoobscura and D. persimilis, two species with overlapping ranges in western North America. Within inverted regions, nucleotide divergence among each pair of the three species is comparable, whereas divergence between D. ps. pseudoobscura and D. persimilis in non-inverted regions is much lower and closer to levels of intraspecific variation. Using molecular markers flanking each of the major chromosomal inversions, we identify strong crossover suppression in F1 hybrids extending over 2 megabase pairs (Mbp) beyond the inversion breakpoints. These regions of crossover suppression also exhibit the high nucleotide divergence associated with inverted regions. Finally, by comparison to a geographically isolated subspecies, D. ps. bogotana, our results suggest that autosomal gene exchange between the North American species, D. ps. pseudoobscura and D. persimilis, occurred since the split of the subspecies, likely within the last 200,000 years. We conclude that chromosomal rearrangements have been vital to the ongoing persistence of these species despite recent hybridization. Our study serves as a proof-of-principle on how whole genome sequencing can be applied to formulate and test hypotheses about species formation in lesser-known non-model systems.
| The transformation of populations into distinct species depends on whether hybridization, recombination, and subsequent gene introgression can be suppressed between diverging species. We use partial genome sequences to reconstruct this evolutionary process in the Drosophila pseudoobscura species subgroup, which includes the hybridizing species pair D. pseudoobscura pseudoobscura and D. persimilis. Recent models suggest that chromosomal inversions can facilitate the persistence of hybridizing species because of their effects on recombination, whereby inverted regions would exhibit higher nucleotide divergence than non-inverted regions. Indeed, D. pseudoobscura-D. persimilis nucleotide divergence outside these inverted regions is lower than within or near inversions, resembling D. ps. pseudoobscura levels of within-species nucleotide diversity. We also observe that recombination suppression in F1 hybrids extends greater than 2 Mbp outside the inversion breakpoints. Furthermore, when genomic sequence of D. persimilis is compared to two sister subspecies—the hybridizing subspecies, D. ps. pseudoobscura, and a non-hybridizing control subspecies, D. ps. bogotana—autosomal divergence is lower in the former, demonstrating recent gene exchange. These lines of evidence support a speciation model in which the two hybridizing species persist despite the presence of recent genic introgression in collinear regions of the genome because of the reduced recombinational effects of the inversions that distinguish them.
| One of the most significant empirical insights in evolutionary biology is that a large number of species naturally hybridize with close relatives [see reviews in 1],[2], and these species pairs often exchange genetic material (“introgression”). These insights contrast previous assertions that considered interspecies hybridization as rare or anomalous [e.g., 3]. However, recognizing that hybridization and introgression are frequent suggests that genetic features may allow such species to remain distinct.
Several recent studies suggested that genomic regions of low recombination may provide a means to create “islands of differentiation” between species [4]–[10]. While introgression may homogenize parts of genomes, regions of low recombination in hybrids, such as those within chromosomal inversions, maintain their distinction despite the influx of foreign alleles in collinear regions. This pattern of localized differentiation should be particularly strong if regions of low recombination also harbor loci with divergently selected alleles or alleles conferring reproductive isolation. However, this hypothesis has been difficult to test rigorously on a genome-wide scale.
Assembled whole genome sequences and inexpensive resequencing technologies can complement locus-specific sequencing studies and genetic mapping studies for clarifying the role of regions of low recombination in species persistence. Fruit flies from the genus Drosophila have been a major focus of many studies of nucleotide divergence between closely related species and mapping studies of traits that prevent introgression such as hybrid sterility. The recent sequencing of multiple closely related Drosophila species [11],[12] places us in a stronger position to test hypotheses concerning gene flow and speciation in this model system.
The Drosophila pseudoobscura species subgroup is comprised of two D. pseudoobscura subspecies (D. ps. pseudoobscura and D. ps. bogotana), and two closely related species, D. persimilis and D. miranda. The D. pseudoobscura subspecies are geographically isolated (D. ps. pseudoobscura ranges across the western half of North America and D. ps. bogotana is restricted to Colombia in South America), share chromosomal arrangements, and represent the earliest stages of species divergence [13]. D. persimilis and D. miranda are restricted to the west coast of North America, where they co-occur with D. ps. pseudoobscura. Both D. pseudoobscura subspecies differ from the close relative species D. persimilis by fixed (or nearly fixed) chromosomal inversion differences on three of their major chromosome arms, and F1 hybrid males from crosses between these species are sterile (though females are fertile). In contrast, D. miranda is an outgroup species which cannot produce any fertile hybrids with D. pseudoobscura or D. persimilis [14]. The relative relationships of these species as (((D. ps. pseudoobscura-D. ps. bogotana) D. persimilis) D. miranda) is well established by DNA sequences, chromosomal inversions, and reproductive isolation [15],[16]. Overall, this system provides us with a pair of taxa that hybridize and have experienced introgression (D. ps. pseudoobscura and D. persimilis [17]–[19]), and two taxa that have not experienced recent introgression from any close relatives (ingroup, D. ps. bogotana, and outgroup, D. miranda).
Nucleotide divergence between the hybridizing species D. ps. pseudoobscura and D. persimilis is high within and near the three chromosomal inversions [20],[21], which are linked to factors conferring hybrid sterility, mating discrimination, and other barriers to gene flow [22],[23]. Based on these observations, we hypothesized that inversions facilitate the distinction of these species despite ongoing natural hybridization. However, it has been difficult to fully disentangle complications that result from ancestral polymorphisms shared between these species [21],[24] and underlying assumptions found in many statistical tests for introgression [25]–[27]. Two recent studies also reached differing conclusions about whether gene exchange between these species occurred during the initial divergence process or later [20],[21].
To better understand the genealogical history of this subgroup, we use published genome sequence assemblies of D. ps. pseudoobscura and D. persimilis [12],[28], along with three novel partial genomic sequences that we generated using 454/Roche technology (Table S1): one from D. miranda, one from a second strain of the North American subspecies, D. ps. pseudoobscura, and one from the South American subspecies, D. ps. bogotana. By providing controls for divergence in the absence of gene flow, these newly obtained genomic sequences allow for more robust analysis than previous studies. Our new results suggest that the chromosomal regions inverted between D. persimilis and D. ps. pseudoobscura arose in allopatry, and that D. pseudoobscura, D. persimilis, and D. miranda all diverged within a relatively short time frame. We also find compelling evidence for autosomal gene exchange between D. persimilis and D. ps. pseudoobscura in collinear regions since the split of D. ps. pseudoobscura and D. ps. bogotana, likely within the past 200,000 years. Overall, our analyses utilize genome sequence data in an existing framework to demonstrate the importance of chromosomal inversions in maintaining the persistence of hybridizing species and to consolidate previous tentative conclusions about divergence in this group. Further, this research serves as a model for how whole genome shotgun sequence data can be used with a reference genome sequence to address fundamental questions regarding evolutionary changes leading to the formation of species.
Figure 1 presents sliding window estimates of polymorphism within the North American subspecies, D. ps. pseudoobscura, and divergence between D. ps. pseudoobscura and each of D. persimilis and D. miranda across four of the five major chromosome arms. Each datapoint within the sliding window represents the fraction of bases differentiating two genome sequences along a 500 kilobase pairs (kbp) interval, iterated every 100 kbp. Very similar plots were generated for intergenic regions or introns alone (not shown). We only scored positions for which aligned sequences were available for all four taxa (Table S2), hence eliminating the possibility that a particular region of high or low divergence would be represented in some estimates but not others. Nucleotide polymorphism estimated within D. ps. pseudoobscura was confirmed to be in the same range as that observed in polymorphism studies of focal genomic regions of this species [18] (see Table S3).
Fixed inversions on chromosomes XL, XR, and 2 distinguish D. ps. pseudoobscura and D. persimilis, and their breakpoints are superimposed on Figure 1. Corroborating previous work, nucleotide diversity within D. ps. pseudoobscura and divergence between D. ps. pseudoobscura and D. persimilis are low in regions near the centromere [21],[29]. The latter observation was previously interpreted “as reflective of ancestral patterns of polymorphism rather than the process of divergence between these species” [21]. Consistent with this interpretation, we observe that diversity within D. ps. pseudoobscura and divergence between D. ps. pseudoobscura and D. persimilis were correlated on every chromosome arm (r = 0.418–0.535, P<0.01 for each) [see also 24].
The species pair, D. ps. pseudoobscura and D. miranda, exhibit a different pattern. There was no consistent decline in divergence between these two species in regions near the centromeres. Furthermore, diversity within D. ps. pseudoobscura was not significantly correlated with divergence to D. miranda along any chromosome arm except chromosome 4 (r = 0.330, P = 0.018), suggesting that D. miranda and D. ps. pseudoobscura are not sharing many polymorphisms. According to a model where D. miranda is the outgroup, we predict that the range (maximum minus minimum) of divergences across windows should be greater for the purportedly more divergent species pair, D. miranda - D. ps. pseudoobscura, than the pair of more recently diverged species, D. persimilis - D. ps. pseudoobscura. While this prediction was met for windows along the collinear chromosome 4, we observed instead a greater range of divergences in the D. persimilis - D. ps. pseudoobscura pairing on the chromosome arms (XL, XR, and 2) that harbor inversions distinguishing these species (see also Table S4, Figure S1). These observations are inconsistent with a more recent divergence of this latter species pair, and are more consistent with the presence of multiple genealogical histories along the genome.
Inversions prevent gene exchange because the products of recombination are not recovered. We confirmed that recombinant products are not recovered within 2.1 megabase pairs (Mbp) of fixed inversions along chromosome XL, XR, and 2 in heterozygotes (D. ps. pseudoobscura - D. persimilis interspecies hybrids). We recovered 0.25%–0.55% recombinants at markers 2.8 Mbp outside of each inversion, indicating that complete recombination suppression extends greater than 2.1 Mbp, but not more than 2.8 Mbp outside inversions. Strong crossover suppression, resulting in less than one percent recombinants, is observed relative to one marker 3.35 Mbp outside of the XR chromosomal inversion. Crossing over is largely restored at 4.55 Mbp outside inversions, with a crossover rate greater than 5% observed from one marker on chromosome 2 (see Table S5).
The lack of recombination and introgression should produce a distinct signature in nucleotide divergence within and near chromosomal inversions. We found that, along the three chromosome arms bearing inversions, nucleotide diversity within D. ps. pseudoobscura was comparable to D. persimilis nucleotide divergence when estimated on sequence greater than 2.5 Mbp outside the inverted regions. In contrast, divergence between D. persimilis and D. ps. pseudoobscura was comparable to divergence between D. miranda and D. pseudoobscura in regions inside and within 2.5 Mbp flanking the inversions. The consistency of this pattern across independent chromosomal arms suggests either that all three inversions arose at approximately the same time as the split from the ancestor of D. miranda, or that the ancestral populations of these species were already separated (i.e., allopatric) when the inversions arose (see Discussion).
Recent gene flow is not expected between the South American D. ps. bogotana and either of the North American taxa D. ps. pseudoobscura or D. persimilis. Analyses of nucleotide sequence data suggests that the D. pseudoobscura subspecies diverged from a common ancestor 200,000 years ago [19],[30]. As such, we can use the isolated subspecies as a “negative control” to test for recent introgression between North American D. ps. pseudoobscura and D. persimilis.
Because of hybridization between the North American taxa, a very simple expectation is that D. persimilis (Dper) should be more similar in sequence to North American (NA) than South American (SA) subspecies of D. pseudoobscura (Figure 2). We limited the dataset to sites where we have 454/Roche sequence reads for both D. pseudoobscura subspecies, and tested this hypothesis using regions far from the inversion on chromosome 2 and all along collinear chromosome 4. Aligned bases were categorized as [Dper = NA≠SA] or [Dper = SA≠NA]. No two bases were scored that were within 500 bp of each other, hence reducing artifacts from non-recombining haplotype blocks. We observed an excess of the first category (7073 vs. 6797, Binomial Sign Test P = 0.0096), indicating that divergence is lower between D. persimilis and North American D. ps. pseudoobscura than between D. persimilis and South American D. ps. bogotana.
The above test does not account for possible faster divergence within the South American subspecies lineage, either through increased mutation rate or more frequent fixation of slightly deleterious alleles. Testing for differences in lineage rates, we did not observe greater divergence between the South American subspecies and the non-hybridizing species, D. miranda (Dmir), than between the North American subspecies and D. miranda (Dxy = 0.019 for both, P = 0.221). Nonetheless, we can test for recent gene exchange more rigorously by specifically counting “shared-derived” base pair substitutions polarized with D. miranda. Counts of [Dmir = SA≠Dper = NA] were compared to counts of individual base pairs in which [Dmir = NA≠Dper = SA], where the latter half of the inequality denotes potential shared-derived bases. Again, we observe a slight, borderline significant excess of the first category (219 vs. 185, Binomial Sign test P = 0.05), suggesting that D. persimilis and North American D. ps. pseudoobscura share more derived bases.
Finally, introgression between species is not expected to be homogeneous outside inverted regions. The Alcohol dehydrogenase (Adh) region has been reported to have introgressed recently between these species using analyses independent of divergence from the South American species, D. ps. bogotana [19]. Further, it can be introgressed in the laboratory and made homozygous in a foreign genetic background with no deleterious effects [23]. We examined base pair counts of [Dmir = SA≠Dper = NA] vs. [Dmir = NA≠Dper = SA] for this region. In this region which bears Adh (chromosome 4 “group1”, extending 4 Mbp starting at position 14.4 million in Figure 1), we again observed a significant and dramatic excess of the first category (27 vs. 10, P = 0.00382).
We applied the same analyses to test for recent gene exchange along X-linked regions from both XL and XR distant from the inversion breakpoints. We observed a nonsignificant difference in number of bases categorized as [Dper = NA≠SA] vs. [Dper = SA≠NA] on this chromosome (1200 vs. 1131, Binomial Sign Test P = 0.079). When we polarized the bases and compared (Dmir = SA≠Dper = NA) vs. (Dmir = NA≠Dper = SA), we observed a nonsignificant difference opposite in direction to our expectation (46 vs. 62). However, there was only 27% as much sequence to analyze more than 2.5 Mbp from inversions on the X-chromosome than on the autosomes.
Genome sequencing has recently become affordable for individual investigators, but how the resultant data can be applied to address evolutionary questions about species formation or diversification has been less clear. Here, we use partial genome sequence data to: 1) evaluate the role of chromosomal inversions in maintaining the distinction between two hybridizing Drosophila species, 2) estimate when gene exchange occurred between these species, and 3) clarify contradictory interpretations from earlier studies that attempted to address related questions. Overall, this research demonstrates how present-day patterns within genomic data can help to infer past processes involved in speciation.
If two species share extensive polymorphism through introgression or incomplete lineage sorting resulting from a recent split, we predict that nucleotide sequence diversity within species should be correlated with average pairwise nucleotide differences between species. Extensive polymorphism sharing was shown previously in the case of D. ps. pseudoobscura and D. persimilis [18],[20],[21],[31]. In contrast, we find that nucleotide sequence differences between D. ps. pseudoobscura and D. miranda were uncorrelated with nucleotide sequence differences between two strains of D. ps. pseudoobscura. This finding suggests that our comparisons to D. miranda are not hindered by introgression or extensive shared ancestral polymorphisms [but see 32].
Previous DNA sequence-based studies observed that D. ps. pseudoobscura and D. persimilis share variation far outside the fixed inversions that distinguish these species [20]. Here, we note that divergence between D. ps. pseudoobscura and D. persimilis is higher and more comparable to differences between two strains of D. ps. pseudoobscura in regions distant from the inversions. In contrast, divergence between D. ps. pseudoobscura and D. persimilis is comparable to that between D. ps. pseudoobscura and the non-hybridizing outgroup, D. miranda, for regions inside and just outside the chromosomal inversions that separate them.
These new results can be used to formulate a hypothesis for the evolutionary history of these species and reconcile previously contradictory inferences. Machado et al [20] suggested that D. ps. pseudoobscura and D. persimilis largely speciated in allopatry, close in time to the split of these species from D. miranda, and recent secondary contact between the first two resulted in the dissolution of differences outside the inverted regions. In contrast, Noor et al [21] noted that significant differences in divergence among the XL, XR, and 2-chromosome inverted regions suggest instead that D. ps. pseudoobscura and D. persimilis speciated under a sympatric “divergence-with-gene-flow” model. In other words, differences in divergence between inversions reveal when each inversion arose as both species evolved in sympatry.
Our study recapitulates both sets of results and allows us to suggest a resolution. As in Noor et al [21], we observe that the XL chromosome arm inversion was most different in sequence between D. ps. pseudoobscura and D. persimilis, followed by chromosome 2 and finally chromosome arm XR (see Figure 1). However, we also observe that divergence between D. pseudoobscura and D. persimilis within each inverted region was similar to the divergence between D. ps. pseudoobscura and D. miranda (shown at single loci by [20]). Because we observe the same XL>2>XR ranking in D. ps. pseudoobscura divergence from D. miranda that was shown previously for divergence from D. persimilis [21], we now interpret this variation among chromosomes as reflective of differences in mutational processes rather than differences in time since separation. Our new, combined observations suggest two possible interpretations. First, the three inversions independently may have arisen very close in time (near the time of the split from D. miranda) from the D. pseudoobscura-D. persimilis ancestor, and these three derived forms segregated exclusively in D. persimilis. Alternatively, and arguably more parsimoniously, the three species diverged close in time, D. persimilis acquired three new inversions sometime after the split from D. pseudoobscura, and secondary contact between D. persimilis and D. pseudoobscura homogenized the noninverted regions.
Many recent studies have analyzed DNA sequence polymorphism and divergence to identify the statistical signature of recent introgression. However, these tests did not typically identify a time frame within which introgression occurred except as variance in the time of divergence [25],[26]. Instead, most tests merely reject or fail to reject a model of divergence in total isolation.
Here, we use a comparison between subspecies to infer the timing of introgression between D. persimilis and D. ps. pseudoobscura. One D. pseudoobscura subspecies co-occurs and hybridizes with D. persimilis while the other subspecies lives isolated on a different continent. Hence, we can attribute differences in divergence between D. persimilis and these D. pseudoobscura subspecies to hybridization that has occurred more recently than the split of the subspecies, estimated to have been 200,000 years ago [19],[30]. We observe a slight but statistically significant difference in divergence across uninverted (collinear) autosomal regions between D. persimilis and the two D. pseudoobscura subspecies, suggesting recent introgression between the co-occurring taxa, but we fail to detect such evidence for introgression across comparable regions of the X-chromosome.
Although we detected a statistically significant signature of introgression along autosomal loci, the signature was faint, suggesting that recent gene exchange has not been extensive. DNA sequence-based studies previously identified the statistical signature of historical introgression [18],[20],[31], but these studies interpreted this gene exchange as ancient based on the lack of longer shared haplotypes [33]. Similarly, an allozyme-based meta-analysis failed to detect differences between D. ps. pseudoobscura populations co-occurring with D. persimilis compared to those elsewhere in North America [34], suggesting a lack of extensive recent introgression. Given the high levels of gene exchange among populations within D. ps. pseudoobscura, the approach used by Kulathinal and Singh [24] does not have enough resolution to detect the low levels of gene flux we infer here. Again, our sparse genomic sequence data helps to refine these earlier results.
A significant difference between sex-linked and autosomal loci in introgression has been a recurring theme in divergence population genetics [e.g.], [ 35], [36]–[38]. However, in most systems, we lack knowledge of the karyotype (e.g., inversion differences) or other factors which may make the sex chromosomes and particular autosomes inappropriate for comparison. In D. ps. pseudoobscura and D. persimilis, however, we observe evidence for introgression on the autosomes while not on the X-chromosome in regions outside the inversions, suggesting that these differences may be reflective of sex-linkage per se. This observation may be consistent with a higher density of factors conferring hybrid sterility or other barriers to gene flow on the X-chromosome than on the autosomes [e.g., 39].
In this study, we used sparse whole-genome shotgun sequences from multiple taxa to infer the evolutionary history of a species group and to identify genomic features associated with their divergence. Our system was well-leveraged in that we initially began the investigation already having an assembled and annotated full-genome sequence for two of the focal species [12],[28] as well as genetic mapping data localizing factors that reduced potential gene exchange [22],[23].
Nonetheless, the cost of next-generation sequencing is dropping for both model and non-model systems, even between the execution of this study and its publication. Because of cost constraints, our study approached these questions using light resequencing (effectively utilizing the power of millions of markers) but producing extensive gaps and a majority of aligned positions being covered by single sequence traces. However, our approach serves as a proof-of-principle for future genomic studies on lesser developed systems. We attempted to reduce systematic biases by applying stringent filters, specific tests (including averaging across 500 kbp windows) and by employing the use of a well-assembled reference genome sequence. Future, more rigorous approaches enabled by less-expensive sequencing technologies will allow researchers greater power to infer historical evolutionary processes such as speciation and historical introgression in non-model systems.
In this comparative study, a total of five genomes representing four species of the obscura subgroup were sampled. Adult females from inbred lines of D. miranda (from Mather, California; San Diego stock #14011-0101.08) and the subspecies, D. ps. bogotana (from El Recreo, Colombia; San Diego stock #14011-0121.152) were each extracted and purified using the Gentra PureGene DNA isolation kit. For D. miranda, genomic DNA was nebulized and single stranded libraries generated before being sequenced at light coverage on a single Roche/454 Life Sciences GS-FLX run at Duke University's IGSP core sequencing facility, yielding approximately 100 Mbp of sequence (see Table S1). D. ps. bogotana genomic DNA was similarly sequenced in one half of one run at Duke University's IGSP core sequencing facility and one half of one run at 454 Life Sciences. These genome sequence traces were submitted to the NCBI Short Read Archive (SRA) as accession SRA008268. Additionally, two previously sequenced and assembled genomes, D. ps. pseudoobscura (Release 2) and D. persimilis (Release 1), were used for comparative analysis [12],[28]. Finally, to estimate nucleotide diversity within D. ps. pseudoobscura, previously sequenced Roche/454 reads (NCBI SRA accession SRA000268) from a second line (from Flagstaff, Arizona; San Diego stock number 14011-0121.151; [24]) were reassembled syntenically to D. ps. pseudoobscura.
All Roche/454 reads were syntenically aligned against reference D. ps. pseudoobscura (Release 2) linkage groups. Individual base calls were filtered to exclude nucleotides that are: within 3 base pairs of an alignment gap, harbor low quality scores (below 10), contain greater than 30% mismatches within a 7 base pair window, are in regions of high divergence (divergence to D. persimilis is greater than 30% in a 7 base pair window). Alignments from the two previously sequenced reference genomes, Drosophila ps. pseudoobscura and D. persimilis were obtained via chain files from the UCSC Genome Browser (genome.ucsc.edu). Site-specific annotation information such as intron and codon position was extracted from D. ps. pseudoobscura Release 2.3 annotations from FlyBase (flybase.org). Chromosome arms (including ordered contigs) 2, 4, XL, and XR were used (see [40] for contig details), representing roughly 80% of the total genome. We did not survey chromosome 3 because of complications from its inversion polymorphism within each of these species [41].
Chromosome arms XL, XR, and 2 differ by single inversions between D. pseudoobscura and D. persimilis, and the breakpoints of these inversions have been mapped [21],[42]. Using microsatellite markers that flank the sides of each inversion, we surveyed the extent of recombination in F1 hybrids between D. ps. pseudoobscura and D. persimilis. The published genome lines of both species (San Diego stock numbers #14011-0121.94 and #14011-0111.49) were used in this cross and recombinants were screened among 384 progeny of F1 females backcrossed to D. pseudoobscura. The following markers were used to assay recombination rate at varying distances from the inversions – chromosome 2 inversion: DPS2019 (2.77 Mbp from inversion on telomeric side), DPS2026 (associated with inversion) and DPS2031 (2.8 Mbp from inversion on centromeric side), XL inversion: DPSX_7446z (2.84 Mbp from inversion on centromeric side), DPSX046 (associated with inversion), DPSX008 (0.4 Mbp from inversion on telomeric side), and DPSXL_3a_0.8 (2.8 Mbp from inversion on telomeric side), XR inversion: DPSXR_6_2.7 (3.35 Mbp from inversion on centromeric side), DPSX063 (associated with inversion), DPSX037nA3 (1.4 Mbp from inversion on telomeric side), DPSX037N (2.1 Mbp from inversion on telomeric side), and DPSX058 (2.8 Mbp from inversion on telomeric side). Primer sequences are available upon request.
|
10.1371/journal.ppat.1005727 | Adapting SHIVs In Vivo Selects for Envelope-Mediated Interferon-α Resistance | Lentiviruses are able to establish persistent infection in their respective hosts despite a potent type-I interferon (IFN-I) response following transmission. A number of IFN-I-induced host factors that are able to inhibit lentiviral replication in vitro have been identified, and these studies suggest a role for IFN-induced factors as barriers to cross-species transmission. However, the ability of these factors to inhibit viral replication in vivo has not been well characterized, nor have the viral determinants that contribute to evasion or antagonism of the host IFN-I response. In this study, we hypothesized that the host IFN-I response serves as a strong selective pressure in the context of SIV/HIV chimeric virus (SHIV) infection of macaques and sought to identify the viral determinants that contribute to IFN-I resistance. We assessed the ability of SHIVs encoding HIV-1 sequences adapted by serial passage in macaques versus SHIVs encoding HIV sequences isolated directly from infected individuals to replicate in the presence of IFNα in macaque lymphocytes. We demonstrate that passage in macaques selects for IFNα resistant viruses that have higher replication kinetics and increased envelope content. SHIVs that encode HIV-1 sequences derived directly from infected humans were sensitive to IFNα –induced inhibition whereas SHIVs obtained after passage in macaques were not. This evolutionary process was directly observed in viruses that were serially passaged during the first few months of infection–a time when the IFNα response is high. Differences in IFNα sensitivity mapped to HIV-1 envelope and were associated with increased envelope levels despite similar mRNA expression, suggesting a post-transcriptional mechanism. These studies highlight critical differences in IFNα sensitivity between HIV-1 sequences in infected people and those used in SHIV models.
| The innate immune system is an important host defense against viral infection. Recently, there has been significant interest in characterizing the innate immune response to HIV-1 infection, in particular the role of type-I interferon (IFN-I). Understanding the interaction of HIV-1 with the innate immune system is particularly important for the development of animal models of infection as innate host factors present potential species-specific barriers to the establishment of persistent infection. One of the most commonly used animal models of HIV-1 infection is chimeric SIV/HIV (SHIV) infection of macaques. Here, we demonstrate that the process of adapting SHIVs for replication in macaques selects for viruses that are resistant to the IFNα response, and we identity important viral determinants that contribute to this resistance. This improved understanding of virus interactions with the innate immune response may facilitate the development of improved animal models of HIV-1 infection.
| The innate immune system presents the first host defense against viral infection. Host cells are able to sense the presence of viral infection and respond by producing type-I interferon (IFN-I), which, in turn, leads to the up-regulation of hundreds of host genes that are potentially antiviral [1,2]. Infection with HIV-1 in people and SIV in non-human primates induces a robust IFN-I response within days of infection [3–7]. IFN-I production, including IFNα, is part of a larger systemic cytokine storm that precedes the establishment of set-point viral load suggesting that the IFN-I response may play a role in limiting viral replication during acute infection and influence disease progression [8]. In support of this hypothesis, a recent study of SIV infection in rhesus macaques demonstrated that blocking the IFN-I response resulted in higher plasma viral loads during acute infection, increased reservoir size and faster progression to AIDS [9].
Despite the presence of a robust antiviral IFN-I response to infection, lentiviruses are able to replicate to high levels during acute infection and establish persistent infection in their hosts. Some recent studies have provided evidence that the innate immune response selects for HIV-1 variants that are relatively resistant to IFN-I during transmission [10,11]. The biological properties that contribute to the ability of some HIV-1 variants to resist the IFN-I response remain unclear.
One possible explanation for differences in IFN-I sensitivity of HIV-1 variants is that they have different abilities to evade or antagonize downstream effectors of the IFN-I response. Over the last decade, host antiviral proteins, referred to as restriction factors, have been identified that act at multiple stages of the lentiviral life cycle and directly inhibit viral replication [8,12]. Many of the restriction factors are induced by IFN-I [8,12]. Because the IFN-I-induced factors are effective at inhibiting viral replication, lentiviruses have evolved mechanisms to evade or antagonize their activity. Indeed, the human orthologs of the IFN-I-induced restriction factors that inhibit HIV-1 replication are largely inactive against HIV-1 because of the specificity of the viral antagonist for the human protein. The mechanisms of restriction factor inhibition and viral antagonism and the importance of these interactions for establishing productive infection in vitro have been carefully elucidated. However, the role of the IFN-I response in limiting viral replication and mechanisms of viral evasion/antagonism in the context of infection in vivo is less clear. Relevant to this, HIV-specific restriction factors have been largely studied for their ability to inhibit HIV-1 variants derived after passage in cell culture and less is known about the IFN-I-induced responses that inhibit viruses replicating in infected individuals.
Due to the selective pressure of restriction factors, lentiviral proteins are adapted to antagonize factors in their respective hosts and often act in a species-specific manner [8,13]. For example, HIV-1 proteins are able to antagonize human restriction factors but are unable to effectively counteract the non-human primate orthologs. For this reason, SIV/HIV chimeric viruses (SHIVs) used to study HIV-1 pathogenesis in macaques encode SIV antagonists of well-characterized macaque restriction factors. The HIV-1 genes encoded in SHIVs typically include the env gene that encodes the Envelope surface glycoprotein (Env). In most cases, SHIVs require multiple rounds of adaptation in lab-culture and/or by animal-animal serial passage in macaques in order to increase replication capacity and pathogenicity [14]. Often the process of animal-animal serial passage is performed within the first two weeks of infection when levels of IFN-I are highest in the animals [4,5,15–17], providing a possible selective pressure to drive changes in the virus. Thus, SHIVs that have been adapted in macaques present a unique opportunity to study the mechanism of adaptation to IFN-I response.
The goals of this study were to determine whether the process of adapting SHIVs for increased replication capacity and pathogenicity in macaques selects for variants that are resistant to the host’s IFN-I response and to identify the biological changes in the virus that contribute to IFN-I resistance. Given the fact that the majority of SHIVs studied to date encode HIV-1 variants derived from cell culture and represent the select group of SHIVs that were able to infect macaques, we also asked whether there are differences in IFN-I sensitivity of these viruses compared to SHIVs encoding HIV-1 sequences isolated directly from infected individuals. We demonstrate that envelope differences selected in vivo allow SHIVs to adapt to the IFNα response; adapted HIV-1 variants encode IFNα resistant Envs, whereas Envs obtained directly from infected individuals early in their infection are sensitive, suggesting that IFNα may have an inhibitory effect on viruses spreading in humans that has not been observed through the study of adapted viruses. These findings suggest that Env plays an important role in evading or antagonizing the IFNα response. Identification of IFNα resistant HIV-1 Env variants may facilitate the development of challenge viruses for macaque models of HIV-1 infection.
In order to test the hypothesis that adaptation of SHIVs results in IFN-I resistance, we compared a panel of nine SHIVs for their ability to replicate in macaque cells in the presence or absence of IFNα. The panel of SHIVs included four viruses that encode HIV-1 sequences isolated directly from infected individuals (MG505, Q23, QF495, and BG505) with the latter three from early infection [18–20], two viruses that encode HIV-1 sequences obtained from individuals during late-stage chronic infection and adapted in lab-culture in human cells (AD8 and SF162) [21,22] and three viruses that encode HIV-1 sequences adapted by animal-animal passage in macaques (AD8-EO, SF162P3 and 1157ipd3N4), two of which represent the animal-passaged derivatives of the lab-adapted viruses [16,21,23] (S1 Table). Most HIV-1 variants circulating in people are unable to use the macaque CD4 receptor for entry into cells [24], therefore, SHIVs that were made from HIV-1 variants isolated directly from individuals encode single amino acid changes that allow them to use the macaque CD4 receptor for entry. HIV-1 variants encoding these changes are able to use the macaque CD4 for entry at levels similar to those of adapted SHIVs [24], and the viruses chosen for study represent those that were infectious in macaque CD4+ T cells. Otherwise, the HIV-1 sequences of these SHIVs are unmodified from the sequences found in the infected individual and will be referred to as circulating SHIVs as they are representative of HIV-1 variants circulating in human populations.
We assessed the ability of the panel of SHIVs to replicate in immortalized pig-tailed macaque (Ptm) CD4+ lymphocytes [25] in the presence and absence of IFNα-2a at a concentration similar to that observed in natural infection (1000 U/ml) [4,5] (Fig 1A, S1 Fig). Intracellular staining for two IFNα-stimulated proteins, MX1 and IFIT1, showed that nearly all immortalized Ptm lymphocytes responded to IFNα treatment (S2 Fig). IFNα sensitivity was measured as the ratio of the area-under curve (AUC) of the replication curve in the IFN-treated cells to the AUC of the replication curve in the untreated cells. For example, SHIV AD8-EO, a pathogenic, macaque-passaged virus, replicated in the presence of IFNα nearly as well as the untreated cells (Fig 1A) and had an AUC ratio (IFN+/IFN-) of 0.96. In contrast, SHIV Q23AE, a circulating SHIV, exhibited a pronounced IFNα-induced inhibition of viral replication corresponding to an ~100-fold reduction in SIV p27 levels at nine dpi and had an AUC ratio of 0.68 (Fig 1A). The other seven viruses exhibited a range of inhibition between these two ends of the spectrum (S1 Fig). We observed the same patterns of IFNα-induced inhibition when we pre-treated the cells with IFNα-2a at 24 hours prior to infection (S3 Fig). Overall, SHIVs adapted by macaque-passage and by lab-culture exhibited higher AUC ratios than circulating SHIVs indicating resistance to IFNα treatment (Fig 1B). Comparing AUC ratios, macaque-passaged SHIVs were significantly more resistant to IFNα treatment compared to circulating SHIVs (0.94 vs. 0.78, p = 0.05).
The replication kinetics of the nine SHIVs were defined using the data from the replication time course studies where replication differences between viruses were evident even in the absence of IFNα. For example, SHIV AD8-EO demonstrated rapid replication kinetics in untreated cells reaching peak virus levels of >106 pg/ml of SIV p27 by six dpi (Fig 1A). In contrast, SHIV Q23AE reached lower peak virus levels of >105 pg/ml of SIV p27 at nine dpi. In order to compare replication kinetics across the panel of nine SHIVs, we determined a summary measure of viral replication based on the slope of a best-fit straight line of the replication curve in untreated immortalized Ptm lymphocytes during the first six days of infection. Comparing replication slopes, macaque-passaged SHIVs replicated faster than circulating SHIVs (0.98 vs. 0.75, p = 0.05), and lab-adapted SHIVs were more similar to the animal-adapted SHIVs (Fig 1C). There was a strong positive correlation between replication kinetics and IFNα resistance as measured by the AUC ratio (Spearman r = 0.88, p = 0.003) (Fig 1D). Representative macaque-passaged and lab-cultured SHIVs also exhibited faster replication than circulating SHIVs in primary Ptm PBMCs (Fig 1E). Thus, macaque-passaged SHIVs exhibited higher replication kinetics and greater IFNα resistance compared to the circulating SHIVs, and the replication kinetics in untreated immortalized Ptm lymphocytes correlate with the ability of the virus to overcome the IFNα response.
In order to identify the viral determinant(s) of replication capacity and IFNα sensitivity, we generated chimeras between the viruses that exhibited the greatest difference in replication kinetics and IFNα sensitivity, SHIV AD8-EO, which is a prototype animal-adapted SHIV, and SHIV Q23AE, which represents a circulating SHIV. Because SHIV Q23AE was generated by cloning full HIV-1 vpu and env genes directly into SHIV AD8-EO [21], the SHIV Q23AE and SHIV AD8-EO are isogenic except for the HIV-1 genes vpu, env and the second exons of tat/rev; thus, biological differences between them must be due to the HIV-1 sequences. Introduction of the entire env gene from SHIV AD8-EO to SHIV Q23AE resulted in complete recovery of replication capacity (Fig 2A). Introduction of the gp120 surface subunit of Env resulted in a modest increase in replication kinetics while introduction of the gp41 trans-membrane subunit did not result in any detectable increase in replication (Fig 2A and 2B). Because regions of the env gene contain overlapping reading frames with vpu and tat/rev, we also introduced the full vpu gene and the second tat exon, including a portion of rev, from SHIV AD8-EO to SHIV Q23AE. Introduction of neither vpu nor the second tat/rev exon resulted in a significant increase in replication kinetics (Fig 2B). Introduction of the full env gene from SHIV AD8-EO to SHIV Q23AE also restored high-level replication capacity in primary Ptm PBMCs where differences in replication are similar to those of the immortalized Ptm lymphocytes (Fig 2C).
The chimeras were then examined for their ability to replicate in the presence of IFNα treatment. Introduction of the entire HIV-1 env gene from SHIV AD8-EO to SHIV Q23AE resulted in a nearly complete recovery of IFNα resistance (Fig 2D). The gp120 chimera exhibited a modest increase in IFNα resistance while neither the gp41 nor vpu chimera demonstrated any detectable increase in IFNα resistance. Because of the poor replication capacity of the tat/rev chimera, we were unable to determine its IFNα sensitivity using the AUC ratio. Thus, the HIV-1 env gene is a major determinant of replication and of resistance to the IFNα response in Ptm cells.
To address the basis for the effect of Env on replication capacity, we measured the amount of Env protein present in virions harvested from infected immortalized Ptm lymphocytes for the panel of nine SHIVs. For the day six viral lysates, the levels of HIV-1 Env in virus expressed from cells infected with the SHIVs adapted in lab-culture or by animal-passage were consistently higher than that of virus from cells infected with the four circulating SHIVs (Fig 3A). These differences are exemplified by SHIV AD8-EO and SHIV Q23AE where there was a >10-fold difference between HIV-1 Env relative to SIV Gag p27. These patterns of Env content were similar at nine dpi (Fig 3B). In the purified virion lysates, we did not observe evidence of contamination from infected cells, for example the presence of unprocessed Gag, although we cannot definitively rule out very low levels of infected cell debris. Because the panel of SHIVs encodes HIV-1 Envs from diverse subtypes, we probed for Env using two different primary antibodies, a polyclonal rabbit sera from animals immunized with a subtype A Env protein [26] and HIVIG, antibodies pooled from HIV-1+ patients (NIH AIDS Reagent Program). We observed the same patterns of Env content using either of the primary antibodies indicating that differences in HIV-1 Env detection were not due to differences in antibody recognition of the diverse proteins (S4 Fig).
In addition to virion-associated Env content, we determined the relative infectivity of prototype macaque-passaged (AD8-EO) and circulating (Q23AE) SHIVs by measuring TCID50 in immortalized Ptm lymphocytes (S5 Fig). We found that the TCID50 values were very similar between the two viruses and in each case about 100-fold lower than the titer defined using TZM-bl cells. Thus, while the input levels of infectious virus were lower based on the TCID50 assay (MOI of 0.0002 rather than 0.02), the infecting virus titer was similar between the two viruses in our experiments. Interestingly, when we normalized TCID50 to p27 levels, we found that the infectivity of SHIV AD8-EO was approximately 30-fold higher than SHIV Q23AE (S5 Fig) suggesting that SHIVAD8-EO may have a higher ratio of infectious to non-infectious particles than SHIV Q23AE. Thus, the approach of using virus titer rather than p27 levels provided the best approach to normalizing infectious virus input (S5 Fig).
We next compared virion-associated Env content to replication kinetics across all nine viruses from Fig 3A in immortalized Ptm lymphocytes. We observed a strong positive correlation between the replication slope and Env content of virions produced at six dpi (Spearman r = 0.90, p = 0.002) (Fig 3C). Overall, SHIVs adapted by macaque-passage and in lab-culture had higher virion Env content and higher replication kinetics compared to circulating SHIVs.
Considering our previous finding that replication kinetics positively correlated with the ability to overcome the IFNα response, we compared virion-associated Env content and AUC ratio (IFN+/IFN-). We observed a positive correlation between Env content in virions and resistance to IFNα treatment (Spearman r = 0.78, p = 0.02) suggesting that Env content is linked to the ability to overcome the IFNα response (Fig 3D).
The observed differences in Env content in SHIV virions could be due to variation in synthesis within the infected cells or to variation in incorporation into newly formed virions. In order to address these two possibilities, we measured the amount of HIV-1 Env expressed in SHIV-infected immortalized Ptm lymphocytes. The pattern of Env detection was similar in cells as in virions: at six dpi, there was higher steady state Env levels in cells infected with SHIVs adapted by macaque-passage or in lab-culture compared to circulating SHIVs despite detection of PrGag at comparable levels (Fig 4A), and the same was true at nine dpi (Fig 4B). For example, SHIV AD8-EO had >30-fold (1.0 vs. 0.04) more HIV-1 Env relative to total Gag compared to SHIV Q23AE. For the nine SHIVs tested, there was a strong positive correlation between the relative steady state intracellular Env expression levels and relative virion-associated Env content at both six dpi (Spearman r = 0.87, p = 0.005) (Fig 4C) and nine dpi (Spearman r = 0.83, p = 0.009) (Fig 4D) suggesting that differences in Env content in virions are reflective of differences in Env levels in the infected cells.
To address whether differences in intracellular HIV-1 Env expression are the result of variation in transcription and splicing of env mRNA, we measured spliced env mRNA and un-spliced viral genomic RNA by reverse transcriptase quantitative PCR (RT-qPCR) for SHIV AD8-EO and SHIV Q23AE. At three and nine dpi, we did not observe any difference between SHIV AD8-EO and SHIV Q23AE with respect to the amount of spliced env mRNA relative to un-spliced viral genomic RNA (Fig 4E), although there was a statistically significant difference at six dpi; SHIV AD8-EO had ~1.5-fold more spliced env mRNA compared to SHIV Q23AE (5.6 vs. 3.8, p = 0.04). Given the small magnitude of this RNA difference compared to protein differences (>30-fold) and that differences were not observed at other time points where protein differences were observed, these findings suggest that low levels of intracellular HIV-1 Env expression are due to post-transcriptional events in SHIV-infected Ptm lymphocytes.
The finding that animal-passaged SHIVs were the most IFNα resistant of all viruses tested suggested that in vivo adaptation results in increased resistance to the IFNα response. To test the hypothesis that the process of adapting SHIVs by serial animal-animal passage in macaques increases IFNα resistance of SHIVs, we examined a collection of related SHIVs derived from a parental SHIV encoding an HIV-1 Env variant obtained without culturing, similar to the other circulating SHIVs described above [17,27]. We tested the IFNα sensitivity of the parental molecular clone (SHIVC109mc), two isolates from the third animal passage–one from early (SHIVC109P3) and one from late in infection (SHIVC109P3N)–and an isolate from the fourth animal passage (SHIVC109P4). For these studies, we applied an assay that allowed us to determined the amount of IFNα-2a required to inhibit 50% of viral replication (IFNα IC50) [11] as a more quantitative method to assess IFN sensitivity. Because these viruses were adapted for replication by serial passage in rhesus macaques, we first tested their IFNα sensitivity in primary Rhm PBMCs (Fig 5A). The parental circulating SHIVC109mc was the most sensitive to IFNα (12.8 U/ml), with adapted SHIVs derived from passage of this virus showing 25–80-fold increased resistance. The three passaged viruses were generally similar in their sensitivity to IFNα to each other (340–1,000 U/ml), suggesting that adaptation to become resistant to IFNα occurred by the time of the third passage.
Very similar results were observed in immortalized Ptm lymphocytes. The parental molecular clone SHIVC109mc was highly sensitive to IFNα treatment (IFNα IC50 1.6 U/ml), similar to SHIV Q23AE (Fig 5B). Each of the macaque-passaged isolates was more resistant to the IFNα response induced in macaque lymphocytes. Both of the isolates from passage three exhibited IFNα IC50 values >5,000 U/ml while the passage four isolate was moderately sensitive (IFNα IC50 330 U/ml). The % Vres values, which measures residual virus replication at the highest IFN level tested, demonstrated similar patterns of IFNα resistance (Fig 5C and 5D). Replication of the parental virus was nearly completely inhibited at the highest concentration of IFNα while the passage three isolates demonstrated higher residual replication. This result was consistent between both Rhm PBMCs and immortalized Ptm lymphocytes although overall residual replication was higher for the passaged isolates in Ptm cells. Interestingly, several SHIVs that were resistant to IFNα exhibited increased replication in the presence of IFNα. This increase in replication could be the result of proliferation of cells that were initially protected from infection at early time points but later became infected.
In order to sample a larger collection of viruses and compare these measures of IFNα sensitivity between circulating and animal adapted viruses, we defined IFNα IC50 and % Vres values for the three animal-adapted and four circulating SHIVs examined at a single IFN concentration in Fig 1. The seven viruses exhibited a range of IFNα IC50 values (1.7–5,000 U/ml). For example, SHIV Q23AE was highly sensitive to IFNα treatment and exhibited a dose-dependent inhibition of viral replication with an IC50 value of 1.7 U/ml (Fig 6A). In contrast, SHIV AD8-EO was completely IFNα resistant and did not exhibit inhibition of viral replication at any of the concentrations; the other animal-passaged SHIVs were similarly IFNα resistant.
The data from these SHIVs allowed us to compare the IFNα sensitivity of SHIVs encoding Envs isolated directly from people and macaque-passaged SHIVs from a total of 11 viruses. The macaque-passaged SHIVs (n = 6) were significantly more resistant to IFNα compared to the circulating SHIVs (n = 5) with respect to both IFNα IC50 value (3,180 vs. 10.6 U/ml, p = 0.0003) and % Vres (76.9 vs. 3.3%, p = 0.01) (Fig 6B and 6C).
We took advantage of the SHIV macaque model to define the role of IFNα in infection and found that infection selects for viruses that are resistant to the inhibitory effects of IFNα in macaques. Pathogenic SHIVs, which have been developed to model HIV-1 transmission and pathogenesis in macaques, are resistant to IFNα, whereas the SHIVs based on HIV-1 variants circulating in humans, including transmitted viruses, are inhibited by IFNα. Differences in sensitivity to IFNα were determined by the HIV-1 Envelope protein, which is considered a key feature of the SHIV models. Our findings underscore critical differences between SHIVs adapted for replication in macaques and HIV-1 variants isolated directly from infected individuals, including those that were recently transmitted, which represent the most biologically relevant targets of HIV-1 vaccine and prevention efforts.
Resistance to IFNα inhibition was associated with higher replication capacity, which resulted from the process of adapting virus in animals. Both increased replication and resistance to IFNα were observed in three SHIVs derived from animal passage compared to their corresponding parental molecularly cloned viruses. In all three cases, the process of adapting SHIVs in macaques, including during the critical window of early infection, led to increased replication and IFNα resistance. The differences were most striking when comparing virus derived from a SHIV constructed from HIV-1 sequences derived directly from an infected individual early in their infection [27] and pathogenic SHIVs derived from it by serial passage during the first few months of infection, which includes a time when the IFNα response is high [3–7]. Interestingly, the IFNα sensitivity of this panel of viruses mirrors those of the in vivo viral replication: the parental cloned virus, SHIVC109mc, was the most sensitive to IFNα treatment and demonstrated the lowest peak viremia. The viruses from the third animal passage (SHIVC109P3 and SHIVC109P3N) demonstrated the greatest resistance to IFNα and the highest peak viremia in vivo, 100–1,000-fold higher than the parental virus. The virus from the fourth animal passage SHIVC109P4 was intermediate in terms of IFNα sensitivity and peak viremia between the parental and passage three isolates. A variety of amino acid substitutions and deletions in variable (V1V2, V3 and V4) and constant (C3) regions of Env occurred during adaptation [17], and thus, there are many potential amino acid changes that could contribute to IFNα resistance. Importantly, selection for IFNα resistance was observed in both Rhm PBMCs and immortalized Ptm lymphocytes, suggesting a similar mechanism of IFNα inhibition in both macaques. Taken together with data showing that blocking the IFNα response early in infection results in faster progression to AIDS [9], these results suggest that the pathogenicity of adapted SHIVs may in part reflect their selection for IFNα resistance. These findings also suggest that macaque IFNα response in vivo exerts a strong selective pressure on SHIVs.
Two of the parental SHIVs examined here (ADA, SF162) were derived from HIV-1 sequences that were isolated during late-stage chronic infection by passaging the virus in vitro [28,29]. These viruses showed greater resistance to IFNα than SHIVs encoding sequences obtained directly from infected individuals early in infection. Conditions under which these viruses were derived in culture may have selected for IFNα resistance, perhaps by selecting those viruses with increased replication kinetics. It is also possible that these HIV-1 variants were already selected for features that made them less sensitive to IFNα inhibition because they were derived from later in infection. The IFNα sensitivity of the SHIVs that have not undergone either cell culture or animal adaptation to IFNα inhibition may help explain why it has been so difficult to identify SHIVs that encode HIV-1 variants isolated directly from infected individuals and are pathogenic in macaques. Indeed, our studies predict that the rare pathogenic SHIVs derived directly from HIV-1 infected people [30–32] may represent a small subset of variants that are resistant to IFNα inhibition, potentially allowing them to antagonize the early IFN storm and seed the viral reservoir to establish a persistent infection.
By generating chimeras between a pathogenic, macaque-passaged SHIV and a circulating SHIV encoding HIV-1 sequences isolated directly from an infected individual, we demonstrated that HIV-1 Env is a critical determinant of the ability of SHIVs to overcome the IFNα response. We found that the underlying mechanism for role of Env in IFNα sensitivity is related to Env protein levels. Interestingly, differences in Env protein levels were not reflected in env RNA levels. There was a ≤1.5 fold difference in spliced env RNA levels between adapted and circulating SHIVs compared to protein differences of >30-fold. These findings suggest that low levels of intracellular HIV-1 Env expression are due to post-transcriptional events in SHIV-infected macaque cells. While Env content correlated with IFNα sensitivity, we observed differences in the IFNα sensitivity of lab-cultured and macaque-passaged SHIVs despite similar levels of Env. Thus, while our results overall suggest that Env content plays an important role in the ability to overcome the macaque IFNα response, other determinants in the envelope protein may also contribute to IFNα resistance.
The finding that IFNα resistance mapped to HIV-1 env and was the result of high Env content was somewhat surprising considering that HIV-1 Env has not previously been implicated in viral antagonism of the IFN-I response, although some studies have suggested that IFN-I treatment predominantly affects very early stages of viral replication [33,34]. These findings raise several interesting possibilities. One is that HIV-1 Envs from adapted SHIVs act directly in evasion or antagonism of the IFNα response by protein-protein interactions with an IFN-induced host factor(s). An alternative hypothesis is that high HIV-1 Env expression/content contributes to increased kinetics allowing the virus to overcome the IFNα response by saturating IFN-induced factor(s). There is some precedent for this model as the ability to saturate IFN-induced restriction factors has been demonstrated in vitro [35–38]. In support of the model that an IFN-induced inhibitory factor is being saturated, we found a strong positive correlation between replication kinetics and IFNα resistance.
IFN–induced, HIV-specific restriction factors are typically species-specific, presumably because HIV-1 has adapted to its human host. Thus, we expect that the IFN-induced factor(s) that limit replication of circulating SHIVs, but not animal adapted SHIVs, may have similar species-specificity. However, we could not directly test whether this pathway is active in human cells with the viruses studied here because SHIVs replicate poorly in human cells due to other host restrictions. A similar correlation between replication capacity and IFNα resistance was recently reported for HIV-1 in human lymphocytes [39].
Differences in virion-associated Env content between SHIVs adapted by lab-culture and/or macaque-passage and those based on circulating variants were reflected by similar differences in Env levels in infected cells. Cells infected with adapted SHIVs had high Env levels whereas those infected with circulating SHIVs had low Env levels. Low Env content could help minimize immune recognition. For example, low Env content has been suggested to reduce antibody avidity [40]. Given that HIV-1 can spread as both cell-free and cell-associated virus [41], one intriguing possibility to explain our finding is that high Env expression in infected cells is leading to increased cell-cell transmission and saturation of IFN-α inhibition of cell-cell virus spread. Interestingly, the pattern of IFNα inhibition was the same whether cells were pre-treated with IFNα or treated just after infection (S3 Fig); even in cells pretreated with IFNα, there is a delay in the inhibition due to IFNα. This could be a result of an effect on cell-cell transmission that is only seen after the initial round of cell-free virus infection. Alternatively, it may indicate the IFN-induced factor is packaged into virions and inhibits later rounds of infection. Finally, a technical explanation for the observed delay in IFNα inhibition could be that the p27 ELISA used to determine virus levels is not sensitive enough to detect differences at very low levels of viral replication.
Overall, the results of this study demonstrate differences in the IFNα sensitivity between SHIVs used to model HIV-1 infection and HIV-1 variants circulating in infected people. They also suggest that the common focus on lab-adapted viral variants may have limited the ability to identify important mechanisms underlying the IFN-I-induced inhibition of HIV-1 variants circulating in people. The results uncover a key role for envelope in the process of adaptation of lentiviruses to the IFNα response and help explain why SHIVs generally do not cause pathogenic infections in macaques without adaptation. These findings may provide insight into the development of improved SHIV challenge viruses for non-human primate models of HIV-1 infection that currently serve as gatekeepers for HIV-1 vaccine and prevention studies.
Full-length proviral SHIV clones encoding the region spanning the vpu and env open reading frames were generated using SHIV AD8-EO as a vector [21]. Expression plasmids encoding vpu and env open reading frames for Q23ENV.17 [18], BG505.W6M.ENV.B1 [20], MG505.W0M.ENV.H3 [20], and QF495.23M.ENV.A3 [19] A204E and G312V variants were amplified using primers designed to introduce an EcoRI site 5’ of the vpu start codon and a SalI site immediately 3’ of the env stop codon. The amplicons were then digested and ligated into the SHIV AD8-EO full-length proviral plasmid using EcoRI and SalI. Chimeras between SHIV AD8-EO and SHIV Q23AE were generated by overlap-extension PCR.
The following full-length proviral plasmids of the parental SHIVs were also used in this study: SHIV AD8 [21] and SHIV SF162 [22].
Full-length, replication-competent virus was generated by transfecting 2x106 HEK 293T cells (American Type Culture Collection, Manassas, VA) with 4 μg of proviral plasmid DNA and 12 μl of Fugene 6 transfection reagent (Roche). Virus was harvested 48 hours post-transfection, passed through a 0.2 μm sterile filter and concentrated ~10-fold using a 100 kDa molecular weight protein concentrator (Amicon).
Replication-competent stocks of SHIV SF162P3 [23], SHIV 1157ipd3N4 [16], SHIVC109mc, SHIVC109P3, SHIV109P3N and SHIVC109P4 [17] were generated by expanding the virus in immortalized Ptm lymphocytes [25]. For each virus, 2x106 cells were infected at an initial multiplicity of infection (MOI) of ~0.02 by spinoculation at 1200 x g for 90 minutes at room temperature. After spinoculation, cells were washed 1x with 1 ml of Iscove’s modified Dulbecco’s medium (IMDM) supplemented with 10% heat-inactivated FCS, 2 mM L-glutamine, 100 U of penicillin/ml, 100 μg of streptomycin/ml and 100 U of interleukin-2/ml (Chiron) (complete IMDM), re-suspended in 2.4 ml of media and plated in a 6-well plate. Every three days, infected cells and cell supernatant were harvested and separated by pelleting at 650 x g for five minutes at room temperature.
Aliquots of replication-competent virus were stored at -80°C. The viral titer of each viral stock was determined by infecting TZM-bl cells and counting the number of blue cells at 48 hours post-infection after staining for β-galactosidase activity [20].
Replication of SHIVs was assessed using immortalized Ptm CD4+ lymphocytes [25] maintained in complete IMDM. One million Ptm lymphocytes were infected at an MOI of 0.02 by spinoculation as described above. In some cultures, recombinant human IFNα-2a (PBL Assay Science, Piscataway, NJ) was added at a final concentration of 1,000 U/ml five hours after the initial infection. Every three days, 400 μl of each cell supernatant was replaced, including with IFNα-2a if appropriate. SIV p27 concentrations were determined using a SIV p27 antigen ELISA (ABL, Rockville, MD).
For some experiments Ptm or Rhm PBMCs were used: these cells were isolated from whole macaque blood (Washington National Primate Research Center, Seattle, WA) from two separate donors using 95% Lymphoprep (STEMCELL, Vancouver, BC). Isolated PBMCs were stimulated for three days prior to infection with IL-2 (20 U/ml) and concanavalin A (5 μg/ml) in RPMI 1640 medium supplemented with 20% FCS and 2 mM L-glutamine. Ptm or Rhm PBMCs from two donors were pooled immediately prior to infection, and 1x106 PBMCs were spinoculated and maintained as described for immortalized Ptm lymphocytes.
The amount of HIV-1 envelope (Env) was determined by semi-quantitative western blot using the LICOR Odyssey system. For virion-associated Env content, supernatants from infected immortalized Ptm lymphocyte cultures were pelleted through a 25% sucrose cushion by ultracentrifugation for 90 minutes at 28,000 rpm. Virus pellets were lysed in 70 μl of radioimmunoprecipitation assay (RIPA) buffer for 10 minutes at room temperature. The concentration of SIV p27 antigen in the virus lysates was determined by ELISA, and virus lysate input was normalized to 5 ng of p27. Western blotting was performed as described previously [42] using rabbit polyclonal anti-HIV-1 Env sera [26] and mouse anti-SIV p27 monoclonal antibody (ABL, catalog no. 4323). Both gp160 and gp120 bands were included in the quantification of Env signal. For western blotting of whole cell lysates, SHIV-infected immortalized Ptm lymphocytes were pelleted by spinning at 650 x g for 5 minutes at room temperature. Cell pellets were washed with 1x PBS and then lysed in 100 μl of RIPA buffer for 10 minutes at room temperature. Western blotting was performed as described for virus lysates.
HIV-1 RNA was measured by reverse transcriptase quantitative PCR (RT-qPCR). Total RNA was isolated from SHIV-infected immortalized Ptm cells using the miRNeasy Mini Kit (Qiagen). For each reaction, 50 ng of total RNA, measured by Nanodrop spectrophotometer was amplified using the Superscript III Platinum SYBR Green One-Step RT-qPCR kit with ROX (Invitrogen). Primers 5’- AGGGACTTGGCAAATGGATTGTAC-3’ and 5’ GTGTAATAGGCCATCTGCCTGCC-3’ were used to amplify gag from unspliced RNA. To amplify splice vpu/env mRNA, the forward primer 5’-AGGAACCAACCACGACGGAGTGCTC -3’, which binds upstream of the splice donor site in the 5’ LTR, and reverse primer 5’-CATTGCCACTGTCTTCTGCTCTTTC-3’, which binds downstream of the Vpu start codon, were used. To amplify macaque β-actin mRNA, primers 5’-CAACCGCGAGAAGATGACCCAGATCATG-3’ and 5’-AGGATGGCATGGGGGAGGGCATAC-3’ were used. Relative levels of HIV-1 env mRNA for each virus was determined using the following equation: 2-ΔΔ = [(CT env—CT beta actin)]—[(CT genomic—CT beta actin)] [43].
For each SHIV, 4.25x106 Ptm or Rhm lymphocytes were infected at an MOI of 0.02 in a final volume of 1.4 ml of complete IMDM using spinoculation. Approximately 2.5x105 infected cells in 200 μl of media were plated in each well of a 96-well plate containing 50 μl of media containing the indicated concentrations of IFNα-2a; experiments were performed in duplicate. Cell supernatants were harvested at 7 days post-infection (dpi) and used to determine the amount of virus using TZM-bl cells (NIH AIDS Reagent Program). For the data analysis, all values were plotted and statistical analyses performed using Prism version 6.0c (GraphPad Software). Percent viral replication was determined by dividing the amount of β-galactosidase activity in the IFNα treated sample by the untreated sample. The concentration of IFNα at which 50% viral inhibition was achieved was interpolated from a non-linear, best-fit curve. The amount residual viral replication at the highest concentration of IFNα (5000 U/ml) was also determined.
TCID50 assay was used to determine the end-point dilution of the SHIV stocks at which infection is detected in 50% of the immortalized Ptm CD4+ lymphocyte culture replicates. Seven serial 4-fold dilutions of the SHIV stocks were prepared in triplicate in 96-well flat-bottomed tissue culture plates. Briefly, the SHIV stocks were diluted 1:12 in complete IMDM and 200 μl of diluted virus was transferred to the first well of a 96-well plate. Next, the virus was serially diluted by transferring 50 μl of the virus to the subsequent well containing 150 μl of complete IMDM. 2 x 105 immortalized Ptm CD4+ lymphocytes, in 50 μl of complete IMDM, were seeded in each well. Every 4 days 125 μl of the culture supernatant was removed from every well and replaced with fresh 150 μl of complete IMDM. On day 12, 100 μl of the culture supernatant from each well was harvested and tested for SIV p27 using a SIV p27 antigen ELISA. TCID50 was calculated using the Spearman-Kaber method.
|
10.1371/journal.pgen.1001096 | The Potential for Enhancing the Power of Genetic Association Studies in African Americans through the Reuse of Existing Genotype Data | We consider the feasibility of reusing existing control data obtained in genetic association studies in order to reduce costs for new studies. We discuss controlling for the population differences between cases and controls that are implicit in studies utilizing external control data. We give theoretical calculations of the statistical power of a test due to Bourgain et al (Am J Human Genet 2003), applied to the problem of dealing with case-control differences in genetic ancestry related to population isolation or population admixture. Theoretical results show that there may exist bounds for the non-centrality parameter for a test of association that places limits on study power even if sample sizes can grow arbitrarily large. We apply this method to data from a multi-center, geographically-diverse, genome-wide association study of breast cancer in African-American women. Our analysis of these data shows that admixture proportions differ by center with the average fraction of European admixture ranging from approximately 20% for participants from study sites in the Eastern United States to 25% for participants from West Coast sites. However, these differences in average admixture fraction between sites are largely counterbalanced by considerable diversity in individual admixture proportion within each study site. Our results suggest that statistical correction for admixture differences is feasible for future studies of African-Americans, utilizing the existing controls from the African-American Breast Cancer study, even if case ascertainment for the future studies is not balanced over the same centers or regions that supplied the controls for the current study.
| This paper discusses and provides unique insight into an important problem raised by the current state of genetic studies into disease susceptibility, namely whether we can reuse genetic data from participants genotyped as controls in one study when cases (people with a disease of interest) are obtained from other studies, or whether each new study needs its own controls. We are interested in whether studies where cases and controls are sampled differently will give correct answers and are as powerful statistically as when new control data is also genotyped. Because of the huge investments made recently in large scale genotyping of cases and controls for various diseases, this is a timely question. This question is especially important in understanding the genetic causes of disease in as-yet relatively understudied population groups, such as African-Americans, in order to speed up progress when this is possible. We give theoretical results about the power of studies that reuse existing control genotypes based on statistical considerations. We also provide analysis of real data from a major study of the genetic causes of breast cancer in African-American women in order to shed practical light upon this issue.
| A genetic association study estimating the main effects of single nucleotide polymorphisms (SNPs) or other genetic variants upon the risk of a rare or common disease in minority populations is a setting in which it is especially attractive to consider the use of existing genotype data as a supplementary or even a primary source of controls. DNA samples may be expensive and difficult to obtain, and response rates are often lower in minority populations [1]. Researchers might consider using an already available “stand in” population sample as controls, provided that genotype frequencies are equivalent to those in the population from which controls would be drawn. There are however two immediate concerns raised, one fundamental and the other technical in nature: The fundamental question is whether or not the controls are sampled from the same underlying population (or populations) as are the cases – or more generally the feasibility and “cost” (generally loss of statistical power) of controlling for case/control differences if they arise. The technical question is whether differences in genotyping, including differences in DNA preparation, and in the actual markers genotyped in cases and controls, i.e. when the platforms are not identical so that imputation is relied upon to make up the difference, may introduce false positive (or false negative) associations.
Consider the problem of conducting a genetic association study aimed at discovering genetic variants related to the risk of a disease, where there already exists extensive genotyping data, perhaps publicly available, for members of similar populations. If the disease under consideration is rare (so that genotype frequencies for controls may be expected to be the same as in the general population) then it is intuitively appealing to consider using existing control data from studies of other rare diseases (or population-based studies if they exist) to reduce the cost or increase the statistical power of an association study. From the classical case-control literature, a study that uses 1∶ matching of controls to each of cases will be equivalent in power to a 1∶1 matched study with cases (and an equal number of controls). Thus a study with a large number of controls, , for each case will have nearly twice the effective sample size of a 1∶1 matched study [2].
Note that 1∶m matched studies (with m>1) are only cost effective if it is more costly or difficult to obtain additional cases than it is to obtain additional suitable controls since the increment in effective sample by (for example) doubling the number of case-control pairs (in a 1∶1 matching) is twice that of adding two new controls to each existing pair (to achieve 1∶3 matching) to a study even though the total number of participants is the same. If however the cost of adding an additional control is far less than that of adding an additional case (because control data is already available) then adding the almost “free” controls is highly attractive although the returns diminish as more and more controls are added, with the increment in effective sample size governed by m/(m+1).
This paper considers these issues from both theoretical and empirical perspectives. We apply a recent generalization [3]–[6] of the testing procedure of Bourgain et al [7] to the situation where population substructure (in the broad sense), including admixture and relatedness between subjects, is estimated from marker data rather than being assumed to be known as the basis for our theoretical considerations of study power. We point out below the relationship between this procedure and that of the more widely known principal components technique [8]. Our empirical investigation of the use of existing controls utilizes data from a genome-wide association study (GWAS) of breast cancer in African American women, namely the African American Breast Cancer (AABC) study, in which cases and controls come from a total of 9 different studies widely distributed geographically throughout the United States. African Americans are a relatively understudied group (compared to European Americans) in studies of genetic susceptibility. African Americans are admixed with Europeans and (in some cases) with Hispanics (themselves an admixed group) and Native Americans [9], [10]. We examine empirically the false positive rates that occur when cases from one geographical location or study within the AABC study are combined with controls from other AABC locations or studies, as well as the success (and cost in terms of loss of effective sample size) of adjustment for the observed population differences in global genetic ancestry when analyzing such illustrative data sets derived from the AABC study.
We utilize an approach derived from that of Bourgain et al [7] which has been discussed extensively in recent papers [3]–[5]. This approach for accounting for relatedness between subjects in association tests adopts a “retrospective” approach towards the problem of testing for disease associations using marker data, in which (as in the Armitage test) the allele frequency of a variant is related to case-control status. In vector notation we have of observed values for a given SNP j. Here is the total number of subjects (cases+controls) in the study, and SNP values are coded as (0,1,2) for the number of copies of a specified allele, , carried by subject . (This coding of SNP genotype implies that we are interested in additive models for the relationship between disease risk and genotype but the approach can readily be extended to other codings). The retrospective approach models the mean of as a function of case-control status. If we define the design matrix C to have rows (1 ci) where ci is case-control status (0 or 1) for subject i then the mean, , of is written asRelatedness between subjects induces a covariance matrix for the number of copies of a given SNP of form(1)with specific to each SNP but with the same matrix for all SNPs. In fact, for known pedigree relationships, and unrelated founders, this matrix has diagonal elements equal to where is the inbreeding coefficient for subject and each off-diagonal element, , is twice the kinship coefficient for the relationship between subjects and [7].
It is worth noting that in general the topic of relatedness includes what is often considered to be population substructure. For example consider two large but isolated populations (freely mixing within each population) that have been separated for many generations. While a random sample of people from the same population (with sample size small relative to the population size) might be considered unrelated to each other when considering that population separately, when considering the two populations together people from one isolated population are considered to be related to each other, relative to those in the other population. In particular, genetic markers will, through a process of random drift and other factors, be able to distinguish members from the two populations, and this will be detectable when calculation of the matrix is performed. A standard method of simulating genetic markers for divergent populations stemming from the same ancestral population (e.g. the Balding-Nichols model [11]) can readily be shown to produce covariance matrices of the form of expression (1).
If and are both known then the best linear unbiased estimate (BLUE) of the regression vector is of weighted least squares formand the variance covariance matrix of the estimates is in the form ofThus inference on the significance of the allele frequency difference between cases and controls may be based upon the Wald test statistic(2)with the (2,2) element of
In general of course, and are not known, except in the case of known pedigrees and unrelated founders, where can be computed from first principles. The estimation of using marker data has been considered by a number of authors and both method of moments [4], [5] and maximum likelihood methods [3] have been considered. A method of moments estimator of can be concisely written [5] as(3)and the estimate of asOne value of this approach, which is exploited here, is that it is relatively easy to compute the power of the Wald test if we can hypothesize a form of the relatedness matrix . For a given form for (below we consider several forms for both isolated population models and more complex admixed populations) then for a given sample size, , a given allele frequency for a causal SNP, and a hypothesized difference in allele frequencies between cases and controls (which can then be related to odds ratios in typical case/control analysis) we can compute the non-centrality parameter of (and hence the power of the test) as(4)We illustrate the computation of this non-centrality parameter for a number of important special cases in the results section below. It is worth noting now, however, that the Bourgain test appears to be reasonably powerful compared to other procedures, and can sometimes be considered as a compromise between the principal components method [8] and genomic control [12]. We attempt to justify this last statement in the results section below.
In addition to the Bourgain test we used several well known tools for addressing population structure in the AABC data. For example we computed eigenvectors through the use of the program EIGENSTRAT [8]. Briefly, each eigenvector explains a proportion of the genetic variation among samples in the analysis so that the leading eigenvector explains the greatest variation, followed by the second eigenvector, and so forth. The full set of eigenvectors form an orthonormal basis so that each eigenvector is scaled on the unit interval and linearly independent from all other eigenvectors. Note that the EIGENSTRAT procedure is operating on the same estimated matrix, , that we have described above.
To assess ancestry within the AABC study in relation to reference populations from HapMap, we performed a principal components analysis based on ancestry informative markers that were genotyped in both the AABC study and the HapMap Phase 3 populations. The 2,546 ancestry informative markers (contained within the Illumina 1M genotyping array which was used in the AABC scan) were selected based on low inter-marker correlation and high correlation to a previously determined eigenvector that explained African and European ancestry.
We quantified percent African ancestry for each of the nine AABC study populations by running the program STRUCTURE for each study population. The program implements a Markov Chain Monte Carlo algorithm that provides the posterior estimates of the proportion of ancestry from each of k clusters for each individual, where k is specified by the investigator. For each AABC study population, we assigned k = 3, including genotypes from the same ancestry informative markers used in PCA genotyped in YRI, CEU, and JPT from HapMap Phase 3.
AABC included 9 epidemiological studies of breast cancer among African American women, which comprise a total of 3,153 cases and 2,831 controls. Below is a brief description of these studies.
Genotyping in stage 1 was conducted using the Illumina Human1M-Duo BeadChip. Of the 5,984 samples from these studies (3,153 cases and 2,831 controls), we attempted genotyping of 5,932, removing samples (n = 52) with DNA concentrations <20 ng/ul by pico green assay. After clustering the genotype data we removed samples based on the following exclusion criteria: 1) unknown replicates (≥98.9% genetically identical) that we were able to confirm (only one of each duplicate was removed, n = 15); 2) unknown replicates that we were not able to confirm through discussions with study investigators (pair or triplicate removed, n = 14); 3) samples with call rates <95% after a second attempt (n = 100); 4) samples with ≤5% African ancestry (n = 36) (discussed below); and, 5) samples with <15% mean heterozygosity of SNPs in the X chromosome and/or similar mean allele intensities of SNPs on the X and Y chromosomes (n = 6) (these are likely to be males).
In the analysis, we removed SNPs with <95% call rate or minor allele frequencies (MAFs) <1%. To assess genotyping reproducibility we included 138 replicate samples; the average concordance rate was 99.95% (>99.93% for all pairs). We also eliminated SNPs with genotyping concordance rates <98% based on the replicates. The final analysis dataset included 3,016 cases and 2,745 controls, with an average SNP call rate of 99.7% and average sample call rate of 99.8%. Hardy-Weinberg equilibrium (HWE) was not used as a criterion for removing SNPs for this analysis.
We use the Balding Nichols model [11] for allele frequency differences between isolated populations. In this model allele frequencies for a SNP in modern data populations are distributed according to a beta distribution with the ancestral allele frequency of that SNP. In this model the variance of the modern day allele frequency is , thus is a parameter specifying the degree of separation between the modern day and ancestral population. As described in Rakovski and Stram 2009 [4] if genotypes are obtained for randomly sampled individual from two modern day isolated populations using this model and the separation of each modern day population from the ancestral population equals (for ) statistic then the covariance matrix, between subjects for the jth SNP will have diagonal terms equal to for members of the first population, diagonal terms equal to for members of the second, off diagonal terms of , , or zero for pairs of individuals who are either both from the first population, both from the second population or from different populations respectively. Here is the frequency in the ancestral population of SNP . Consider now a study in which all cases come from one isolated population, and all controls from another. Assume for simplicity that (both populations have the same degree of separation from their ancestral source), and that the number of cases and controls are both equal to so that total sample size is (the calculations below can be readily altered for different matching fractions if necessary). Thus we can write the variance of the estimator of the case-control difference for SNP aswith the first column of C being a vector of 1's and the second column of C a vector of 1's and 0's indicating case-control (and population) status. Using a readily derived formula for the inverse of an matrix of compound symmetric formwe can easily write(5)Thus the non-centrality parameter, = of a test of association does not increase linearly in N, but rather is bounded above by the value . This can impose severe limitations on the power of any study in which there are such differences. To put this in perspective, consider two isolated populations which are each separated from their ancestral population with an F value of 0.0005, and consider an allele that exhibits 40 percent frequency in the ancestral population. The variance of the difference between the two isolated modern day populations in the frequency of this allele is equal to so that we would expect by chance that there is a about a 1.5 percent difference in allele frequencies between cases and controls for such an allele. Consider now the detection, in a study of 5,000 cases and 5,000 controls, of a disease-causing allele of the same frequency associated with a 10 percent difference in allele frequencies between cases and controls ( = .2). The difference in allele frequencies is approximately 6 times larger than expected due to population differences, and can be seen to correspond to an odds ratio for disease, under a multiplicative risk model, of 1.5 per copy of the risk allele. From equation (2) the non-centrality parameter will be equal to 34.73 in this case; on the other hand if between cases and controls is 0 the non-centrality parameter will equal 208.33. Thus a study that would, given no differences between cases and controls in population of origin, have overwhelming power (>.9999) to reject the null hypothesis at a genome wide level significance (p<) is, under this alternative, reduced to having power of only 56 percent after correcting for the differences in origins of cases and controls. The survey of European populations by Nelis et al [21] estimates fixation indices, , (which can be equated to under the Balding Nichols model) between populations in SNP allele frequencies which range from less than .001 between neighboring populations to 0.023 for Southern Italy versus parts of Finland. Because our values as defined above are between present day and ancestral populations the fixation indices calculated between present day populations by Nelis et al need to be multiplied by ½ to be consistent with our definition of . Thus the example we have given () corresponds only to the nearest neighbor populations in Europe and would appear to throw into doubt any thought of using control data not perfectly matched in ancestry to cases.
While the calculations given above appear to be pessimistic regarding the usefulness of shared control data it is important to note that the completely isolated population model is naïve and makes assumptions not applicable to the study subjects for the AABC study or indeed for most modern populations. Therefore we broaden our discussion to admixed populations, specifically, when the DNA from both cases and controls come from groups that are admixed from the same two ancestral populations. We consider this in two parts, first deriving results for comparisons between “completely” admixed populations, i.e. where the two populations have different levels of admixture between the ancestral populations, but when there is no within-population heterogeneity in ancestry. Next we focus on the much more realistic setting of incompletely admixed populations serving as cases and controls.
Our analysis focuses upon (1) estimating a more appropriate model for the distribution of ancestry in the data from the AABC data than the homogeneous “complete” admixture described above, (2) checking the adequacy of this model, enriching it if necessary, and (3) describing the implications of the model for the likely power to detect associations in studies in which all cases come from outside the AABC study populations, and controls are chosen from within the AABC. We can partly mimic such studies by making up “pseudo” case-control studies using the data from the different study sites within the AABC study.
We have adopted a somewhat non-standard approach in relying upon the Bourgain test rather than principal components [8] or related methods [22]–[24] to control for population structure in a GWAS of a minority population with cases/controls drawn from multiple studies with different designs and recruitment approaches. We have done this mainly because we can give certain theoretical results for the Bourgain test when assuming specific forms for the true kinship matrix using this procedure. It is worth noting that the Bourgain test can be regarded as a random effects version of the usual principal components method. In particular the Bourgain model can be alternatively described as a model for the mean of conditional upon all eigenvectors of as(10)Now consider the coefficients as being independent random effects with mean zero and variances equal to times the associated eigenvalues, of . Averaging over all the random will yield the unconditional mean and variance covariance matrix . Our “smoothed” estimate, , of is motivated by expression (10), and choosing a value of is analogous to choosing the number of eigenvectors to be used as fixed effects by EIGENSTRAT. The random effects framework also highlights the relationship between the Bourgain procedure and the genomic control method of Devlin and Roeder [25]. In genomic control one additional parameter that governs the dispersion of the test statistic is used to assess the association between and case-control status. In the smoothed version of the Bourgain test introduced here, we choose a total of such parameters.
We have shown that if cases and controls come from genetically distinct populations but ones that have only recently diverged (so that the parameter is very small) then some limited power remains to detect true marker associations so long as the true value of is very large compared to the “typical” differences between cases and controls seen with the other markers. This is also analogous in interpretation to the genomic control method. However our explicit description of the upper bound on the noncentrality parameter of the association test for such a study clearly shows the limits of this design, and by implication, the limits of the genomic control procedure as well. Basically neither of these two methods behaves “properly” from a statistical point of view as sample size increases, i.e. the non-centrality parameter under an alternative hypothesis (and hence power) does not increase correspondingly. In genomic control the overdispersion parameter that the procedure corrects for increases with sample size, while for the Bourgain test the noncentrality parameter is bounded from above.
For case-control studies involving two or more similarly admixed populations that differ in admixture fraction, the key issue in assessing the power of a study using cases from one population and controls from another is in determining the within-population heterogeneity of the admixture fractions, relative to the between population differences in average admixture. If the within-population heterogeneity is small then the situation is equivalent to the case of isolated populations, i.e., there will be a bound on the power of a study to detect an effect with the bound determined by the upper limit on the noncentrality parameter as a function of as computed above.
Despite the concerns raised in our theoretical considerations, in our assessment of the observed marker data from the AABC study we tentatively conclude that reuse of controls data from this study in future work may be statistically feasible. While there are clear differences in average admixture fraction between studies these are dwarfed by the within-study heterogeneity. Other signs of hidden structure in the AABC studies (as evidenced by additional eigenvalues which are significant by the Tracy-Widom test) do not appear to have a very large impact (Figure 5) on the power of our hypothetical study using the CBCS and MEC cases and controls respectively. Control for the first few (1–200 in our case) eigenvectors appears to dramatically reduce false positive associations with very little power loss (about a 7 percent reduction in effective sample size) relative to studies of homogeneous sets of cases and controls. We have used a specific set of ancestry informative markers in our analysis but the existence of genome-wide data for the AABC allows for considerable latitude in selecting SNPs to control for admixture, and even randomly selected SNPs, if enough are considered, can be used for admixture correction. Our use of the Bourgain test when considering the feasibility of a particular study design allows us to consider noncentrality parameters (and hence power) in particularly simple and helpful ways. While we have focused on the Bourgain method to correct for admixture differences in the AABC study our specific finding (that little loss in power is anticipated when re-using control data from this study) is likely to apply also to fixed-effects methods such as treating principal components or STRUCTURE estimates of percentage ancestry from ancestral populations as covariates. Our reasoning is based upon the close relationship between the principal components methods and the random effects rationale for the Bourgain test as given in equation (10) and also on the high correlation seen between STRUCTURE estimates of African ancestry in the AABC study and the first eigenvector from principal components.
|
10.1371/journal.pgen.1007251 | Synchronous termination of replication of the two chromosomes is an evolutionary selected feature in Vibrionaceae | Vibrio cholerae, the causative agent of the cholera disease, is commonly used as a model organism for the study of bacteria with multipartite genomes. Its two chromosomes of different sizes initiate their DNA replication at distinct time points in the cell cycle and terminate in synchrony. In this study, the time-delayed start of Chr2 was verified in a synchronized cell population. This replication pattern suggests two possible regulation mechanisms for other Vibrio species with different sized secondary chromosomes: Either all Chr2 start DNA replication with a fixed delay after Chr1 initiation, or the timepoint at which Chr2 initiates varies such that termination of chromosomal replication occurs in synchrony. We investigated these two models and revealed that the two chromosomes of various Vibrionaceae species terminate in synchrony while Chr2-initiation timing relative to Chr1 is variable. Moreover, the sequence and function of the Chr2-triggering crtS site recently discovered in V. cholerae were found to be conserved, explaining the observed timing mechanism. Our results suggest that it is beneficial for bacterial cells with multiple chromosomes to synchronize their replication termination, potentially to optimize chromosome related processes as dimer resolution or segregation.
| Most bacteria encode their genetic information on a single chromosome. The pathogenic bacterium Vibrio cholerae is an exception to this rule and carries two chromosomes of different sizes, each having one origin of replication. A very basic research question is how the replication of the two chromosomes is timed starting from their replication origins. If they start simultaneously, the smaller chromosome would finish replication earlier than the larger chromosome. Interestingly, the timing in V. cholerae is such that the smaller chromosome starts replication after a time delay, resulting in synchronous replication termination of the two chromosomes. Here we answer the question whether it is the termination synchrony which is under evolutionary pressure, or whether a certain duration of the delay between the two chromosomes to start replication is of biological importance. To this end, we analyzed replication in different species of the Vibrionaceae phylogenetic group with differently sized chromosome pairs. Our results indicate that a synchronous termination of the two chromosomes in this group of bacterial species is under evolutionary selection, suggesting it to be potentially important for the process of cell division.
| The diversity of regulatory systems of DNA replication has been studied in multiple bacteria [1–4]. An especially interesting group of bacteria with regard to DNA replication are those with multiple chromosomes. While a single chromosome is the norm in bacteria, about 10% of species in a diverse set of phyla carry more than one chromosome [5]. The best studied system in this respect is that of V. cholerae, the causative agent of the cholera disease [6, 7]. The genome of strain O1 El Tor N16961 is divided into two chromosomes of about 3 (Chr1) and 1 Mbp (Chr2) respectively [8]. Chr1 carries most of the essential genes [8, 9]. Replication at the origin of Chr1 (ori1) is initiated by the initiator protein DnaA, as is the case in almost all known bacteria [10]. Chr2 encodes its own initiator, RctB.[8, 11]. Notably, no RctB-like proteins have yet been found outside the phylogenetic group of Vibrionales. The structure of its central two domains (of four in total) resembles that of several plasmid replication initiators [12, 13]. RctB binds to a set of so-called iterons within ori2 to initiate replication [14, 15], which contain the sequence GATC, methylated at the adenine by the Dam methyltransferase [16]. Binding of RctB was shown to be specific for fully-methylated GATCs, which in conclusion renders Dam essential in V. cholerae, unlike in E. coli [17–19]. RctB also binds to another type of sequence, the so-called 39-mers, which are also located at ori2 [20]. However, the binding of RctB to the 39-mers does not activate replication as does its binding to the iterons; on the contrary, this suppresses initiation [20]. The balance between the activating and repressing action of RctB, in conjunction with a handcuffing mechanism, is thought to generate tight control of Chr2 replication in a cell-cycle-dependent manner [20, 21]. It was found that the two chromosomes start replication with a time delay in between [22–25]. In the search for regulatory mechanisms of communication between the two chromosomes, it was found that Chr1 was insensitive to the blockage of Chr2 replication [26]: it was shown that Chr1 controls replication of Chr2 through a short sequence about 800 kbp downstream from ori1 [27]. This site was later named crtS, for ‘Chr2 replication triggering site’ [25]. Replication of crtS triggers the replication of Chr2, which is initiated after a short delay [25]. Moving the crtS site to other positions on Chr1 led to a corresponding shift in Chr2 initiation. The mechanism underlying the triggering effect of crtS is not yet fully understood but might involve physical contacts that were observed to occur between crtS and ori2 [25].
Replication of Chr2 in V. cholerae starts after about two-thirds of Chr1 are replicated. This timing leads to termination of Chr2 replication at about the same time as that of Chr1. To better understand the mechanism underlying this phenomenon, we investigate here if it is the Chr2 replication starting after two-thirds of Chr1 is replicated which is important to the cell, or if the orchestrated termination of both chromosomes is the driving force of evolutionary selection. To this end, we tested whether the V. cholerae paradigm applies to other species of the Vibrionaceae and derive general rules of replication control.
While early studies suggested a synchronous replication start of the two V. cholerae chromosomes, more recent studies support a time delay between Chr1 and Chr2 initiation [22, 23, 25, 28]. In synchronized V. cholerae cell cultures, such a time delay should lead to a situation with only Chr1 replicating in all cells short after initiation and later Chr2 replication. However, to date no synchronization method for V. cholerae has been available. Here we test if a synchronization method established for Escherichia coli can be used to synchronize V. cholerae populations [29]. The method is based on the induction of the stringent response as a cellular answer to nutrient limitation. In E. coli, addition of serine hydroxamate (SHX) blocks re-initiation of DNA replication, while ongoing replication rounds are finished, leading to cells with fully replicated chromosomes. Transfer of the cells to SHX-free medium then leads to a synchronous re-start of DNA replication. The stringent response in V. cholerae can also be induced by SHX treatment [30]. Consequently, addition of 0.9 mg/ml SHX to an exponentially growing V. cholerae batch culture resulted in clear inhibition of growth (Fig 1A). Flow cytometry analysis of the cellular DNA content shows asynchronous replicating cells before SHX treatment and cells with either 1+1 or 2+2 fully replicated chromosomes (Chr1 and Chr2) after SHX treatment for 150 minutes (Fig 1B). After transfer to growth medium without SHX, the DNA content of the cells increases gradually as would be expected for a synchronously replicating population (Fig 1B and 1C). To analyze replication of the two chromosomes individually, we performed marker frequency analyses using high density custom microarrays. We used the genome sequence of strain N16961 as a reference, which is very similar to the strain A1552 used here. A1552 (El Tor biotype, Inaba serotype) is a pathogenic strain of V. cholerae that was isolated from a traveler to Peru in 1992 (Strain DSM 106276 in the German Collection of Microorganisms and Cell Culture, DSMZ)[31]. However, during initial experiments we observed patterns indicating some sort of chromosomal rearrangements within the V. cholerae A1552 in comparison to strain N16961. For the latter, it was found that the strain used in most labs actually carries a chromosomal inversion between two operons encoding ribosomal RNAs ([25], Supporting S1A and S1B Fig. We used a set of PCRs to check potential additional inversions between ribosomal operons within the A1552 strain (Supporting S2 Fig). Results suggested a secondary inversion between rRNA operons A and C. To support this finding, we sequenced the A1552 genome with a combination of Illumina short read and Pacific Bioscience long read sequencing (GenBank Accession CP024867 and CP024868; see supporting S1 Text).
Adjusting the genomic positions to the new genome sequence, we were able to follow the replication activity of the two V. cholerae chromosomes after release from the stringent response (Fig 2A–2D). Indeed, Chr1 initiated replication first as seen by a higher copy number of genomic loci near the replication origin 10.5 minutes after shifting to SHX-free medium. At later time points, this region of higher copy number increased gradually in size, indicating bi-directional replication towards the terminus region. Chr2 replication was not detected until 28 minutes after release from stringent response. Notably, its replication was more difficult to detect due to its smaller size and the large integron not providing reliable copy number values. A stepwise function was fitted to a total of 13 copy-number plots of Chr1 to determine average replication fork positions at different time points (Supporting S3 Fig)[32]. Interestingly, one replication fork runs about 60 kb ahead of the other on Chr1, similar to what was found for E. coli (Fig 2E). Based on the progression of the replication forks, we calculated a replication rate of 22 kbp/min or 360 bp/s for the replication in V. cholerae (Fig 2E).
Clearly, the secondary chromosome in V. cholerae starts to replicate after about two-thirds of the primary chromosome are replicated, causing a synchronous termination of both chromosomes. Two different models could be derived from these observations for the DNA replication within the family of Vibrionaceae. First, it is of biological relevance to the cells to replicate two-thirds of the primary chromosome before starting replication of Chr2. Second, it is of biological relevance to the cells to terminate the two chromosomes at approximately the same time. Here we wanted to test both hypotheses to pave the way for a general understanding of the delay in initiation timing of DNA replication of Chr2 within the Vibrionaceae. The replication start control model, in which replication of two-thirds of Chr1 is important before Chr2 replication is initiated, implies that in Vibrio species with smaller secondary chromosomes than that of V. cholerae, replication of Chr2 ends before that of Chr1 (Fig 3A). In Vibrio species with larger secondary chromosomes, replication of Chr2 would terminate after Chr1. In contrast, the replication end control model would imply that replication of a small secondary chromosome starts later than that of a larger one (Fig 3A). To test both models we used a comparative genomics approach. Recently, replication of a sequence called crtS on Chr1 was found to trigger Chr2 initiation in V. cholerae [25, 27]. If such a site also appears in other Vibrio species, its position on Chr1 could be used as proxy for the time of Chr2 initiation. Based on the sequence of the V. cholerae crtS site, we searched the database for similar sequences occurring only once per genome in Vibrionaceae and generated a multiple sequence alignment (Fig 4A). The most conserved sequence parts were then used to find a set of 129 sequences and generate a corresponding sequence logo (Fig 4B). To test experimentally, if initiation at secondary replication origins in Vibrio species other than Vibrio cholerae is triggered by crtS sites, we analyzed the replication of mini-chromosomes, each driven by one of eleven secondary replication origins from different species of the Vibrionaceae. For a corresponding mini-chromosome based on V. cholerae ori2, it was shown that the copy number increases in an E. coli strain carrying a copy of the crtS site; this did not occur in a strain lacking crtS [27, 33]. As readout for the replicon copy number, we measured how well each strain tolerated increased amounts of antibiotic (Fig 5, see Method section for details, [34]). This method is based on the logic that an increased replicon copy number correlates with an increased copy number of the resistance gene, and so correlates with a higher antibiotic tolerance [35, 36]. A significant increase in copy number was observed for 8 out of 11 mini-chromosomes in an E. coli strain carrying the V. cholerae crtS site integrated into the chromosome in comparison to a strain without crtS (Fig 5, compare red and grey bars). Mini-chromosome copy number was similarly increased in strains carrying either the V. nigripulchritudo or the V. parahaemolyticus crtS. (Fig 5, green and blue bars). For two of the mini-chromosomes (synVivuII and synVihaII), the copy number appeared to be high already in the strain without crtS and one mini-chromosome (synPhopII) showed no crtS-dependent copy number increase (Fig 5). In summary, the data showed that replication origins of secondary chromosomes in Vibrionaceae are triggered by crtS sites in general, suggesting this mechanism to be conserved. The data also suggested that crtS sites do not function specifically on the ori2 of their corresponding species, but appear to be interchangeable.
To test the replication start and end control models, the position of the crtS sites on the primary chromosomes of 29 fully sequenced Vibrionaceae species was determined, and the relative distance to ori1 and ter1 calculated (See method section for details). A correlation of the length of two-thirds of one Chr1 replichore to the distance of the crtS site to ori1 would be expected if the start control model for replication in Vibrionaceae held true (Fig 6A, grey dots). However, the data from our comparative genomics approach showed no such correlation (Fig 6A, black and red dots). To test the replication end control model, the distance of the crtS site to ter1 was plotted against the length of a Chr2 replichore (Fig 6B). Here, the values derived from comparative genomics resembled the theoretical data quite well, where the shift between the two respective regression lines correlate with the delay between crtS replication and ori2 initiation observed in V. cholerae [25]. Our findings support the replication end control model to explain the replication timing of the two chromosomes in Vibrionaceae (Fig 3).
To further test DNA replication in Vibrionaceae, we performed marker frequency analysis by next-generation sequencing of eleven different strains from the Vibrionaceae group. If the replication start control model holds true, one would expect copy numbers of ter2 to be higher than ter1 in Vibrio species where Chr2 is smaller than one-third of Chr1. In species with Chr2 bigger than one-third of the corresponding Chr1, the copy number of ter2 should be below that of ter1 (Fig 3B). If the replication end control model was applicable, one would expect the copy numbers of ter1 and ter2 to be equal in all Vibrionaceae. Chromosomal DNA was isolated from exponentially growing cultures and from cells in stationary phase. The DNA samples were then analyzed by Illumina sequencing and copy numbers plotted according to chromosomal positions. Copy numbers of the two chromosomes were close to one in stationary phase and showed a flat distribution in most cases, as expected for non-replicating cells (Supporting S5 Fig, supporting S4 Table). In all analyzed cases of exponentially growing cells, the copy number plots formed typical triangular shapes, with the replication origin at the highest point and copy numbers declining towards the termini (Fig 7). Note that the data were not normalized to the copy numbers of stationary phase cells. Two lines were fitted to each of the chromosomes and their intersection assigned as the minimal and maximal copy number as described [37]. The position of the maxima corresponded well with the positions of ori1 and ori2 with about 39 kbp deviation on average (below 1% of genome size), supporting good data quality (Supporting S2 Table). Also, data correlated well in biological replicates (Supporting S6 Fig, supporting S2 and S3 Tables). The copy number of ori2 was lower than the copy number of ori1 in all strains, consistent with a conserved replication mode within the Vibrionaceae. In fact, the copy number of ori2 was also lower than that of the region carrying the corresponding crtS site on Chr1 in all studied strains, suggesting that crtS-based triggering of Chr2 replication is a conserved mechanism. The genomic plots showed the copy numbers of ter1 and ter2 within individual strains to be very similar, although some variation occurred (Fig 7). To test the two proposed models of replication start versus replication end control, we plotted the ter1/ter2 ratio for the analyzed strains (Fig 7L). Values were around one, with some variation supporting the replication end control model. Notably, we found no good correlation between crtS position, Chr2 size and how well the ter1/ter2 ratio matches 1. The ratio of copy numbers between the Chr1 position two-thirds of replichore size from ori1 and the ori2 copy number was higher, with a mean of 1.4. This indicates that replication in Vibrionaceae does not follow the replication start model (Fig 7L).
In V. cholerae, the secondary chromosome initiates its replication after about two-thirds of the primary chromosome has been replicated. As a consequence, the two chromosomes terminate at approximately the same time. It has been shown that this replication pattern can be changed by moving the crtS sites to other positions, either further away or closer to the terminus [25]. Such engineered strains have no dramatic deficiencies in cell viability. Why has evolution shaped the replication timing to be as it is found in V. cholerae and other species of Vibrionaceae? We approached this question by asking if the selection pressure lies at the start or end of Chr2 replication timing (Fig 3). By analyzing replication rules in multiple Vibrio species, we show here that it is in fact the timing of the end of replication relative to Chr1 which is under selection, and not the start. In other words, the delay between the start of Chr1 and Chr2 replication respectively seems to be unimportant, but it appears to be more important that the two chromosomes terminate replication at approximately the same time. This begs the question of why synchronous termination of both chromosomes in Vibrionaceae is important. One reason could be the coordination of chromosome segregation and cell division. The chromosomal region opposite the replication origin is the part of the chromosome where dimer resolution occurs at the dif site [38]. In addition, chromosome segregation is coordinated with cell division through interactions of the Ter domain(s) with the divisome in E. coli, as well as in V. cholerae [39–41]. Interestingly, it was found that in engineered V. cholerae strains, in which Chr2 terminates long before Chr1, the two copies of ter2 remain at the middle of the cell until cell division, and segregate approximately at the same time as ter1, like in wildtype cells [25]. Cohesion of ter1 and ter2 with their respective sister ter sequences near the division site thus seems to be important for segregation and the synchronized termination of the two chromosomes might facilitate this mechanism.
We could imagine an alternative explanation of why termination of the two chromosomes is conserved, which at present is more speculative. It could be that the replication pattern is the result of two opposing selection pressures. One driving force in Vibrio evolution could be the simultaneous replication of the two chromosomes. For cell cycle regulation and to limit the overall replication time to a minimum, it could be beneficial for the cell to not replicate one chromosome after the other. On the other hand, the secondary chromosome could be viewed as an invader which the cell needs to keep at bay as a second form of selection pressure. Indeed, Chr2 is thought to originate from a plasmid that the Vibrionaceae acquired early in evolution [20]. The cell might suppress Chr2 replication as far as it can to limit the danger of Chr2 taking over. Indeed, there is evidence that replication origins act as selfish genetic elements [42]. With overlapping replication cycles in fast growing cells, one could essentially imagine patterns in which Chr2 initiates replication before Chr1 with regards to the cell cycle. However, in different growth conditions, it is always Chr1 that initiates before Chr2 [22]. This finding might support a selective process active in Vibrio species to keep “Chr1 first”. In this context, it is also interesting that the initiator protein RctB actually has the capacity to mediate higher copy numbers of Chr2, since many different single amino acid changes lead to copy up of Chr2 numbers [14, 34, 43]. However, selection obviously works against these copy-up mutations in Vibrios occurring in nature. The selection pressure resulting from this Chr2 suppression would be to keep the Chr2 copy number as low as possible. The combination with the second selection pressure for simultaneous replication of the two chromosomes would finally result in termination synchrony. It is noteworthy that the termination synchrony appears to tolerate some deviation, as seen in the variation in copy number ratios of ter1 to ter2 (Fig 7). This observation suggests the evolutionary process leading to termination synchrony to be slow compared to chromosome rearrangements leading to changed distances between relevant genetic loci (ori1, crtS) within the system. One form of such chromosome rearrangement will be discussed in the next section. Besides the observed deviation in termination of the two chromosomes in Vibrionaceae, the delay between crtS replication and ori2 initiation also appears to be variable (Fig 7). We did not find any correlation between this delay and other relevant parameters, such as crtS site position, Chr2 size, the distance of ori1 to the crtS site, or between any combinations of these (Supporting S7 Fig). Factors influencing the delay duration remain to be discovered. One of the longest delays between crtS and ori2 replication was observed for Photobacterium profundum (Fig 7J). Interestingly, the respective ori2 mini-chromosome was not triggered by any of the three crtS sites tested in E. coli (Fig 4). The reason remains to be discovered since we found no obvious deviation of the Photobacterium crtS-site sequence from the other sequences (Fig 4). We also observed no increase of copy numbers for the mini-chromosomes with ori2 copies of V. harveyi and V. vulnificus (Fig 5). However, here the copy numbers of mini-chromosomes were high already in strains without crtS. Notably, the used assay is not able to detect an even further increase in copy number. The copy number of the secondary chromosomes in V. harveyi and V. vulnificus are not increased, suggesting that the observed copy-up phenomenon is mini-chromosome specific.
Most published studies on DNA replication in V. cholerae have used the O1 El Tor strain N16961. This is the strain for which the first V. cholerae genomic sequence was published [8]. However, this strain was later found to not be transformable via natural competence due to a frameshift mutation in the regulator hapR [44, 45]. In contrast, the closely related O1 El Tor strain A1552 encodes a fully functional system of natural competence [44]. We thus decided to use strain A1552 for our experiments, expecting results directly comparable to studies using strain N16961, as we have frequently used the published N16961 genome sequence for primer design for A1552 sequences and never experienced any deviation. Our probe design for DNA microarrays was therefore also based on the N16961 genomic sequence. However, the initial plots showed non-continuous slopes from origin to terminus, indicating some sort of chromosomal rearrangement [46]. Interestingly, an inversion around ori1 was also found recently in strain N1691, in contrast to the published genome sequence (Supporting S1B Fig and [25]). This inversion at ribosomal operons B-H was also detected in strain A1552. Additionally, a second inversion was found at rRNA operons A-C by diagnostic PCR and whole genome sequencing, including long reads to cover the large rRNA encoding operons. These operons were indeed the point at which the inversions occurred. This is certainly due to the high similarity of rRNA operons to each other presenting a good target for homologous recombination as mechanism of inversion. Such inversions at homologous sequences such as rRNA operons or transposons have been observed when comparing genome arrangements of related strains or during long term evolution experiments [47, 48]. A reconstruction of Yersinia evolution by comparative genomics suggests that as much as 79 inversions have happened to shape the genome arrangement seen today [49]. Inversions such as those found here in V. cholerae might thus happen quite frequently. This is certainly interesting in the context of chromosomal macrodomains that often rely on biased distributions of DNA motifs in the origin-to-terminus orientation [50–52]. However, it is also relevant for the crtS-based regulation of Chr2 initiation, in which chromosomal distances between crtS and origin or terminus certainly matter [25], as also seen in our study. In fact, the distance between the crtS site and ori1 is changed by 142 kbp in the strain A1552 analyzed here compared to the N16961 strain analyzed before. This difference correlates well with an earlier termination of Chr2 relative to Chr1 in strain A1552 in comparison to N16961 (Fig 7, [25]). We cannot exclude that crtS-site positions in the different Vibrio species we analyzed are also re-localized slightly as, in some cases, we analyzed sub-strains not exactly resembling the strain with an available genome sequence. However, the analyses of many Vibrios instead of only individual strains should even out such uncertainties.
Bacterial populations in batch cultures are mixtures of cells in all different cell cycle stages, ranging from newly-born cells to large cells shortly before division. Such cultures can be used in many ways to study mechanisms of DNA replication and related processes. However, often a synchronized cell culture is desirable in studies investigating temporal resolution in DNA replication. In bacteria, different methodologies have been developed to generate synchronous populations. These include differential density centrifugation for Caulobacter and related bacteria [53], the baby machine [54], the baby cell column [55], and the blocking, and latter release, of DNA replication by the temperature shift of strains carrying temperature-sensitive protein mutants involved in initiation of DNA replication [56, 57]. The later approach was first developed in E. coli based on screens for mutant strains with temperature sensitive DNA replication mechanisms. Mutations appeared either in the initiator protein DnaA, or the DnaC protein responsible for loading the helicase DnaB. It was shown that amino-acid exchanges which rendered the E. coli DnaA temperature sensitive could help to rationally design a temperature sensitive DnaA in other bacteria [58]. Since V. cholerae and E. coli are relatively closely related and their DnaA proteins are highly similar, we mutated the V. cholerae dnaA according to the temperature sensitive E. coli DnaA. Notably, DnaC could not be used because V. cholerae lacks a homolog. The constructed V. cholerae DnaA showed temperature-sensitive activity in the heterologous E. coli system, but we were not able to exchange the natural V. cholerae dnaA with the mutated allele. As alternative way of synchronization, we tested the stringent-response-based method established by Ferullo and colleagues. [29]. Here, initiation of replication is blocked by addition of serine hydroxamate (SHX), which induces the stringent response. Transfer to SHX-free medium leads to synchronous initiation of DNA replication in E. coli. Ferullo and his collaborators suggested that this method should be transferable to other bacteria with a stringent response system such as that in E. coli. We demonstrated here that this assumption is true and predict that more, but not all, bacterial species could be synchronized using the stringent response. A negative example would be Bacillus subtilis, where the elongation of DNA replication, not the initiation, is blocked by inhibiting primase, an essential component of the replication machinery, during the stringent response [59]. Although we clearly showed the synchrony of the culture, including the linear increase of cellular DNA content as well as temporally-separated initiation of Chr1 and Chr2 replication, the synchronization could be further improved. In a perfectly synchronized culture, all cells would initiate DNA replication at the same point in time, which is hardly feasible with currently available synchronization methods. For example, the initiation of chromosomal replication in the widely applied DnaC-ts system in E. coli differs in the range of some minutes between cells [60, 61]. Furthermore, in the system established here, cells initiate replication after the first cells have started replication, as can be seen by the increasing copy number of ori1 in comparison to ter1 over time after release from the stringent response (Fig 2). One possibility to limit replication initiation to a narrower window could be to add SHX for a second time which should limit initiation to the time between release and re-addition of SHX.
The chromosomal replication origin in E. coli was found to be asymmetric, this being the root of an offset between the two replisomes [32]. The offset varied from strain to strain between 40 to 130 kbp. We observed a similar offset for the two replication forks on Chr1 in V. cholerae in synchronized cells (Fig 2E). Regarding the sequence of the replication origins, the replication fork that runs ahead is the same in V. cholerae and E. coli. It was suggested that the asymmetry of replication is caused by the asymmetry of the replication origin itself, where the initiator protein DnaA multimerizes on the right side to melt an AT-rich region on the left side [32]. Intuitively, one could imagine the replication to start more easily in the direction where no initiator complex sits in the way. In the context of Chr2 regulation, the observed offset is interesting because the exact time point of crtS site replication, and with it the time of Chr2 initiation, depends on ori1 orientation. The frequent chromosomal inversions around ori1 discussed above might consequently lead to frequent changes in Chr2 replication timing and might also explain the observed deviation from exact termination synchrony in other Vibrios (Fig 7).
All strains, plasmids and oligonucleotides used in this study are listed in supporting S5–S7 Tables. Unless indicated otherwise, cells were grown in LB medium, Marine broth, or AB medium supplemented with 25 μg/ml uridine, 10 μg/ml thiamine and 0.2% glucose with 0.5% casamino acids (AB Glu CAA) or 0.4% sodium-acetate (AB So-Ac) [62]. Antibiotic selection for E. coli was used at the following concentrations if not indicated otherwise: Ampicillin 100 μg/ml, kanamycin 35 μg/ml, spectinomycin 100 μg/ml. For growth curves, cells were grown in a 96-well plate at 37°C in a microplate reader (Victor X3 Multilabel Plate Reader, PerkinElmer). OD450 was measured every 6 min for 18 h.
ori2-based mini-chromosomes synVihaII and synVitaII were constructed as described in [34]: ori2s with parAB and rctB were amplified from gDNA of the respective strain. For synVihaII, the ori2 region was amplified with primers 1515/1517 and 1410/1516 from gDNA of V. harveyi. For synVitaII, the ori2 region was amplified with primers 1164/1557 from gDNA of V. tasmaniensis. Both ori2 regions were assembled with AscI-digested synVicII-1.351 per Gibson assembly [63] and transformed in E. coli XL1Blue cells. For integration of crtS in E. coli MG1655, integration cassettes were constructed by MoClo assembly [64]. For pMA161, the crtS was amplified with primers 1474/1475 from gDNA of V. nigripulchritudo and for pMA892 with primers 1443/1444 from gDNA of V. parahaemolyticus. All PCR products were assembled in pMA349 by MoClo assembly as described in [65] and transformed into E. coli TOP10 cells. For pMA451, the backbone pMA327 was assembled with pICH50900, pMA709, pMA710, pMA431 and pMA161. For pMA454, the backbone pMA327 was assembled with pICH50900, pMA709, pMA710, pMA431 and pMA892. All assemblies were transformed in E. coli DH5α λpir cells. Integration cassettes were cut out with BsaI, integrated in E. coli AB330, transferred in E. coli MG1655 per P1-transduction and recombined to remove the resistance as described in [65]. oriII-based mini-chromosomes were added to wild type and crtS strains by conjugation or transformation.
V. cholerae A1552 grown in AB Glu CAA was treated with 0.9 mg/ml serine hydroxamate (SHX) at an OD450 of around 0.15 (exponential phase). After an incubation of 150 min, the cells were harvested by centrifugation and re-suspended in fresh medium without SHX. Samples for flow cytometry and CGH were taken every 3.5 min if not indicated otherwise.
Unless described otherwise, the cells were harvested and washed twice in TBS (0.1 M Tris-HCl pH 7.5, 0.75 M NaCl). They were fixed in 100 μl TBS and 1 ml 77% ethanol and stored at least overnight at 4°C. The samples were washed in 0.5 M sodium-citrate and treated with 5 ng/ml RNase A in 0.5 M sodium-citrate for 4 hr at 50°C. They were stained with 250 nM SYTOX Green Nucleic Acid Stain (Thermo Fisher Scientific) and analyzed on Fortessa Flow Cytometer (BD Biosciences). The SYTOX Green fluorescence was measured through a 530/30 nm bandpass filter. V. cholerae A1552 cells grown in AB So-Ac were fixed with ethanol and stained as described above. These cells served as the standard and were measured alternatingly with the samples. Data was processed with the software FlowJo (Treestar, Ashland, USA). For display as density maps, the sample data was aligned to the corresponding standard and converted into density maps by R.
CGH was performed as described [37]. For hybridizing, Agilent SurePrint G3 Custom CGH Microarrays, 8x60K (Design ID: 074887) were used. They were designed on V. cholerae N16961 (NC_002505.1 and NC_002506.1) with a probe length of 60 bp and a probe distance of 7 bp. Probes with multiple hybridization sites were excluded. As a reference, DNA from stationary V. cholerae A1552 grown in AB Glu CAA was used. Probe signal ratio values were merged in 1000 bp windows. A Lowess fitting was applied to the microarray data to get a locally weighted average (shown as green line in CGH plots). For Chr1, a stepwise function was fitted on the data. The stepwise function divides the plot into five parts: two flat parts at the edges (not yet replicated), one flat risen part in the middle (already replicated) and an increasing and decreasing part (replicating at that moment). The four points at the transition from one of these lines to the next were defined by chromosomal position (x1 to x4) and the heights (h1 to h4). These values were estimated based on plots of the raw data and then used for fitting in conjunction with the stepwise function using nls (nonlinear least-squares) of the R statistics software. Mean positions of the replication forks were calculated as the middle of the increasing (left fork) or decreasing (right fork) part. Progression and asymmetry of the forks was calculated in Excel. For Chr2, copy number values of 10.000 bp windows were compared to the mean copy number of the corresponding plot and displayed as above or below the mean.
crtS sequences of Vibrionaceae in Fig 4A were found with BLAST [66]. The alignment was done with ClustalOmega [67] and the layout with MView [68]. Additionally, crtS sites in 114 Vibrionaceae from NCBI were found with fuzznuc (http://www.hpa-bioinfotools.org.uk/pise/fuzznuc.html) by using the sequence CAGnATATGTAACTnATGCTTTCGG with a maximum of three mismatches. This search resulted in only one hit per genome. The consensus was visualized with WebLogo 2.8.2 [69].
For comparative genomics of Vibrionaceae, data of 29 fully sequenced and annotated strains from NCBI were used. Positions of ori1 and ori2 were either found at dOriC [70] or by assigning the intergenic region between gidA/mioC (ori1) or rctB/parAB (ori2). The genes were either found by annotation or with BLAST [66]. One half of each chromosome was defined as a replichore, and ter1 was then calculated as the opposite position to ori1 on Chr1 (ori1 + 1/2 Chr1). The position of crtS was found using the consensus sequence with fuzznuc. Expected values of both parameters were calculated by using the same data on both X and Y axis (two-thirds Chr1 replichore and Chr2 replichore, respectively).
Analysis of mini-chromosome copy numbers was as described [34]. Cells were grown in LB medium containing either 100 or 500 μg/ml ampicillin at 37°C in 96-well plates in a microplate reader (infinite M200pro multimode microplate reader, Tecan). The 150 μl of main culture was inoculated 1:1000 and growth curves recorded for 15 hr. Statistical significance of differences between wt and crtS strains was calculated by a two sample t-test. For better visualization, 1 divided by the time needed to reach an OD of 0.1 was defined as a measure of the copy number.
For Illumina sequencing of the Vibrionaceae, cells were grown in marine broth at either 28°C or 10°C (Photobacterium profundum) to either exponentially or stationary phase. Genomic DNA was prepared by incubating resuspended frozen cell pellets in 300 μl TE-buffer with 1.2% SDS and 4 mM EDTA for 5 min at 65°C. After adding 750 μl isopropanol, the precipitate was incubated in 500 μl TE with 50 μg RNase A for 90 min at 65°C and additional 15 min at 37°C with 50 μg proteinase K. DNA was isolated with phenol/chloroform. Final DNA was resuspended in deionized sterile water and quantified using a NanoDrop (ThermoFisher Scientific). Genomic DNA was sequenced by applying the Nextera XT library kit and a MiSeq v3 reagent kit with 150 cycles on an Illumina MiSeq (Illumina, USA).
For PacBio sequencing, V. cholerae A1552 [31] was grown in AB Glu CAA at 37°C to stationary phase. Genomic DNA was prepared as above. Final DNA was resuspended in TE-buffer and quantified using a NanoDrop (ThermoFisher Scientific) and a Qubit Fluorometer (Life Technologies). The SMRTbell template library was prepared according to the instructions from PacificBiosciences, Menlo Park, CA, USA, following the Procedure & Checklist—20 kb Template Preparation Using BluePippin Size-Selection System. Briefly, for preparation of 15kb libraries ~8μg genomic DNA libraries was sheared using g-tubes from Covaris, Woburn, MA, USA according to the manufacturer´s instructions. DNA was end-repaired and ligated overnight to hairpin adapters applying components from the DNA/Polymerase Binding Kit P6 from Pacific Biosciences, Menlo Park, CA, USA. Reactions were carried out according to the manufacturer´s instructions. BluePippin Size-Selection was performed according to the manufacturer´s instructions with a size selection cutoff of 4 kb (Sage Science, Beverly, MA, USA). Conditions for annealing of sequencing primers and binding of polymerase to purified SMRTbell template were assessed with the Calculator in RS Remote, PacificBiosciences, Menlo Park, CA, USA. SMRT sequencing was carried out on the PacBio RSII (PacificBiosciences, Menlo Park, CA, USA) taking one 240-minutes movie.
Genome assembly was performed with the RS_HGAP_Assembly.3 protocol included in SMRT Portal version 2.3.0. Both chromosomal contigs were successfully assembled and trimmed, circularized, as well as adjusted to dnaA (Chr1) and rctB (Chr2) as the first gene.
Quality improvement of the PacBio HGAP assembly was performed by a mapping of all corresponding Illumina short reads using the Burrows-Wheeler Aligner (BWA) using bwa aln and bwa sampe [71]. Illumina reads were mapped onto the obtained chromosome and plasmid sequences with subsequent variant and consensus calling using Varscan2 [72] and GATK [73]. A final quality score of QV60 was attained. Automated genome annotation was carried out using Prokka [74]. The genome sequence was submitted to GenBank (Accession Number: CP024867; CP024868).
In the first step, reads from the exponential and stationary phase were mapped on the respective Vibrio replicons using qalign from the QuasR R package. Subsequently, replicon-wide coverage was calculated by bedtools genomecov using the 5' ends of the reads. Single base coverage was smoothed by a 5 kbp sliding window averaging with a shift of 1 kbp. Windows with an internal standard deviation that exceeded three times the difference between the median and the third quartile of standard deviations of windows within 500 kbp were removed. These are windows, with an average coverage that does not properly reflect the coverage of individual bases. Furthermore, windows with an internal standard deviation below three times the difference between the median and the 1st quartile of standard deviations of windows within 500 kbp were removed. These are windows with many bases of low or mostly zero coverage indicating deviations of reads from the template sequence. The procedure removes unreliable window averages (data points) taking the noisiness of the data and regional specificities into account. Sequence bias was removed as follows: firstly, the coverage of exponential and stationary phase samples was normalized to the total amount of mapped reads to remove the bias of total read counts in the samples. Then, ratios of exponential and stationary phase coverage were determined. Ratios were subsequently corrected for a systematic sequence-dependent local bias [25], using the second exponential phase sample.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.